Bay eu he A bei spapie e zu Wi scha sin o ma ik
Leh s uhl ü
Wi scha sin o ma ik
In o ma ion Sys ems
Managemen
Bay eu h Repo s on In o ma ion Sys ems Managemen
No. 52
Oc obe 2011
Sebas ian Hude
The BabelNEG Sys em - A P o ocol-gene ic
In as uc u e o Elec onic SLA Nego ia ions in he
In e ne o Se ices
ISSN
1864-9300
Die A bei spapie e des Leh s uhls ü
Wi scha sin o ma ik dienen de Da s ellung
o läu ige E gebnisse, die i. d. R. noch ü
spä e e Ve ö en lichungen übe a bei e we den.
Die Au o en sind deshalb ü k i ische Hinweise
dankba .
The Bay eu h Repo s on In o ma ion Sys ems
Managemen comp ise p elimina y esul s
which will usually be e ised o subsequen
publica ions. C i ical commen s would be
app ecia ed by he au ho s.
Alle Rech e o behal en. Insbesonde e die de
Übe se zung, des Nachd uckes, des Vo ags,
de En nahme on Abbildungen und Tabellen –
auch bei nu auszugsweise Ve we ung.
All igh s ese ed. No pa o his epo may
be ep oduced by any means, o ansla ed.
Au ho s: In o ma ion Sys ems Managemen
Wo king Pape Se ies
Edi ed by:
P o . D . To s en Eymann
Con ac :
Uni e si ä Bay eu h
Leh s uhl ü Wi scha sin o ma ik (BWL VII)
P o . D . To s en Eymann
Uni e si ä ss asse 30
95447 Bay eu h
Ge many
Email: [email p o ec ed]
ISSN
Sebas ian Hude (Uni e si y o Bay eu h)
1864-9300
The BabelNEG Sys em – A
P o ocol-gene ic In as uc u e o
Elec onic SLA Nego ia ions in he In e ne
o Se ices
Disse a ion
zu E langung des G ades eines Dok o s de Wi scha swissenscha en
de Rech s- und Wi scha swissenscha lichen Fakul ¨a
de Uni e si ¨a Bay eu h
Vo geleg
on
Sebas ian Hude
aus
Schwein u
Dekan: P o . D . Ma kus M¨os l
E s be ich e s a e : P o . D . To s en Eymann
Zwei be ich e s a e : P o . D . Guido Wi z
Tag de m¨undlichen P ¨u ung: 04. 10. 2011
Acknowledgmen s
The esea ch p esen ed in his doc o al hesis has been ca ied ou du ing my assignmen
a he Depa men o In o ma ion Sys ems Managemen a he Uni e si y o Bay eu h.
Along he pa h owa ds his hesis, I was in ol ed in se e al esea ch p ojec s and he
academic li e in gene al, p o iding me a e y challenging bu also s imula ing wo k
en i onmen . I owe a deb o g a i ude o many people who ha e assis ed me along he
way and influenced bo h my wo k and my unde s anding o he opic o elec onic SLA
managemen .
In pa icula , I wish o exp ess my g a i ude o my p ima y supe iso , P o esso D .
To s en Eymann o he oppo uni y he ga e me a his depa men . The discussions
we had, his many concep ual and me hodical sugges ions and no leas his cons an
encou agemen undamen ally affec ed my wo k and c ucially con ibu ed o i s success.
I am g a e ul o P o esso D . Guido Wi z om he Mobile and Dis ibu ed Sys ems
G oup a he O o-F ied ich Uni e si y Bambe g. He decisi ely influenced my way o
assessing and designing dis ibu ed so wa e in as uc u es du ing his supe ision o
bo h my diploma and co-supe ision o my doc o al hesis.
Fu he mo e I am deeply indeb ed o my colleagues a Bay eu h Uni e si y o hei
elen less assis ance wi h concep ual, echnical and me hodological p oblems along he
way. Depending on hei academic o igin in compu e science o business adminis a-
ion / economics hey all p o ided me wi h aluable inpu and helped me unde s and
he in icacies o he esea ch a ea I wo ked in. In spi e o all difficul ies occu ing in
such a demanding p ojec , hey managed o c ea e an almos homelike en i onmen o
which I am deeply g a e ul. In pa icula , I would like o hank S e an K¨onig whom I
had he hono o ge o know a he e y beginning o my unde g adua e s udies and
wo k oge he e e since as well as Ch is oph Niemann o he many discussions on
esea ch me hodology and concep ual help he acco ded me. A special hanks goes o
Axel P¨u ckhaue o his cons an echnical suppo o ou wo k; me ha ing been able
o con inuously wo k on my esea ch is g ea ly owed o his effo s. I am p oud o ha e
been a membe o his eam and o be able call hese people my iends.
To s en Eymann always encou aged an exchange o ideas wi h he na ional and in-
e na ional esea ch communi y. On many occasions he ga e me he oppo uni y o
isi con e ences and wo kshops, gi e alks and in gene al mee esea che s om all o e
he globe, wo king on simila opics. Among hese, I owe much g a i ude o P o . D .
Udo K iege (O o-F ied ich Uni e si y Bambe g) o belie ing in me and acquain ing me
wi h D . Heiko Ludwig om IBM Resea ch and he a ea o au oma ed SLA nego ia ions
in gene al. This opened many doo s o me in he academic wo ld. I consequen ly wan
o hank D . Ludwig o esco ing me and my wo k o a long ime and o he endless
phone calls we had o e he yea s, in which he consul ed me in much mo e han jus my
iii

esea ch opic. I am also e y g a e ul o esea che s I was lucky o mee wi hin he col-
labo a i e p ojec s I wo ked in, such as CATNETS, SORMA o eRep, as well as he ones
I me a esea ch o ums, such as he Open G id Fo um o a Dags uhl Cas le. Among
hese I pa icula ly owe Philipp Wiede (Geo g-Augus Uni e si y G¨o ingen), D .-Ing.
Bas ian Kolle (High Pe o mance Compu ing Cen e S u ga ), D . Shamima Pau-
obally (Uni e si y o Wes mins e ), P o . Ome F. Rana (Ca diff Uni e si y), D .-Ing.
Nicolas Repp (Technical Uni e si y o Da ms ad ) and P o . F ances B azie (Technical
Uni e si y o Del ) o hei ime and he aluable inpu hey ga e me o my wo k.
Wi h all my hea I hank my wonde ul amily and iends. My pa en s Ka in and
We ne aised, suppo ed, augh and lo ed me in a way I am bu beginning o g asp and
hey s ill con inue o do so. Wi hou he uncondi ional help, pa ience and unde s anding
ecei ed om hem and my much alued b o he Fabian his wo k would ne e ha e been
possible. I am o e e indeb ed o hem o all hey did o me h oughou my li e.
Finally my mos hea el g a i ude goes o Ve ena. Ha ing me he a he beginning
o my ime in Bay eu h and coming o know he suppo , cons an encou agemen and
he uncondi ional lo e and companionship, I conside he g ea es godsend in my li e.
Bay eu h, Oc obe 2011
Sebas ian Hude
i
Abs ac
Visions o he nex -gene a ion In e ne o Se ices a e d i en by digi al esou ces aded
on a global scope. Fo he esul ing economic se ing, au oma ed on-line echniques o
handling se ices and esou ces a e needed, as well as o ad e ising and disco e ing o
o he on- he-fly nego ia ion o p ope e ms o hei use.
This hesis p esen s he esul s o my disse a ion p ojec . They comp ise a se ice
in as uc u e, able o suppo he s uc u ed disco e y and p o ocol-gene ic nego ia ion
o elec onic se ice le el ag eemen s (SLAs) and hus se ices hemsel es.
The need o such an a i ac is deduced om a de ailed scena io analysis, ex apola -
ing pas and cu en de elopmen s in dis ibu ed business in o ma ion sys ems. Based
on economic heo y, he need o bo h nego ia ion p ocesses as such and he possibili y
o adap o diffe en p o ocols a un ime is in e ed in a second s ep.
The equi emen s o my p o o ype sys em a e hen de i ed om he scena io model
and unde lying economic heo ies. I discuss concep ual ounda ions, comp ising heo e -
ical p inciples o he design, o mula ion, disco e y, nego ia ion, and subsequen usage
o (elec onic) SLAs in dis ibu ed in o ma ion sys ems.
A e ha ing p esen ed hese undamen al concep s, he ac ual in as uc u e design,
hus he p oposed solu ion o he s a ed esea ch p oblem, is de ailed. The unde lying
idea is o decouple he good o be sold ( he SLA) om he nego ia ion p o ocol, enabling
a se ice p o ide o apply diffe en nego ia ion p o ocols o he same se ice o e ime.
Fu he mo e, o he consume side a p o ocol-gene ic nego ia ion componen is de-
signed, capable o au oma ically adap ing o diffe en p o ocols, as offe ed by he se ice
p o ide s. The concep ual copula be ween hese wo sides is a se o s uc u ed se ice de-
sc ip ion documen s, defining no only he se ice- ele an unc ional and non- unc ional
pa ame e s, bu also he applied nego ia ion p o ocol in a machine- eadable way.
The las chap e p esen s a ho ough assessmen o he in as uc u e design, includ-
ing he ac ual implemen a ion o he de eloped mechanisms and da a s uc u es in a
Ja a-based simula ion en i onmen , he concep ual demons a ion o he sys em’s effec-
i eness (wi h ega d o he s a ed equi emen s) as well as a simula i e demons a ion
s ep p o ing he adap abili y o he se ice consume agen s. Wi h hese assessmen
s eps, he sys em’s capabili y o ulfill all s a ed equi emen s could be shown, hus
p o ing i s effec i eness o sol e he esea ch p oblem.
Con en s
Lis o Figu es ix
Lis o Tables xi
Abb e ia ions xiii
1. Resea ch P oblem and App oach 1
1.1. In oduc ionandMo i a ion.......................... 1
1.1.1. Scena ioModel............................. 2
1.1.2. Resea ch P oblem . . ......................... 7
1.2. S uc u eo hisThesis ............................ 9
1.3. Resea chMe hod................................ 11
1.3.1. Gene alConside a ions ........................ 11
1.3.2. Design Science in In o ma ion Sys ems Resea ch . ......... 13
1.3.3. Implemen a iono heDesignSciencePa adigm .......... 19
2. Objec i es and Founda ions 23
2.1. Requi emen sAnalysis............................. 23
2.2. Concep ual Founda ions ............................ 26
2.2.1. Se iceLe elAg eemen sin heIn e ne o Se ices ........ 26
2.2.2. Disco e yPhase ............................ 40
2.2.3. Nego ia ionPhase ........................... 45
2.3. Rela edWo k.................................. 55
2.3.1. P ojec sBuildingonWS-Ag eemen ................. 56
2.3.2. App oaches Offe ing Significan P og ess Beyond WS-Ag eemen . 59
2.3.3. P ojec sFocusingonEconomicAspec s ............... 62
2.3.4. Ini ial App oaches Towa ds P o ocol-Gene i y in SLA Nego ia ions 64
3. Design and De elopmen 71
3.1. Abs ac DesignIdea.............................. 71
3.2. Se ice Desc ip ion Documen s ........................ 73
3.2.1. Se iceType .............................. 73
3.2.2. Ex endedSLATempla e........................ 75
3.2.3. Se iceIden ifica o .......................... 83
3.2.4. FinalSLA................................ 84
3.3. P o ocolDesign................................. 85
3.3.1. Disco e yPhase ............................ 85
ii
Abb e ia ions
EST Ex ended SLA Templa e
FIPA Founda ion o In elligen Physical Agen s
FPA Fixed P ice Auc ion
FPSB Auc ion Fi s -p ice-sealed-bid Auc ion
GB Gigaby e
GC G id Compu ing
GGF Global G id Fo um
GN Gene ic Nego ia o
GT Gua an ee Te m
HPC4U Highly P edic able Clus e o In e ne -G ids (P ojec )
ID Iden ifie
IETF In e ne Enginee ing Task Fo ce
IL In as uc u e Laye
IoS In e ne o Se ices
IS In o ma ion Sys ems
ISP Independen Se ice P o ide
ITIL IT In as uc u e Lib a y
NACK Non-Acknowledgemen (Message Type)
NC Nego ia ion Coo dina o
Nex GRID A chi ec u e o he Nex Gene a ion G ids (P ojec )
OASIS O ganiza ion o he Ad ancemen o S uc u ed In o ma ion
S anda ds
OCL Objec Cons ain Language
OGF Open G id Fo um
OGSA Open G id Se ices A chi ec u e
OLA Ope a ional Le el Ag eemen
P2P Pee - o-Pee
xi

Abb e ia ions
QoS Quali y-o -Se ice
RA Regis y Agen
RDA Re e se Du ch Auc ion
REA Re e se English Auc ion
REPAST Recu si e Po ous Agen Simula ion Toolki
RFC Reques o Commen s
SC Se ice Consume
SDT Se ice Desc ip ion Te m
SI Se ice Iden ifica o
SimIS Simula ing and In e ne o Se ices (Toolki )
SLA Se ice Le el Ag eemen
SLO Se ice Le el Objec i e
SLP Se ice Loca ion P o ocol
SLS Se ice Le el Specifica ion
SO Se ice O ien a ion
SOA Se ice O ien ed A chi ec u e
SOAP Simple Objec Access P o ocol
SOC Se ice O ien ed Compu ing
SORMA Sel -O ganizing ICT Resou ce Managemen (P ojec )
SOS Se ice O ien ed Sys em
SP Se ice P o ide
SRT Se ice Re e ence Te m
ST Se ice Type
UDDI Uni e sal Desc ip ion, Disco e y and In eg a ion
UML Unified Modeling Language
URI Uni o m Resou ce Iden ifie
USD US Dolla
x
Abb e ia ions
WS Web Se ice
WS-Ag eemen Web Se ices Ag eemen
WS-Disco e y Web Se ices Dynamic Disco e y
WSCM Web Se ices Composi ion Managemen
WSLA Web Se ice Le el Ag eemen
WSOL Web Se ices Offe ing Language
XML Ex ensible Ma kup Language
x i
1. Resea ch P oblem and App oach
As o now, compu e ne wo ks
a e s ill in hei in ancy, bu as
hey g ow up and become
sophis ica ed, we will p obably
see he sp ead o ’compu e
u ili ies’ which, like p esen
elec ic and elephone u ili ies,
will se ice indi idual homes and
offices ac oss he coun y.
(Leona d Klein ock, UCLA p ess
elease announcing he launch o
he ARPANET in 1969)
1.1. In oduc ion and Mo i a ion
Visions o 21s cen u y’s in o ma ion sys ems (IS) show highly specialized digi al se -
ices and esou ces, which collabo a e con inuously and wi h a global each. Today’s
In e ne o mainly human in e ac ions e ol es o a global, socio- echnical in o ma ion
in as uc u e, whe e humans as well as so wa e agen s, ac ing on hei behal , con-
inuously in e ac o exchange da a and compu a ional esou ces. Possibly millions o
se ice p o ide s (SPs), consume s (SCs) and a mul i ude o in e media ies like b oke s
o wo kflow o ches a o s a e p esen , o ming a global economic en i onmen . This
ision is commonly e e ed o as he In e ne o Se ices (IoS) (Ruggabe 2007; Sch o h
and Janne 2007).
Building on cu en ly applied compu ing pa adigms, such as Se ice-o ien ed (SOC)
(Fos e 2005), G id (GC) (Fos e , Kesselman, and Tuecke 2001) o Cloud Compu ing
(CC) (Buyya e al. 2009), he IoS ision defines highly dynamic ne wo ks o composable
se ices, offe ed and consumed on demand and on a global scope. I igo ously ocuses
on he goal o an In e ne -based se ice economy, simila o he eal-wo ld se ice sec o .
Digi al se ices a e offe ed o e elec onic se ice ma ke s, pu chased by espec i e cus-
ome s and hen combined wi h in e nal o o he ex e nal se ices o business wo kflows
o a ying complexi y. In ha , i allows e en e y small and specialized companies o
find a niche in he digi al economy whe e hey can compe e wi h he ubiqui ous in e -
na ional en e p ises, which in u n ha e o ace a much highe compe i ion on he global
ma ke (Theseus 2009).
Economic success o he IoS c ucially depends on new business models, as well as hei
suppo ing echnical in as uc u e, enabling ading p ocesses down o he le el o an
indi idual se ice, and he subsequen cha ging based solely on i s usage and deli e ed
1
1. Resea ch P oblem and App oach
quali y-o -se ice (QoS). Such models imply he need o mechanisms gua an eeing QoS
o each se ice in oca ion, e en ac oss en e p ise bounda ies.
Since a scena io like he IoS inhe en ly lacks he applicabili y o cen alized QoS
managemen , gua an ees mus be ob ained o his end in he o m o bi- o e en mul i-
la e al se ice le el ag eemen s (SLAs), assu ing se ice quali y ac oss indi idual si es
(Ludwig e al. 2003a).
In he ollowing he esul s o my disse a ion p ojec , dealing wi h he au oma ed
managemen (mo e specifically, he disco e y and nego ia ion) o elec onic se ices and
espec i e SLAs in he IoS, a e p esen ed.
1.1.1. Scena io Model
In his subsec ion a de ailed scena io model, subsequen ly ac ing as he con ex o his
hesis, is de i ed. Fo his pu pose cu en ends in dis ibu ed IS a e analyzed fi s .
Building on hose de elopmen s, a gene ic scena io model, an icipa ing he u u e IoS
en i onmen , can be induced in a second s ep.
Se ice O ien a ion and Se ice O ien ed A chi ec u es
In ecen yea s a new pa adigm o designing and implemen ing business IS has been
es ablished: Se ice O ien a ion (SO). The main idea behind his concep is ha e e y
unc ionali y offe ed by humans, o ganiza ional en i ies o compu e sys ems is consid-
e ed an abs ac se ice, each o which can again be combined wi h o he s o c ea e mo e
complex composi e se ices.
Be o e de ailing he indi idual assump ions and design p inciples p esen in his
pa adigm, a se o ela ed key concep s ha e o be defined and dis inguished om each
o he , namely SOC, Se ice O ien ed A chi ec u es (SOAs) and Se ice O ien ed Sys-
ems (SOS).
Each o hese concep s builds on he basic idea o SO, howe e each ep esen s a
diffe en pe spec i e on his ision. In o de o dis inguish hese pa adigms, a well-
known concep in IS esea ch can be employed, he dis inc ion be ween he ask laye
and he ask ope a o laye (Fe s l and Sinz 2008, pp. 2-5). The ask laye comp ises
all abs ac asks and hei combina ion o p ocesses, whe eas he ask ope a o laye
con ains all human o au oma ed ope a o s p esen in a gi en en e p ise sys em, which
on hei pa can execu e asks assigned o hem.
Applied o he SO ealm, a SOA defines all se ices, and he e o e abs ac unc ional-
i ies, exis ing in a gi en sys em, hus ep esen ing he ask laye . Jus as wi h adi ional
en e p ise IS he ope a o s ( ask ope a o laye ) p o iding he indi idual se ices can be
bo h humans1o compu e p og ams (also e e ed o as elec onic se ices o se ice
ins ances2). The design pa adigm conce ned wi h he defini ion and implemen a ion o
such elec onic se ices is called SOC.
1Fo he emainde o his hesis human ask ope a o s will be omi ed; he p ima y ocus o my wo k
lies on elec onic se ices, he au oma ed ask ope a o s.
2Al hough echnically a dis inc ion be ween se ice ( ask laye ) and se ice implemen a ion ( ask op-
e a o laye ) would also be app op ia e he e, in he majo i y o he scien ific li e a u e only he e m
2
1.1. In oduc ion and Mo i a ion
All o he abo emen ioned concep s apply o ei he he abs ac ision o se ice-based
sys ems o he way he implied asks a e s uc u ed. An ac ual se ice-based sys em is
called SOS, ep esen ing he se o elec onic se ices offe ing he asks defined in a SOA
and consequen ly i s au oma ed ask ope a o s.
Acco ding o he gene al ag eemen in he li e a u e (see o example Papazoglou and
Geo gakopoulos 2003; S ini asan and Teadwell 2005) an elec onic se ice (ins ance),
can husbedefinedas ollows:
•An elec onic se ice is an indi idually add essable so wa e componen ha p o-
ides some unc ionali y o a se ice eques e .
•Se ices can be accessed o e an elec onic ne wo k, such as an en e p ise in ane .
•Indi idual se ices can be composed o highe -le el, mo e complex se ices, esul -
ing in possibly mul iple le els o se ice complexi y.
•Se ices only ad e ise de ails, such as hei capabili ies, in e aces o accep ed
p o ocols ha a e needed o in e ac wi h hem. Technical implemen a ion de ails
o he se ice a e hidden om he se ice eques o s.
•Rega ding hei in e ac ion, elec onic se ices a e loosely coupled. This means
ha hei in e ac ions a e no ha dcoded in each indi idual se ice, bu e e y
se ice eques e disco e s and binds a gi en o he se ice, i in e ac s wi h, a un
ime.
Among o he s, he main ad an age o he esul ing SOSs is a much highe flexibili y,
when compa ed o adi ional sys ems. Due o he loose coupling se ice eques e s
do no ely on ha d-coded links, bu on dynamic se ice disco e y and in oca ions.
Addi ionally, SOSs allow o he dynamic ins an ia ion and emo al o se ice ins ances
o cope wi h load fluc ua ions. The new se ice ins ances jus ha e o be egis e ed o
he disco e y sys em and can immedia ely be in oked (allowing o a be e dis ibu ion
o incoming se ice eques s o a ailable ins ances).
Due o he p ominen ad an ages o his concep he SO ision has b oadly been
adop ed wi hin he las yea s o whole business IS as well as wi hin indi idual so -
wa e sys ems. Significan esea ch wo k has been dedica ed o he defini ion o echnical
se ice s anda ds (such as Web Se ices (WS) (Boo h e al. 2004), he Uni e sal Desc ip-
ion, Disco e y and In eg a ion (UDDI) (Bellwood e al. 2004) and he Simple Objec
Access P o ocol (SOAP) (Gudgin e al. 2007)) suppo ing he defini ion, desc ip ion and
disco e y as well as he in e ac ion wi h elec onic se ices.
G id Compu ing
In pa allel o hese de elopmen s, egula ly applied wi hin a gi en en e p ise, a pa adigm
o dis ibu ed IS has eme ged, which is mos ly dealing wi h he coo dina ion o elec-
onic esou ces ac oss o ganiza ional bounda ies: GC. I has been in oduced in he
se ice is used, e e ing o bo h concep s depending on he con ex . I will also p oceed his way, as
especially wi h my ocus on elec onic ask ope a o s his dis inc ion is o mino impo ance.
3

1. Resea ch P oblem and App oach
ea ly yea s o he new millennium as a new pa adigm o dis ibu ed execu ion o
esou ce-demanding compu ing asks (Fos e 2002; Fos e and Kesselman 2004; Fos e ,
Kesselman, and Tuecke 2001).
Acco ding o wo o he mos ecognized G id esea che s, Ian Fos e and Ca l Kessel-
man, GC is mainly conce ned wi h ”coo dina ed esou ce sha ing and p oblem sol ing
in dynamic, mul i-ins i u ional i ual o ganiza ions” (Fos e , Kesselman, and Tuecke
2001, p. 201), emphasizing he c oss-o ganiza ional na u e o G id sys ems.
G ids can hus be defined as:
•Sys ems ha “coo dina e[...] esou ces ha a e no subjec o cen alized con-
ol”, also add essing p oblems such as secu i y, policy o paymen occu ing when
esou ces a e sha ed ac oss o ganiza ional domains (Fos e 2002, p. 2). Those e-
sou ces ange om compu a ional, s o age and ne wo k esou ces o code eposi-
o ies (Fos e , Kesselman, and Tuecke 2001).
•In doing so, G ids employ “s anda d, open, gene al-pu pose p o ocols and in e -
aces” (Fos e 2002, p. 2), suppo ing he sha ing p ocess.
•The final o e all goal o GC is “ o deli e non i ial quali ies o se ice”, ollowing
he ision o he indi idual “u ili y o he combined sys em [being] significan ly
g ea e han ha o he sum o i s pa s” (Fos e 2002, p. 3). This las s a emen
explici ly confi ms GC’s ocus on he use , aiming a maximizing he pe o mance
o a use ’s applica ion, un on he dis ibu ed G id nodes (Schop and Ni zbe g
2002).
Wi h he defini ion o he Open G id Se ices A chi ec u e (OGSA) (Fos e e al.
2002a) he GC ision has been in eg a ed wi h eme ging SO p inciples. The adi ional
goal o GC, o execu e indi idual jobs on a se o dis ibu ed esou ces, emained he
same, howe e a much mo e p ecisely desc ibed a chi ec u e o hose esou ces was
gi en. Acco ding o he ision o SO, each G id esou ce was conside ed o be a se -
ice. Se ice seman ics and managemen in e aces o he basic building blocks in such
sys ems, G id Se ices, we e gi en, hus con e ging he me apho s used in business
ansac ions (i.e. elec onic business ela ionships wi h ex e nal o in e nal SPs) and in
compu ing echnology. The cen al concep o a dis inc se ice has become he glue
be ween hose wo lds.
This concep did no only p o ide a common basis o compu e scien is s and po en ial
business use s bu also enabled he eme gence o a powe ul abs ac ion concep in
dis ibu ed compu ing: Vi ualiza ion. “Vi ualiza ion enables consis en esou ce access
ac oss mul iple he e ogeneous pla o ms [and] also [ he] mapping o mul iple logical
esou ce ins ances on o he same physical esou ce...” (Fos e e al. 2002b, p. 40). F om
a use pe spec i e i ualiza ion hus aims a comple ely hiding no implemen a ion bu
also deploymen de ails o a gi en se ice. Se ice eques o s only deal wi h he offe ed
in e aces and do no need any in o ma ion o he ac ual implemen a ion o compu e
ha dwa e he se ice ins ance is deployed on.
4
1.1. In oduc ion and Mo i a ion
Cloud Compu ing
CC as he mos cu en de elopmen builds on GC, SO and Vi ualiza ion echnologies
(Fos e e al. 2002b) in o de o implemen “pa allel and dis ibu ed sys em[s] consis -
ing o a collec ion o in e connec ed and i ualized compu e s ha a e dynamically
p o isioned and p esen ed as one o mo e unified compu ing esou ces based on SLAs,
es ablished h ough nego ia ion be ween he se ice p o ide and consume s” (Buyya
e al. 2009, p. 601).
CC can hus be cha ac e ized as ollows:
1. Clouds hea ily build on i ualiza ion echnologies. All offe ed compu ing e-
sou ces a e i ualized, hus hiding he implemen a ion de ails om he end use .
2. Cloud se ices a e offe ed by independen and ex e nal SPs.
3. Cloud esou ces a e dynamically p o isioned on demand. Compu ing powe is
supplied ha can dynamically scale up o down as he demand o he hos ed
se ices a ies.
4. Each CC offe ing is based on possibly p e-nego ia ed SLAs. The e o e he main
goal o each Cloud se ice is o mee ce ain QoS le els.
5. CC esou ces a e accessed o e s anda d In e ne p o ocols.
The main diffe ence be ween CC and GC is p obably ha a cloud ies o p esen a
cen alized “image” ha manages and schedules i s esou ces in he backg ound (e.g. a
comme cial da a cen e s as wi h Amazon3) as opposed o G ids which explici ly offe
access o decen alized esou ces wi h local policies (Begin 2008).
Clouds in oduce a new abs ac ion laye be ween he aw esou ces and he use s:
a i ualiza ion laye . All esou ces a ailable a a gi en da a cen e a e pooled as an
inpu o he i ualiza ion laye , which in u n offe s he a ailable esou ces as disc e e
compu ing blocks o he use s in he o m o i ual se e s. By in oducing his i u-
aliza ion laye Clouds manage o b eak down he po en ial M:N ela ionships be ween
use s and esou ces o wo se s o 1:N ela ionships (use s o i ualiza ion laye and
i ualiza ion laye o esou ces espec i ely), he e o e educing he complexi y o he
o e all sys em. In con as , G id sys ems combine esou ces, loca ed in diffe en o ga-
niza ional domains, o execu e esou ce demanding jobs in pa allel. Subsequen ly he
indi idual pa ial esul s a e collec ed om he indi idual G id nodes and combined o
a single esul o he o e all job.
Common Vision: In e ne o Se ices
In he las subsec ions he s a e o he a in dis ibu ed compu ing was ske ched in
e ms o h ee commonly applied pa adigms o sys em design and ope a ion. The
p esen ed de elopmen s expose a high deg ee o simila i y in e ms o he mechanisms
and echnologies used.
3h p://aws.amazon.com/
5
1. Resea ch P oblem and App oach
Se ice
O ien a ion
G id
Compu ing
Cloud
Compu ing
In e ne o
Se ices
composable
elec onic
se ices
se ice
consump ion
ac oss company
bounda ies
on demand
in oca ion
based on SLAs
Figu e 1.1.: Rela ion o IoS, SOC, GC and CC
Each o he depic ed scena ios builds on e y simila echnical in as uc u es, based
on he In e ne as a communica ion in as uc u e. Also, he applied compu a ional
abs ac ion o all o hose sys ems is always a se ice.
On he o he hand, hey diffe sligh ly in he way he indi idual se ices a e managed
and used on a highe abs ac ion le el. This is especially no iceable when looking a he
applied in oca ion pa adigms, he poin o con ol wi hin each in oca ion, he o e all
sys em configu a ion as well as he scope o he employed sys ems. Howe e , no ha ing
diffe en in as uc u es bu only using hem diffe en ly should no p e en a de elop-
men o consolida ing and in eg a ing compu ing pa adigms in o de o implemen mo e
powe ul and efficien global sys ems.
Thus, many expe s in IS esea ch and indus y sha e a common ision o he nex
gene a ion In e ne , based on highly dynamic ne wo ks o composable se ices, offe ed
and consumed on a global scope, ul ima ely leading o inno a i e business models and
suppo ing he ansi ion om alue chains o alue ne s (Blau e al. 2009; Schei haue
and Winkle 2008).
This ision builds on he a o emen ioned pa adigms o SO, GC and CC by s ill ollow-
ing he se ice pa adigm, he o ches a ion o in e nal and he cho eog aphy o ex e nal
se ices as well as hei on demand consump ion. In addi ion o combining all hese con-
cep s, he IoS pu s a much s onge ocus on new business models and he comme cial
applica ion o he SO ideas.
Summa izing he IoS scena io model esul s in he ollowing se o cha ac e is ics:
•The IoS is composed o a se o elec onic se ices.
•Se ices a y in complexi y and he e o e ange om aw (ha dwa e) esou ces o
e y complex wo kflows.
•They can be s and-alone ( hey only ail i an in e nal e o occu s) o composi e
(meaning ha a se ice depends on o he se ice(s) which can po en ially ail,
ul ima ely also causing he composi e se ice o ail).
•Each o he se ices is deployed a an abs ac in as uc u e node, ep esen ing a
se e o on a highe le el a da a cen e o o ganiza ion.
6
1.1. In oduc ion and Mo i a ion
•These nodes a e in e connec ed ia he In e ne .
•Se ices and nodes can dynamically appea and disappea again, due o he IoS
being an open sys em.
•Each o he se ices (mo e p ecisely, hei espec i e managemen componen s)
can adop diffe en oles, anging om SPs and SCs o media ing oles such as a
se ice b oke .
Fo he emainde o his hesis his abs ac scena io model is assumed as he p oblem
con ex o my disse a ion p ojec .
1.1.2. Resea ch P oblem
The IoS scena io, as desc ibed in he las subsec ion, is mo e and mo e becoming eali y.
Howe e , especially when employing his eme ging global in as uc u e o business
wo kflows, s ill se e al se ious issues emain unsol ed un il oday.
Especially he need o gua an eed eliabili y and se ice quali y becomes mo e p omi-
nen , as no longe he ques ion o “who p o ides he se ice?” ma e s bu only whe he
she is able o achie e he eques ed esul . The assu ance o such QoS e en becomes
c ucial when ex e nal se ices a e o be in eg a ed in o business c i ical wo kflows.
Howe e , in globally dis ibu ed se ice sys ems no cen al QoS con ol can be easily
implemen ed. Such sys ems inhe en ly lack any ype o con ol hie a chy, hus QoS man-
agemen mus be implemen ed in a decen alized way. Resea che s ag ee ha he mos
p omising mechanisms cu en ly a ailable o his ask a e bi- o e en mul ila e al SLAs
in which he in ol ed ansac ion pa ne s assu e each o he ce ain QoS gua an ees
(p o ide side) o financial se lemen (consume side) (Kelle e al. 2002a; Seidel e al.
2007). Such se ice con ac s ensu e se ice quali y ac oss indi idual si es and he e o e
ac oss o ganiza ional bounda ies in a decen alized way. Rep esen ing quali a i e gua -
an ees placed on se ices, SCs can benefi om SLAs because hey make non- unc ional
p ope ies o se ices p edic able. On he o he hand, SLAs enable SPs o manage hei
capaci y, knowing he expec ed quali y le els.
A c ucial phase h oughou he SLA-based se ice managemen li e cycle is he ne-
go ia ion o he espec i e SLA documen . A nego ia ion basically “cons i u e[s] he
p ocess o wo o mo e pa ies communica ing in o de o p oceed om some conflic
si ua ion o an ag eemen ” (Hude 2006, p. 13). Bichle e al. define nego ia ions as
“ he decision-making app oach used o each consensus whene e a pe son, o ganiza ion
o ano he en i y canno achie e i s goals unila e ally” (Bichle , Ke s en, and S ecke
2003, p. 312).
Based on hese defini ions a nego ia ion ep esen s a configu a ion o wo o mo e
pa ies (SP and SC in he IoS scena io), a guing abou some abs ac good (a SLA,
go e ning a espec i e se ice in oca ion). In such si ua ions he in e es s o he in ol ed
pa ies collide and lead o a conflic si ua ion conce ning he subjec o he nego ia ion
(SPs will p obably wan o deli e a low quali y o much money whe eas SCs p obably
ha e qui e opposing in en ions).
7
1. Resea ch P oblem and App oach
Design Science Resea ch P ocess
Up un il now a significan amoun o diffe en DS esea ch p ocess models ha e been
de eloped in science and indus y; Peffe s e al. (2008) gi e a good o e iew on he
mos p ominen ones. The same au ho s also a emp ed o c ea e a consolida ed p ocess
model ha is consis en wi h he ones ound in he li e a u e. I p o ides a nominal
p ocess desc ip ion o DS esea ch (as guideline o a) esea che s conduc ing DS wo k
as well as b) e iewe s when assessing i ) (Peffe s e al. 2008). This model has been
adop ed as a basis o his hesis, as i p o ides he mos s uc u ed and comp ehensi e
p ocess desc ip ion o DS cu en ly a ailable.
I comp ises six dis inc ac i i ies:
1. P oblem Iden ifica ion and Mo i a ion
2. Objec i es o a Solu ion
3. Design and De elopmen
4. Demons a ion
5. E alua ion
6. Communica ion
P oblem Iden ifica ion and Mo i a ion This s ep aims a “defin[ing] he specific e-
sea ch p oblem and jus i y[ing] he alue o i s solu ion” (Peffe s e al. 2008, p. 52). This
phase ma ks one o he mos c ucial s eps du ing any DS esea ch effo , as i iden ifies
he goals and applica ion con ex o he a i ac s o be de eloped subsequen ly. In he
con ex o IS esea ch, a p oblem is conside ed ele an whene e i can be desc ibed as
an “unsol ed and impo an business p oblem...” (He ne e al. 2004, p. 84) occu ing
in he con ex o “ he in e ac ion o people, o ganiza ions and in o ma ion echnology”
(He ne e al. 2004, p. 85).
DS esea che s always ha e o make su e ha hei esea ch a emp can be dis in-
guished om ou ine design, and is hus wo hy o a esea ch effo . He ne e al. p esen
a se o cha ac e is ics a p oblem should exhibi o i s sol ing being a legi ima e DS
p ocess (He ne e al. 2004, p. 81):
•uns able equi emen s and cons ain s based upon ill-defined en i onmen al con-
ex s.
•complex in e ac ions among subcomponen s o he p oblem and i s solu ion.
•inhe en flexibili y o change design p ocesses as well as design a i ac s.
•a c i ical dependence upon human cogni i e abili ies (e.g. c ea i i y) o p oduce
effec i e solu ions.
14

1.3. Resea ch Me hod
•a c i ical dependence upon human social abili ies (e.g. eamwo k) o p oduce
effec i e solu ions.
Hence, gi en a leas one o he abo emen ioned equi emen s is ulfilled o he p ob-
lem add essed in his hesis, he DS me hod can be conside ed app op ia e o my
disse a ion p ojec .
Objec i es o a Solu ion In he nex phase, he p oblem defini ion de eloped in he fi s
s ep is used o de i e p ecise equi emen s o he a i ac s o be de eloped. These can
ei he be quan i a i e o quali a i e, he o me defining measu able me ics “in which a
desi able solu ion would be be e han cu en ones”, he la e defining how he newly
designed a i ac will “suppo solu ions o p oblems no hi he o add essed” (Peffe s
e al. 2008, p. 55). Especially in case o quan i a i e objec i es being iden ified, his s ep
equi es he esea che o ha e ex ensi e knowledge on cu en ly a ailable solu ions o
he iden ified p oblem and hei efficiency.
Design and De elopmen This s ep ma ks he co e p ocess o all DS p ojec s, as i
comp ises he ac ual c ea ion o he a i ac , p o iding he esea che s wi h a solu ion
o he iden ified esea ch p oblem. This includes he de e mina ion o he “a i ac ’s
desi ed unc ionali y and i s a chi ec u e” (Peffe s e al. 2008, p. 55).
A lo o e y de ailed p ocess models o his phase ha e been de eloped, anging om
he adi ional Wa e all Model o So wa e Enginee ing (Royce 1987) o he i e a i e
Spi al Model (Boehm 1986).
Whiche e app oach is used, he esea che is bound o build on he common knowl-
edge cu en ly a ailable in he espec i e esea ch discipline (He ne e al. 2004, p. 80),
also called Jus ifica o y Knowledge in (G ego and Jones 2007, p. 322). On he o he
hand he esea che is u ged o apply igo ous esea ch me hods o he wo k, which can
be assessed based on he “applicabili y and gene alizabili y o he [de eloped] a i ac ”
(He ne e al. 2004, p. 88).
He ne e al. define such an a i ac o be “c ea ed o add ess an impo an o ga-
niza ional p oblem [and i being] desc ibed effec i ely, enabling i s implemen a ion and
applica ion in an app op ia e domain” (He ne e al. 2004, p. 82). They also iden ified
ou diffe en ypes o a i ac s ha can be he esul o a DS ac i i y (He ne e al.
2004, p. 77):
•Cons uc ( ocabula y and symbols)
•Model (abs ac ion and ep esen a ion)
•Me hod (algo i hms and p ac ices)
•Ins an ia ion (implemen ed and p o o ype sys ems)
G ego and Jones ca ego ize hem in o wo dis inc g oups, namely “ heo ies [o ]
imma e ial a i ac s” (cons uc s, models o me hods)” and “ins an ia ions [o ] ma e ial
a i ac s” (ins an ia ion) (G ego and Jones 2007, p. 321).
15
1. Resea ch P oblem and App oach
Demons a ion Du ing his s ep he a i ac ’s capabili y o sol e he add essed esea ch
p oblem is o be shown. Mos DS esea ch models in eg a e his phase wi h he nex
one, as bo h aim a assessing he de eloped a i ac . The diffe ence be ween bo h is
basically whe he he capabili y o he de eloped a i ac o sol e he p oblem a all is
assessed (demons a ion) o how good i does so (e alua ion). This dis inc ion allows
o a much mo e s uc u ed assessmen , especially in cases whe e a i ac s a e designed
add essing p oblems ha we e no add essed be o e a all. He e he fi s ques ion o ask
is whe he he a i ac ulfills i s equi emen s and ques ions abou he quali y o he
solu ion (e alua ion) a e seconda y. In hese cases he “ esea ch con ibu ion lies in he
no el y o he a i ac and in he pe suasi eness o he claims ha i is effec i e” (Ma ch
and Smi h 1995, p. 260).
E alua ion A e ha ing shown ha he a i ac essen ially ulfills i s pu pose, he
e alua ion s ep add esses he quali y o he designed solu ion. Acco ding o Peffe s e
al. his in ol es he “compa i[son o ] he objec i es o a solu ion o ac ual obse ed
esul s om use o he a i ac in he demons a ion” (Peffe s e al. 2008, p. 56). By
building on a se o e alua ion c i e ia (which should be de i ed om he objec i es o
he solu ion) as well as objec i e e alua ion me hods his s eps basically measu es how
good he a i ac sol es he s a ed p oblem.
Following He ne e al., all e alua ion c i e ia o IS a i ac s o igina e in he business
en i onmen hey a e supposed o be applied in; common examples o such c i e ia
a e o example comple eness, unc ionali y, consis ency o pe o mance (He ne e al.
2004, p. 85). I is up o he ac ual esea che s o iden i y he c i e ia app op ia e o he
espec i e wo k and speci y he way hose a e calcula ed based on he obse a ion made
om using he a i ac .
Pas de elopmen s ha e shown ha e en a i ac s ha we e wo se han hei compe i-
o s in he adi ional me ics no only kep being used by a significan se o cus ome s,
bu o en e en ou li ed hei “be e ” p edecesso s. Such echnology is o en called
dis up i e as opposed o he sus aining echnology i se s ou o ou class (Ch is ensen
1997). These new echnologies ob iously exhibi ed some cha ac e is ics ha we e no
pa o adi ional e alua ion in he espec i e con ex , bu ha p o ed o be e y much
impo an o e ime (a good example o his a e he 3,5” and 2,5” ha d discs). This
phenomenon can easily be obse ed in DS esea ch as well and has o be aken in o
accoun when defining he me ics o a gi en e alua ion se ing: e en i a new a i ac
ende s o be wo se han p esen ones on some scale i migh be much be e in a new,
no ye conside ed me ic, which could become e y impo an in he u u e. In such
DS effo s i is c ucial o explici ly define he cha ac e is ic(s) in which he designed a -
i ac is a ying om i s compe i o s and why his me ic can be o eseen o be o u u e
impo ance. An ac ual assessmen o his asse ion can only be made o e a significan
pe iod o ime du ing which his new me ic has o p o e i s impo ance (long- e m
e alua ion o dis up i e esea ch a i ac s).
Based on he s uc u e o he a i ac and he iden ified e alua ion c i e ia an app o-
p ia e e alua ion echnique mus be selec ed as a second s ep (see (He ne e al. 2004)
16
1.3. Resea ch Me hod
o (Buche , Riege, and Saa 2008) o examples). The mos p ominen ones, in he
con ex o DS esea ch, a e p obably building and applying o so wa e p o o ypes in
a p oduc i e en i onmen , compu e simula ions, su eys o lab expe imen s (Buche ,
Riege, and Saa 2008, p. 81) as well as analy ical (i possible) o desc ip i e e alua ions
(He ne e al. 2004, p. 86).
Communica ion The final s ep wi hin e e y DS effo should be he communica ion
o he esul s o ellow “ esea che s and o he ele an audiences, such as p ac icing
p o essionals” (Peffe s e al. 2008, p. 56). He ne e al. define he wo classes o po en ial
audiences as “ echnology-o ien ed as well as managemen audiences” (He ne e al. 2004,
p. 90), gi en he applica ion domain o IS esea ch.
The challenge occu ing in his final s ep is a) o iden i y espec i e ou le s ha
p omise a good isibili y among he add essed audiences (e.g. con e ences o jou nals)
as well as o b) iden i y all esul s in e es ing o hese g oups. This includes no only
he a i ac i sel , bu also “ he [add essed] p oblem and i s impo ance, [..] i s u ili y
and no el y, he igo o i s design, and i s effec i eness” (Peffe s e al. 2008, p. 56).
Resul s o Design Science Effo s: Design Theo ies
A ques ion, occu ing du ing he communica ion s ep o DS esea ch a he la es , is
how “design knowledge[, being he esul o any DS effo ,] can be cap u ed, w i en
down and communica ed” (G ego and Jones 2007, p. 313). An abs ac guideline o DS
esea che s is needed o s uc u ing hei achie ed esul s in a way hey can communica e
hem. G ego and Jones p oposed such a guideline by consolida ing o me wo ks on
DS heo ies om Walls e al. (Walls, Widmeye , and El Sawy 1992) and Dubin (Dubin
1978). The main idea is no o iew he designed a i ac as he p ima y esul o a DS
p ocess, bu a he ha i is no mo e han he nucleus a ound which a comp ehensi e
design heo y is o be defined. The scien ific findings o a DS p ocess a e expec ed o
be gene alizable (much mo e han he sole a i ac would be) heo ies which should be
able o ac as pa o he heo e ical basis o u u e esea ch effo s.
The p oposed s uc u e o a DS heo y iden ifies eigh diffe en ca ego ies o abs ac
componen s ha should be he esul o an ideal DS p ocess (G ego and Jones 2007,
p. 322):
•Pu pose and Scope
•Cons uc s
•P inciples o Fo m and Func ion
•A i ac Mu abili y
•Tes able P oposi ions
•Jus ifica o y Knowledge
17
1. Resea ch P oblem and App oach
•P inciples o Implemen a ion
•Exposi o y Ins an ia ion
Pu pose and Scope This aspec defines he high-le el goals o he de eloped design
heo y (in which he de eloped a i ac is g ounded / which was de i ed om he de el-
oped a i ac ). In doing so, he scope o bounda ies o he heo y, and sys ems o which
i applies espec i ely, a e shown. The equi emen s a e o be s a ed on a “me a-le el”
in ha he goal o a DS esea che should be o define a design “ heo y ha is sui ed
o a whole c1ass o a i ac s ha a e ypified by hese equi emen s” (G ego and Jones
2007, p. 325).
Cons uc s Cons uc s ep esen he basic ocabula y o any heo y, consis ing o i s
(“indica i e, a he han [comple e]”) se o “en i ies o in e es ” (G ego and Jones
2007, p. 325). I can hus be seen as an ex ended glossa y o “physical phenomena o
abs ac heo e ical e ms” (G ego and Jones 2007, p. 325) used in a gi en heo y, such
as “so wa e agen ” o “SLA”.
P inciples o Fo m and Func ion This componen desc ibes he “s uc u e, [...] shape
[...], p ope ies and unc ions” (G ego and Jones 2007, p. 326) o an a i ac . I hus
defines he “bluep in ” o indi idual ins an ia ions ollowing he espec i e design he-
o y.
A i ac Mu abili y Gi en he inhe en ly changing con ex o any IS a i ac his aspec
o a design heo y co e s he capabili ies o a de eloped a i ac when e ol ing o e ime
and adap ing o new applica ion se ings and o ganiza ional en i onmen s.
Tes able P oposi ions This componen desc ibes a se o “ es able p oposi ions o
hypo heses” (G ego and Jones 2007, p. 327) on he beha io o he de eloped a i ac .
By his means i p o ides he main inpu o assessmen o he a i ac , aking place in
he demons a ion and e alua ion s eps as p esen ed abo e.
Jus ifica o y Knowledge The heo e ical basis on which he design heo y is g ounded
is known as jus ifica o y knowledge. I hus ep esen s ela ed a i ac s and espec i e
heo ies unde lying he cu en DS effo . This concep is o be dis inguished om o he
DS p ojec s aiming a he same p oblem ( egula ly known as ela ed wo k in scien ific
pape s), as hose a e basically compe i o s o he cu en ly designed app oach whe eas
he jus ifica o y knowledge only p o ides a common g ound o all hose effo s.
P inciples o Implemen a ion This aspec desc ibes “ he means by which he design
is b ough in o being” (G ego and Jones 2007, p. 328). An abs ac guideline o imple-
men ing he design heo y in an ac ual a i ac is gi en, hus depic ing how he esul s
o he DS effo can be applied in eal-wo ld se ings.
18
1.3. Resea ch Me hod
Exposi o y Ins an ia ion An ac ually ins an ia ed a i ac implemen ing he de eloped
design heo y “con ibu es o he iden ifica ion o po en ial p oblems in a heo ized
design and in demons a ing ha he design is wo h conside ing” (G ego and Jones
2007, p. 329). I he e o e p oo s ha he heo y can be ins an ia ed and ha such
an a i ac ul ima ely sol es he add essed esea ch p oblem ( he demons a ion and
e alua ion phases o he esea ch p ocess assess he de eloped design on he basis o
such an ins an ia ion)6.
Any DS effo (such as he one p esen ed in his hesis) should aim a desc ibing he
p oposed p oblem solu ion om all hese pe spec i es in o de o allow o gene aliza ion
and applica ion o he esul s in eal-wo ld se ings as well as in u u e scien ific p ojec s.
1.3.3. Implemen a ion o he Design Science Pa adigm
Design Science P ocess
This disse a ion p ojec ollows he DS p ocess model ha was in oduced in he las
subsec ion excep o one addi ional phase conce ning he Me ics Ope a ionaliza ion
and Tes bed Implemen a ion. Such effo s a e also p esen in he model p oposed by
Peffe s e al. (Peffe s e al. 2008, p. 54), howe e , hey a e no explici ly p esen ed as a
dis inc phase, which unde ep esen s hei alue o he o e all p ocess in my opinion.
The comple e p ocess as adop ed o his hesis is shown in figu e 1.3:
As a fi s s ep, he esea ch p oblem has been iden ified (i has al eady been de ailed
in sec ion 1.1): he lack o mechanisms o a comp ehensi e and ully au oma ed dis-
co e y and flexible nego ia ion o SLAs in dis ibu ed se ings such as he IoS. I has
been deduced om a de ailed scena io model o he IoS a e in es iga ing espec i e
li e a u e on se ice-based sys ems as well as he esea ch p oblems s a ed wi hin cu en
na ional and in e na ional esea ch p ojec s.
The p oblem add essed in his hesis clea ly ulfills a leas some o he c i e ia He ne
e al. (He ne e al. 2004) impose on a DS p ojec . The complex na u e o se ice-
based se ings and he in ica e in e ac ions and dependencies be ween he diffe en
componen s o he IoS can p obably be seen as he mos p ominen one.
Nex , he objec i es o solu ion o he espec i e p oblem a e s a ed in 2.1. They
p ima ily o igina e in a ho ough analysis o scien ific li e a u e abou he s a ed scena io
and esea ch p oblem.
A e he objec i es o his DS p ojec a e defined, wo basically independen phases
can begin: he design and de elopmen o he ac ual a i ac (i.e. he SLA managemen
in as uc u e) as well as he defini ion o he e alua ion me ics and he es bed imple-
men a ion. These wo phases ha e no ac ual con ac poin s, so hey can be conduc ed
independen ly. The only equi emen ega ding he o e all p ocess is ha he es bed,
wi hin which he designed and implemen ed a i ac is o be assessed, mus be finished
be o e s a ing he demons a ion and e alua ion phases. The es bed used wi hin his
6Ha ing p oduced an exposi o y ins an ia ion does no necessa ily ende a design effo o be alid
DS, as in he case o jus ha ing such an a i ac wi hou an app op ia e heo y o design “ he le el
o knowledge is ha o a c a -based discipline” (G ego and Jones 2007, p. 329).
19

1. Resea ch P oblem and App oach
P oblem
Iden i ica ion
and Mo i a ion
Objec i es o
a Solu ion
Design &
De elopmen
Demons a ion
E alua ion
Communica ion
Me ics
Ope a ion-
aliza ion
& Tes bed
Implemen a ion
dis ibu ed SLA
managemen
in he IoS
p o ocol-gene ic
and au oma ed
disco e y and
nego ia ion o
SLAs
p o ocol
desc ip ion
language and
agen -based
in as uc u e
agen s ac ing
in di e en
nego ia ion
se ings
(e ec i eness)
SimIS Toolki
disse a ion
hesis and
scien i ic
publica ions
quali y o he
solu ion
(e iciency)
Figu e 1.3.: Resea ch P ocess applied o his Thesis
hesis is he IoS simula ion oolki SimIS (“Simula ing an In e ne o Se ices”)7(K¨onig,
Hude , and Eymann 2010), which I co-de eloped. In pa allel, he me ics o be used
wi hin he demons a ion and e alua ion phases, a e de i ed on he basis o he esea ch
ques ion and scena io defini ion a hand.
Du ing he design and de elopmen o he ac ual a i ac , he mechanisms and da a
s uc u es needed o sol ing he s a ed esea ch p oblem a e defined and subsequen ly
implemen ed in a p oo -o -concep p o o ype.
Once bo h his exposi o y ins an ia ion as well as he es bed a e finished, he demon-
s a ion and e alua ion s eps a e unde aken. To his end, he de eloped so wa e com-
ponen s a e deployed in a simula ed IoS scena io (as pa ame e ized using SimIS) and
subsequen ly assessed ega ding he equi emen s defined be o e.
Following Buche e al., simula ion is a alid e alua ion (and demons a ion) echnique
wi hin DS whene e he applica ion o a p o o ype in a eal-wo ld expe imen is impos-
sible (see o example Buche , Riege, and Saa 2008; Hude , Niemann, and Eymann
2010). This can be due o p agma ic (e.g. in es iga ions on fluid beha io in he co e o
7h p://sou ce o ge.ne /p ojec s/simis
20
1.3. Resea ch Me hod
he sun), heo e ic (e.g. wha -i ques ions on diffe en alues o na u al cons an s) o
e hical easons (Ha mann 1996, p. 87). In he case o his p ojec a p agma ic eason
is p esen , as a global IoS as I en ision i , is s ill only a ision o u u e sys ems and
no al eady implemen ed eali y in IS. Ini ial s eps owa ds his ision can al eady be
obse ed in cu en in as uc u es ( o example SAP’s BusinessByDesign8, Enomaly’s
Spo Cloud9o SalesFo ce10), bu he comp ehensi e IoS ision in i s en i e y is no
p esen ye . Simula ion he e o e ep esen s an app op ia e e alua ion ool o explo ing
hese scena io se ings, “ ha canno (ye ?) be in es iga ed . . . by expe imen al means”
(Ha mann 1996, p. 87).
As a concep ually las s ep he esul s o his p ojec a e communica ed in he o m
o scien ific publica ions (see Hude (2006, 2009, 2010); Hude and Eymann (2010,
2011a,b); Hude , Ludwig, and Wi z (2006, 2007, 2008, 2009); Hude e al. (2009)) in
bo h, he compu e science and business communi ies, con ibu ions o conjoin esea ch
p ojec s as well wi h he publica ion o his hesis as a whole.
Resul ing Design Theo y
In o de o desc ibe all ele an inpu and (in e media e) esul s o my esea ch p ojec ,
he emainde o his hesis is p ima ily s uc u ed acco ding o he esea ch p ocess jus
ske ched. Ne e heless, I also ied o inco po a e G ego and Jones’ s uc u e o a design
heo y when p esen ing my esul s. In he ollowing I e y sho ly wan o p o ide some
o e iew in o ma ion on whe e in his hesis which o he men ioned heo y componen s
can be ound. These ema ks can hus ac as an al e na i e app oach o an ou line o
his hesis and a e supposed o guide he eade whene e she is pa icula ly in e es ed
in a dis inc aspec o he heo y esul ing om my wo k.
•The Pu pose and Scope o my wo k can be ound in sec ion 1.1 and mo e pa icu-
la ly in 1.1.1, whe e my ac ual esea ch goal is p esen ed.
•Nex , he Cons uc s used wi hin his hesis a e implici ly in oduced in sec ion
2.2. In his sec ion also he Jus ifica o y Knowledge is desc ibed as an inpu o
my wo k and whene e a dis inc cons uc , ha is needed in he u he hesis, is
in oduced, i is highligh ed espec i ely.
•The main pa o his hesis, he design o he SLA managemen in as uc u e in
chap e 3 basically comp ises he P inciples o Fo m and Func ion, whe eas he
Tes able P oposi ions can be ound in 4 as pa o he demons a ion and e alua ion
expe imen s.
•Also loca ed in chap e 4 is he desc ip ion o he implemen ed p oo -o -concep
p o o ype (Exposi o y Ins an ia ion).
8h p://www.sap.com/ge many/sme/solu ions/businessmanagemen /businessbydesign/index.epx
9h p://www.spo cloud.com/
10h p://www.sales o ce.com
21
1. Resea ch P oblem and App oach
•The las wo aspec s o a design heo y a e conce ned p ima ily wi h he ime
a e he ac ual esea ch p ocess and a e hus men ioned in chap e 5: A sho
guideline o how o use my sys em in a p oduc i e en i onmen is gi en in 5.3
(P inciples o Implemen a ion) and finally he sys em’s capabili y o cope wi h
changing en i onmen s and po en ial ex ensions o he sys em o inc ease his
capabili y a e ske ched in 5.4 (A i ac Mu abili y).
22
2. Objec i es and Founda ions
In his chap e he objec i es o my wo k as well as he concep ual ounda ions upon
which his hesis builds a e p esen ed.
2.1. Requi emen s Analysis
In in es iga ing he iden ified esea ch p oblem and he esul ing esea ch ques ion, a
se o equi emen s o he in ended solu ion can be iden ified. They a e s uc u ed
acco ding o he phases o he o e all se ice li e cycle (see figu e 2.1) hey e e o, i.e.
he disco e y o nego ia ion phase, as hose a e he main ocus o his hesis. Wi hin he
nego ia ion phase equi emen s ano he le el o ca ego iza ion is in oduced, ollowing a
well-known dis inc ion o nego ia ion esea ch: nego ia ion objec ( e e ing o he en i y
nego ia ed abou ), decision making s a egy ( e e ing o he way nego ia o s ac du ing
a nego ia ion) and nego ia ion p o ocol (defining he way nego ia o s can communica e
du ing nego ia ion) (Jennings e al. 2001, pp. 200-201):
Disco e y Phase: Du ing he Disco e y Phase, a sys em o ma ke egis ies is needed
o suppo he disco e y and publica ion o se ice offe ings. This ensu es effec i e and
efficien disco e y p ocesses in which po en ial ansac ion pa ne s a e made known o
each o he , a p e equisi e o any economic ansac ion.
R1 Ma ke egis ies should be p esen o he se ice disco e y phase (Buyya e al.
2009).
Nego ia ion Phase - Nego ia ion Objec : Nex he equi emen s o he nego ia ion
phase a e p esen ed, beginning wi h he ones conce ning he nego ia ion objec ; in he
IoS hese objec s a e se ice in oca ions and he e o e SLAs. Due o he a ie y o
diffe en se ices he complexi y o such se ice in oca ions, espec i e SLAs o highly
a ying complexi y mus be nego iable.
R2 Se ice managemen in he IoS should be based on (elec onic) SLAs o a ying
complexi y (Ba os, Dumas, and B uza 2005; Blau, S ¨osse , and Block 2008; Buyya
e al. 2009; Ludwig e al. 2006; Neumann, S ¨osse , and Weinha d 2007; Neumann
e al. 2008; Pau obally, Tamma, and Woold idge 2007).
On he o he hand, in o de o delimi he po en ial ag eemen space a p io i ( o
example as needed in fixed p ice p o ocols) a possibili y o define some SLA e ms o
be non-nego iable is needed. These co espond o ha d-coded se ice cha ac e is ics,
23
2. Objec i es and Founda ions
on-demand [...]. Elec onic con ac s speci y he way how hese in e ac ions a e ca ied
ou and which con ac ual pa ies a e in ol ed. An impo an aspec o a con ac o
IT se ices is he se o Quali y o Se ice (QoS) gua an ees and he obliga ions o he
a ious pa ies. This is commonly e e ed o as a Se ice Le el Ag eemen (SLA)”
(Kelle and Ludwig 2003, p. 58).
Building on hese defini ions, he cons i uen cha ac e is ics o a SLA can be defined
as:
•A SLA ep esen s some so o nego ia ed ag eemen be ween he p o ide and
consume o a se ice.
•I specifies he gua an eed cha ac e is ics o a gi en se ice in oca ion in e ms
o measu able se ice aspec s, such as a ailabili y o pe o mance.
•Addi ionally a se o o ganiza ional asse ions, such as he cos associa ed wi h a
se ice in oca ion o he penal ies o no deli e ing he gua an eed se ice quali y,
is defined.
•SLA documen s in his ein a e used o c oss-o ganiza ional as well as in e nal
se ice deli e ies.
Hence, he SLA concep as employed h oughou his hesis is defined as ollows.
Defini ion 2.8 A SLA is a s uc u ed documen , desc ibing a nego ia ed, bila e al4
ag eemen be ween a SP and a SC on he e ms and condi ions o he in oca ion(s)
o a (se o ) (elec onic) se ice(s). This ag eemen bo h obliges he SP o deli e o
he SC he espec i e se ice(s) in he s a ed quali y and he SC o ecip oca e his wi h
a defined compensa ion paymen . Hence, a SLA con ains a desc ip ion o he deli e ed
se ices’ unc ionali y, gua an ees on he deli e ed se ice quali y, as well as asse ions
on he associa ed cos s and penal ies in case o a iola ion. The se ice quali y gua an ee
is u he defined as a se o Se ice Le el Objec i es (SLOs). Finally, a SLA is always
alid o a dis inc pe iod o ime.
Hence, a SLA documen basically desc ibes a empo a y business ela ionship be ween
a SP and a SC, du ing which a defined se o se ices is deli e ed. I hus helps answe ing
he ques ions “Who is p o iding he se ice(s) o whom?”, “Wha se ices a e p o ided
and unde which condi ions?” and “Wha consequences a ise om ei he a se ice being
deli e ed as gua an eed o no ?”.
When defining hese espec i e QoS le els wi hin a SLA documen , SLO elemen s a e
used:
Defini ion 2.9 A Se ice Le el Objec i e (SLO) desc ibes one pa icula QoS aspec
wi hin a SLA documen , hus consis ing o a commonly known and measu able se ice
cha ac e is ic oge he wi h a espec i e a ge alue. This a ge alue hen ep esen s
he QoS gua an ee o he espec i e se ice cha ac e is ic.
4Also mul i-la e al SLAs a e possible. Howe e , hey only play a mino ole in se ings, such as he
IoS, and a e he e o e no u he conside ed in his hesis.
30

2.2. Concep ual Founda ions
These defini ions al eady show a undamen al gap be ween a SLA’s business- ela ed
ole and i s implica ions on he echnical in as uc u e. On he one hand i is used as
a con ac ual ag eemen be ween SP and SC, wi hin o e en ac oss en e p ise bound-
a ies, wi h all esul ing equi emen s om a legal pe spec i e. On he o he hand i
egula ly ac s as an inpu o he, o en e en au oma ed, managemen o he echnical
IT in as uc u e deli e ing he eques ed se ices. I can clea ly be seen, ha bo h
usage scena ios pose qui e diffe en equi emen s on he SLA documen and i s ole in
he o e all se ice managemen p ocess.
Some scien is s e en acc edi his ac in defining a SLA only being a pa o a com-
p ehensi e con ac be ween a SP and a SC (Ka aenke and Ki n 2007; Leff e al. 2003),
ocusing on he “ope a ional defini ion o a se ice” (Ka aenke and Ki n 2007, p. 104).
Simila ly, hey dis inguish be ween SLAs a ge ing end use s and hose a ge ing o he
SPs (Be ge 2005, pp. 29-32).
In he ollowing I will add ess his issue by sho ly assessing he abs ac SLA concep
as unde lying his hesis in he ligh o bo h high-le el business ela ionships and he
e y p ecise ask o au oma ed SLA-based esou ce managemen .
SLAs as Ins umen s o Business Rela ionship Managemen
When in es iga ing SLAs om a business pe spec i e, scena ios like applica ion se ice
p o isioning, web hos ing o IT ou sou cing a e o p ima y in e es (Kelle and Ludwig
2003). All hose se ings a e cha ac e ized by bipa i e ela ionships be ween SPs and
SCs, whose cha ac e is ics a e egula ly defined in e ms o a SLA documen (Goo,
Kisho e, and Rao 2009). I subsequen ly ac s as a legally binding con ac be ween he
in ol ed pa ies, deno ing he igh s, obliga ions and compensa ion paymen s associa ed
wi h he espec i e business ansac ion.
The ongoing end o concen a ing on he co e business and hus ou sou cing a lo o
he suppo ing unc ions (such as mos pa s o he IT in as uc u e) in an en e p ise o
specialis p o ide s cons an ly inc eases he need o SLAs as a ounda ion o such busi-
ness ela ions. Addi ionally, he echnical possibili ies o mo e and mo e fine-g ained
defini ion (and hus p o iding and pu chasing) o indi idual se ices was one o he main
business- ela ed d i e s o he IoS as a model o u u e en e p ise IS.
SLA documen s play a e y undamen al ole in such IT ou sou cing scena ios (Masche,
Mckee, and Mi chell 2006). When elying on ex e nal se ices o one’s own economic
pe o mance, some deg ee o planning eliabili y becomes necessa y. This can be ealized
wi h he binding gua an ees s a ed in a SLA.
On he o he hand, mos companies find hemsel es in a posi ion whe e hey no only
pu chase ex e nal se ices, bu also sell hose o ex e nal cus ome s (po en ially e en by
combining hem wi h in e nal ones). The o me ly in e nal and e y echnology-o ien ed
IS o a company is inc easingly mo ing owa ds a cus ome -o ien ed in as uc u e o
deli e ing se ices a leas as pa s o an economic ansac ion (Boehmann and K cma
2004).
In such cases, a SLA documen helps he espec i e en e p ise (ac ing as a classical
SP in his case) o plan and alloca e hei in e nal esou ces in a way ha he p omised
31
2. Objec i es and Founda ions
se ice gua an ees a e me . Se e al bes p ac ice amewo ks o managing such en e -
p ise IS in an efficien way ha e been published in he ecen yea s, he IT In as uc u e
Lib a y (ITIL) (Office o Go e nmen Comme ce 2007) and Con ol Objec i es o In-
o ma ion and Rela ed Technology (COBIT) (IT Go e nance Ins i u e 2007) p obably
being he wo mos p ominen ones. Especially ITIL defines Se ice Le el Managemen
as he undamen al se o managemen ac i i ies conce ned wi h he nego ia ion, mon-
i o ing and en o cemen o se ice deli e y p ocesses acco ding o ag eed-upon SLAs
(Maye l e al. 2005; Schaa 2008).
The ole o a SLA, om a business pe spec i e, is hus qui e clea : i defines all igh s
and obliga ions o an ac ual business ansac ion and hus ep esen s he legal basis o
any in e ac ion be ween SC and SP. I has o co e a lo o e ms and elemen s ela ed
o hese legal equi emen s in addi ion o he ac ual se ice (quali y) desc ip ion. E en
SLAs be ween indi idual depa men s wi hin a la ge coope a ion ela e o his concep ,
howe e hey will egula ly s a e in e nal p ices and be able o omi some o he legal
aspec s due o he o e a ching legal amewo k o he pa en company.
Finally, SLAs allow a SP o ully exploi he economic measu e o p ice diffe en ia ion
when defining indi idual se ice (quali y) le els and hei associa ed p ices. Exploi ing
diffe en u ili y unc ions on he SC sides, a SP can hus maximize i s p ofi , gi en i is
possible o en o ce diffe en quali y le els du ing he ac ual se ice deli e y p ocess.
SLAs as Ins umen s o (Au oma ed) Resou ce Managemen
In con as o his e y business-cen ic iew on SLAs, such documen s also play an im-
po an ole in he (po en ially au oma ed) managemen o echnical IS in as uc u es.
Fi s a emp s o apply SLAs o au oma ed esou ce managemen ha e been done in
he elecommunica ions indus y (see o example Chak a o y e al. 2003; Man a e
al. 2006) and subsequen ly in, mainly non-comme cial, g ids (see o example Pa kin,
Badia, and Ma a 2008; Seidel e al. 2007)5.
The main ocus he e is o ex ac he SLOs s a ed in he SLA, de i e espec i e
equi emen s posed on he managed in as uc u e, calcula e how his in as uc u e
mus be o ches a ed in o de o ulfill hem and finally implemen he esul .
An impo an pa o his ask is ma ked by he scheduling o incoming eques s and
hei assignmen o he a ailable se ices o aw compu ing and s o age esou ces (Seidel
e al. 2007). Fu he mo e, a SLA can e en be employed o well- ounded p edic ions on
po en ial iola ions and subsequen en o cemen measu es (Padge , Djemame, and Dew
2005).
In o de o allow such au oma ed managemen o IT se ices a digi ally p ocessable
e sion o a SLA is needed (Leff e al. 2003).
Defini ion 2.10 An elec onic SLA is a digi al equi alen o a pape -based SLA, s uc-
u ing i s con en s in o machine- eadable elemen s.
5Going back e en u he , e y ini ial wo ks on objec -o ien ed so wa e enginee ing al eady applied he
concep o a con ac as he main communica ion concep be ween indi idual so wa e componen s,
go e ning he deli e ed unc ionali y jus as wi h oday’s SLAs (Meye 1992).
32
2.2. Concep ual Founda ions
The main ocus o such an elec onic SLA hus lies in he in o ma ion needed o
he au oma ed scheduling and esou ce alloca ion, such as he s a ed SLOs, associa ed
cos s and penal ies as well as in ol ed pa ies (as needed o example in sys ems wi h
indi idual use s p io i ies). Hence, an elec onic SLA aims a closing he gap be ween he
high-le el business con ex , as defined in a SLA be ween he in ol ed business pa ne s,
and he e y echnical aspec s o he unde lying IS.
This aspec becomes e en mo e p ominen as cu en dis ibu ed and high-pe o mance
compu ing in as uc u es mo e and mo e e ol e om pu ely academic and non-p ofi o
comme cial in as uc u es (Leff e al. 2003). In such cases SLA-based se ice p o ision-
ing and he possibili y o offe “diffe en se ice le els o diffe en clien s by dynamically
alloca ing esou ces o execu ion o indi idual [...] se ice eques s” (Dan, Ludwig, and
Pacifici 2003, p. 1) becomes inc easingly c ucial o economic success.
As opposed o he business pe spec i e p esen ed in he las subsec ion, om a echni-
cal poin o iew a SLA simply has o define he gua an eed se ice quali y and unc ion-
ali y (in e ms o machine in e p e able se ice me ics). This allows au oma ed se ice
managemen sys ems o calib a e he a ailable IT esou ces in a way ha he o e all
u ili y o he SP is maximized, based on he cu en ly ac i e SLAs and hei associa ed
compensa ion paymen s and penal ies.
An In eg a ed View on SLAs
This dis inc ion be ween in e -o ganiza ional, legally binding business con ac s and in-
e nal ag eemen s o he (au oma ed) managemen can also be ound in he ITIL ame-
wo k. I dis inguishes ex e nal con ac s, called SLAs in ITIL, om in e nal ag eemen s,
which a e much mo e ela ed o he unde lying IS in as uc u e, called Ope a ional Le el
Ag eemen s (OLA) (Schaa 2008, p. 46).
When aking a close look on hese wo concep s, as well as he s a emen s made in he
las wo subsec ions, i becomes qui e ob ious on an abs ac le el ha hey do no diffe
much. In bo h cases, an ag eemen documen is c ea ed, s a ing he in ol ed pa ies,
he unc ional and non- unc ional aspec s o he se ice o be exchanged as well as a se
o o ganiza ional and o he ( o example legal) asse ions on he espec i e ansac ion.
Cu en p ojec s, especially in he domain o high-pe o mance compu ing, a e e en
wo king on mechanisms o au oma ically map SLAs o OLAs, di ec ly employed o he
au oma ed in as uc u e managemen (Hasselmeye e al. 2006b).
This shows he s uc u al equi alence o bo h ag eemen ypes. On an abs ac iew
a SLA hus consis s o a se o elemen s. Be ge gi es a e y ex ensi e o e iew on such
elemen s, co e ing bo h he IT- ele an se ice aspec s and he legal and o ganiza ional
elemen s (Be ge 2005, p. 68).
He defines a SLA o con ain:
•con ac ual elemen s: defining he in ol ed pa ies and legal condi ions.
•se ice- ela ed elemen s: speci ying he unc ional and QoS asse ions o he deli -
e ed se ice, along wi h he associa ed cos s.
33
2. Objec i es and Founda ions
•managemen - ela ed elemen s: mainly de ailing he applied moni o ing mecha-
nisms and penal ies.
Especially he second, bu also he hi d, se o elemen s could po en ially be employed
di ec ly as OLAs (gi en hey can be ep esen ed in a machine-p ocessable way) o he
managemen o he IT se ices in ol ed.
Gi en his equi alency o in e nal and ex e nal SLAs and also aking in o accoun he
abs ac ion le el o my esea ch goal I will no dis inguish be ween hose wo concep s
o he emainde o his hesis. In he con ex o an au oma ed SLA nego ia ion in-
as uc u e, he p ima y equi emen posed on a SLA documen as he objec unde
nego ia ion, is ha i mus consis o a se o defined elemen s. Each o hose mus
ha e a defined iden ifie (ID) and a alue, which is unde nego ia ion. Consequen ly
hese elemen s can ei he con ain simple me ics and hei alues o e en complex legal
exp essions on compensa ion paymen s in case o a SLA iola ion.
Backing up his assump ion, he IoS scena io u he blu s he dis inc ion be ween
in e nal and ex e nal se ices and hus ag eemen s abou hei in oca ions. E e y unc-
ionali y offe ed wi hin o ou side a gi en o ganiza ion can be pu chased and in eg a ed
in o a gi en wo kflow. Subsequen ly he documen desc ibing he espec i e ansac ion
will ul ima ely no diffe any mo e om in e nal o c oss-o ganiza ional in oca ions.
No e ha a he momen SLA documen s c ossing en e p ise bounda ies a e nego i-
a ed manually and signed by human beings. An in as uc u e as defined wi hin his
hesis add esses he eme ging need o au oma ed se ice managemen om a echnical
pe spec i e. Such an app oach howe e implies ha so wa e agen s, used as nego ia-
o s, mus be able o de i e legally binding con ac s in o de o implemen an IoS as
en isioned.
This aspec is cu en ly hea ily discussed by lawye s and legisla i e ins i u ions, as i
demands a significan change o he body o laws o a leas hei in e p e a ion by he
in ol ed judges (see o example We ig and Zehendne 2004).
Fo my hesis I will assume ha a solu ion o his p oblem will be ound. This is
ques ionable, howe e ini ial app oaches o elec onic nego ia ions c ossing company
bounda ies a e al eady applied in some isola ed indus ies. He e, a amewo k con ac
go e ns he o e all ede a ion o companies aking pa in his “isola ed” se ice economy.
This amewo k con ac p o ides he legal basis o so wa e agen s o ac ually conclude
legally binding con ac s.
Gi en such de elopmen s and he p og ess achie ed in compu e science and ju isp u-
dence in ha a ea indica e ha he assump ion o legal con o mi y o au oma ed nego-
ia ions can be assumed alid.
Desc ip ion Languages o Elec onic SLAs
A SLA documen ma ks an impo an inpu o he configu a ion o an elec onic se ice
which is pa o a business ansac ion. In o de o allow he au oma ed nego ia ion and
managemen o such ag eemen s a s uc u ed digi al documen ype is needed, speci ying
how he SLA elemen s can syn ac ically be exp essed, a SLA desc ip ion language.
34
2.2. Concep ual Founda ions
Such a digi al documen schema egula ly also allows o he defini ion o SLA em-
pla es (Ludwig e al. 2005).
Defini ion 2.11 A SLA empla e is a no comple ely filled ou SLA documen , which is
o be finalized du ing a nego ia ion p ocess. In doing so, a SLA empla e defines he SLA
elemen s, ha a e unde nego ia ion, op ionally gi es s a ing alues o hose and finally
s a es he ules, ollowing which he final SLA can be de i ed du ing he nego ia ion.
A empla e hus ma ks he s a ing poin o e e y SLA nego ia ion, p o iding he
SC wi h offe ed QoS me ics along wi h ini ial alues offe ed by he SP. Based on hese
alues he SC can hen c ea e a coun e offe o accep he offe ed SLA di ec ly, de-
pending on he nego ia ion p o ocol applied. Addi ionally, SLA empla es can be used
o implemen efficien disco e y p ocesses, as hey can be published and e ie ed by
po en ial ansac ion pa ne s acco ding o many disco e y p o ocols o elec onic da a
( he mos p ominen o which will be p esen ed in sec ion 2.2.2).
In he ollowing he mos well-known SLA defini ion languages a e sho ly ske ched.
SLAng The Depa men o Compu e Science a he Uni e si y College London de el-
oped a language o ep esen ing elec onic SLAs, building on he Ex ensible Ma kup
Language (XML): he SLAng language 6. SLAng aims a p o iding a clea defini ion o
he quali y a ibu es associa ed wi h a digi al se ice.
In con as o some o he o he languages p esen ed in he ollowing, SLAng only
specifies non- unc ional QoS a ibu es o a se ice, he unc ional aspec s a e no co -
e ed wi h his language. Hence SLAng SLAs con ain elemen s s a ing he con ac ual
pa ne s, con ex s a emen s, such as du a ion o he ag eemen and he ac ual se ice
gua an ees, called Se ice Le el Specifica ions (SLS) (Skene, Lamanna, and Emme ich
2004) in SLAng.
The SLAng language was de i ed using an Applica ion Se ice P o ide (ASP) e -
e ence model subsequen ly defining which ypes o SLAs a e a ailable wi hin SLAng.
In addi ion o e ical SLAs go e ning se ice deli e y om mo e echnical laye s ( o
example communica ion) o highe -le el ones ( o example applica ion se ices) SLAng
also p o ides means o ho izon al SLAs go e ning subcon ac ing o se ices on he
same laye (Skene, Lamanna, and Emme ich 2004).
Based on his a chi ec u al scena io SLAng exposes a wo-laye design: he language
syn ax is defined in e ms o a Unified Modelling Language (UML) 2.0 (2005) model,
whe eas he seman ic o he SLA documen s a e s a ed in e ms o Objec Cons ain
Language (2010) cons ain s. Based on hese cons uc s SLA designe s can speci y a
numbe o diffe en se ice me ics o each o he a ailable SLA ypes, such as pe o -
mance, a ailabili y o secu i y (Lamanna, Skene, and Emme ich 2003).
Finally, SLAng allows o he de ailed defini ion o iming cons ain s o each se ice
gua an ee, so-called schedules, desc ibing which gua an ee is o be alid a wha poin in
ime. This e y gene ic iming concep and he o mal seman ics jus men ioned allow
6h p://uclslang.sou ce o ge.ne /index.php
35

2. Objec i es and Founda ions
Figu e 2.2.: SLAng “ e e ence model o Applica ion Se ices P o isioning” (Skene,
Lamanna, and Emme ich 2004, p. 182)
o he au oma ed alida ion and in eg a ion o diffe en SLA documen s. Addi ionally,
no only he ac ual se ice p ope ies, bu also a ailable moni o ing and backup solu ions
as well as possible in e ac ion beha io s o he in ol ed pa ies can be modeled building
on he exp essi eness o he UML language (Skene, Lamanna, and Emme ich 2004).
Howe e he designe s seem o ha e s opped wo king on ha language in 2006 which
makes i no e y easonable o ely on his concep o u u e SLA documen s, especially
since he au ho s o he co esponding websi e hemsel es ad ice SLA designe s no o7.
WSOL Ano he SLA language was designed by Tosic e al. and is called he Web Se -
ices Offe ing Language (WSOL) (Tosic, Pa el, and Pagu ek 2002). As al eady hin ed in
he name, his SLA language explici ly aims a suppo ing he Web Se ices echnologies
and he e o e is compa ible wi h ela ed s anda ds such as WSDL.
WSOL assumes web se ices o be offe ed in a ying classes, ma king a disc e e con-
figu a ion o he se ice along wi h i s unc ional and non- unc ional a ibu es. Taking
his in o accoun a se ice offe ing, which can be exp essed using he WSOL mecha-
nism is “a o mal ep esen a ion o one class o se ice o one web se ice” (Tosic e al.
2002, p. 1). WSOL documen s hus s a e a gi en se ice class and consequen ly he SLA
associa ed wi h he se ices o ha class.
In doing so, WSOL offe s a se o language cons uc s o build a SLA documen , such
as cons ain s,s a emen s and cons ain g oups. The cons ain s can be u he sub-
ca ego ized in o unc ional (such as p e- o pos -condi ions o some se ice in oca ion)
and non- unc ional ( o example pe o mance o a ailabili y o a se ice) cons ain s.
A s a emen is basically some impo an in o ma ion abou he espec i e se ice ha
canno be coded as a cons ain , o example he p ice o a gi en se ice.
Finally, a cons ain g oup allows o he, po en ially ecu si e, g ouping o indi idual
cons ain s, allowing o example he assignmen o diffe en moni o ing uni s o each o
7h p://uclslang.sou ce o ge.ne /index.php
36
2.2. Concep ual Founda ions
he g oups. Fo such g oups WSOL also offe s a empla e cons uc .
Building on his e y low le el language he au ho s also designed mechanisms o a
s uc u ed Web Se ices Composi ion Managemen (WSCM) as well as a managemen
in as uc u e, applying WSOL documen s o dis ibu ed se ice managemen (Tosic
e al. 2004).
WSLA One o he mos comp ehensi e SLA languages o da e is he Web Se ice
Le el Ag eemen (WSLA) (Ludwig e al. 2003b) language de eloped by IBM8.This
app oach aimed a p o iding a specifica ion o he defini ion and moni o ing o SLAs
in a Web Se ice en i onmen . Employing such a language an au oma ed managemen
o elec onic se ices based on he abs ac no ion o a con ac was en isioned (Kelle
e al. 2002b).
The main design p inciples o WSLA a e an unambiguous specifica ion o Web Se ice
SLAs, enabling an au oma ed moni o ing o he espec i e se ices, ease o SLA c ea ion
by in oducing SLA empla es and offe ing XML Schema specifica ions o he WSLA
language as well as he inco po a ion o a dis ibu ed moni o ing amewo k, in which
ele an moni o ing and e alua ion asks can be ou sou ced o independen hi d pa y
moni o ing se ices (Kelle and Ludwig 2003).
To his end, WSLA no only defines he SCs and SPs (so-called p ima y pa ies)o
he SLA bu also seconda y pa ies suppo ing he enac men o he con ac , so called
suppo ing pa ies. Such pa ies can be ep esen ed by measu emen o condi ion e alua-
ion se ices, implemen ing he moni o ing o he SLA o managemen se ice esponsible
o aking espec i e managemen ac ions once a iola ion has occu ed (Ludwig e al.
2003a).
A WSLA documen hus no only desc ibes he se ice ela ed me ics, bu also how
hey a e supposed o be measu ed (measu emen di ec i es)incaseo awme icsand
how hey can be compu ed in case o highe -le el me ics (in e ms o agg ega ion unc-
ions o e lowe -le el pa ame e s) (Ludwig e al. 2003a).
Finally, a SLA adhe ing o he WSLA specifica ion con ains a se o obliga ions s a ing
heSLOsandac ion gua an ees o he espec i e SLA. A SLO ep esen s a gua an ee
conce ning one o mo e se ice me ics as opposed o an ac ion gua an ee which defines
a p omise o pe o m an ac ion unde a gi en condi ion.
Up o da e he desc ip ion o SLA elemen s conce ning he alidi y o a gi en gua an ee
(in WSLA implemen ed as alidi y pe iods), he me ics i builds on, how o measu e
o calcula e hem as well as quali ying condi ions on hei measu emen , has no been
designed in a mo e de ailed way as done wi h he WSLA language. Ne e heless WSLA
is mos ly ocused on he sole pu pose o desc ibing a SLA documen and does no co e
how o c ea e o e en nego ia e one.
Hence, a new SLA language has been de eloped based on he WSLA ideas, and pa ly
by he same au ho s ha ep esen s he cu en ly mos widely used SLA language in
se ice based sys ems: Web Se ices Ag eemen (WS-Ag eemen ). This s anda d simpli-
fies he WSLA da a model and accompanies i wi h a se o defined p o ocol p imi i es
8h p://www.ibm.com/us/en/
37
2. Objec i es and Founda ions
Ag eemen
Laye
Se ice
Laye
Ag eemen
Fac o y
Fac o y Applica ion
Ins ance
Responde
P o ide Consume
Ini ia o
inspec ()
c ea e()
c ea e()
oo()
Figu e 2.3.: WS-Ag eemen A chi ec u al Model (And ieux e al. 2007, p. 12)
and in e aces o ag eemen c ea ion and moni o ing.
WS-Ag eemen The WS-Ag eemen specifica ion is a s anda diza ion effo conduc ed
in he OGF in o de o acili a e c ea ion and moni o ing ag eemen s be ween a SP and
a SC. The specifica ion d a (And ieux e al. 2007) defines an XML ep esen a ion o
ag eemen s and ag eemen empla es, a simple ag eemen es ablishmen p o ocol as well
as co esponding in e aces o c ea ing an ag eemen and moni o ing i a un ime.
WS-Ag eemen defines wo oles in c ea ing ag eemen s: ag eemen ini ia o and
ag eemen esponde . These wo oles a e comple ely independen om SP and SC,
as bo h he consume and he p o ide shall be able o ini ia e he ag eemen c ea ion
p ocess.
WS-Ag eemen depic s a laye ed se ice model consis ing o wo laye s: he se ice
laye and on op o ha he ag eemen laye .These ice laye ep esen s he domain-
and applica ion specific pa o he p oposed a chi ec u e. I con ains he ac ual se ices
he ag eemen s a e c ea ed o . “The ag eemen laye p o ides a Web se ice-based
in e ace ha can be used o c ea e, ep esen and moni o ag eemen s” (And ieux e
al. 2007, p. 12). In o de o acili a e ag eemen c ea ion, ag eemen empla es can be
offe ed jus as in he WSOL app oach.
In o de o c ea e a SLA, he ag eemen ini ia o p oposes an ag eemen , op ionally
de i ed om a empla e. The ag eemen esponde hen checks he offe ed ag eemen
and decides o accep o ejec he offe acco ding o i s esou ce si ua ion.
Ag eemen s and ag eemen empla es a e bo h defined using he XML language. The
high-le el elemen s a e illus a ed in figu e 2.4.
38
2.2. Concep ual Founda ions
Figu e 2.4.: WS-Ag eemen SLA Model (And ieux e al. 2007, p. 14)
E e y ag eemen and ag eemen empla e is iden ified by an ag eemen id a ibu e.
This id has o be unique be ween he ag eemen ini ia o and esponde . The main body
o any ag eemen consis s o an op ional name elemen used o human unde s andabili y,
a con ex sec ion and he ag eemen e ms. Ag eemen empla es addi ionally con ain a
c ea ion cons ain sec ion (And ieux e al. 2007).
The con ex sec ion con ains in o ma ion abou ini ia o and esponde o he ag ee-
men along wi h o he me ada a conce ning he ag eemen as a whole. The ag eemen
e ms on he o he hand define he main pa o an ag eemen . They deno e he obliga-
ions o he in ol ed pa ies esul ing om he ag eemen . The WS-Ag eemen defines
wo ypes o e ms: se ice e ms and gua an ee e ms (GTs).
Se ice e ms “p o ide in o ma ion needed o ins an ia e o o he wise iden i y a se ice
o which his ag eemen pe ains and o which gua an ee e ms can apply” (And ieux
e al. 2007, p. 17). The se ice e ms a e u he subca ego ized as se ice desc ip ion
(SDTs),se ice e e ence (SRTs) and se ice p ope y e ms (SPTs). Se ice desc ip ion
e ms define he unc ionali y o he se ice o which he ag eemen is c ea ed. Se ice
e e ence e ms poin o an exis ing se ice endpoin , o which he ag eemen ela es,
whe eas se ice p ope y e ms define “measu able and exposed p ope ies associa ed
wi h a se ice” (And ieux e al. 2007, p. 20). These measu able aspec s o a se ice a e
desc ibed as a se o a iables. Each a iable ela es o an a ibu e o he se ice and
is associa ed wi h a me ic o enable e alua ion o his a iable.
In o de o define assu ances on se ice quali y o he desc ibed se ices, addi ional
gua an ee e ms can be specified ep esen ing he se ice le els bo h pa ies a e ag eeing
on. Each gua an ee e m specifies he obliga ed pa y o he gua an ee which is needed
o enabling consume -side gua an ees. I can also define an op ional quali ying condi ion
which mus be me o he gua an ee o be alid (And ieux e al. 2007). Quali ying
condi ions a e asse ions o e se ice a ibu es and / o ex e nal ac o s such as da e o
39
2. Objec i es and Founda ions
nego ia ions in he IoS). All o hose scena ios a e made up o indi idual nodes eques ing
esou ces o goods om each o he , he e o e ulfilling he necessa y condi ion.
Also a leas one sufficien condi ion can be ound o he IoS scena io, whe e he
p esen se ices a e only a ailable o a gi en imeslo and he e o e exhibi all cha ac-
e is ics o pe ishable goods. Also he complex s uc u e implies e y dynamic demand
and supply, ulfilling he hi d i em lis ed.
A e ha ing shown he necessi y o nego ia ions in he an icipa ed IoS se ing, I will
now de i e a defini ion o he me e e m nego ia ion, as o be used o he emainde o
his hesis.
Since he con ex s in which nego ia ions ake place a e nume ous he e a e many
diffe en defini ions o nego ia ion p ocesses desc ibing he concep om diffe en pe -
spec i es. P ui o example desc ibes nego ia ions as “a o m o decision making in
which wo o mo e pa ies alk wi h one ano he in an effo o esol e hei oppos-
ing in e es s” (P ui 1981, p. xi). “The pa ies fi s e balize con adic o y demands
and hen mo e owa d ag eemen by a p ocess o concession making o sea ch o new
al e na i es” (P ui 1981, p. 1).
In my hesis I will use he ollowing defini ion, as i ocuses much mo e on elec onic
en i onmen s:
Defini ion 2.13 A nego ia ion ep esen s an “i e a i e communica ion and decision
making p ocess[...] be ween wo o mo e agen s (pa ies o ep esen a i es) who: 1)
canno achie e hei objec i es h ough unila e al ac ions; 2) exchange in o ma ion com-
p ising offe s, coun e -offe s and a gumen s; 3) deal wi h in e dependen asks; and 4)
sea ch o a consensus which is a comp omise decision” (Bichle , Ke s en, and S ecke
2003, p. 318).
This defini ion explici ly desc ibes he in o ma ion exchanged in he communica ion
p ocess as offe s and coun e -offe s. Addi ionally he pa ies in ol ed in a nego ia ion
a e u he specified o be he pa ies, affec ed by he consensus o be eached, o al e -
na i ely agen s ac ing on hei behal , which fi s he IoS scena io e y well.
The heo e ical ounda ions o he se ice nego ia ion phase p esen ed in he ollow-
ing a e subdi ided in o he al eady in oduced pe spec i es on nego ia ion p ocesses:
p o ocols,objec s and s a egies / pa icipan s18.
Nego ia ion P o ocols
In gene al, a p o ocol desc ibes he ules o an in e ac ion be ween wo o mo e pa ies.
This in ol es he defini ion o s a es he indi idual pa icipan s can be in, he possible
ac ions o he pa icipan s dependen o he cu en s a e and, op ionally, pa icula
s imuli like ecei ed messages o example. Ano he ace o p o ocols is o define he
messages and message o ma s used in he in e ac ions.
18Acco ding o he esea ch ques ion o be answe ed in his hesis, I will p ima ily ocus on he p o ocol
pe spec i e.
46

2.2. Concep ual Founda ions
Defini ion 2.14 A nego ia ion p o ocol desc ibes he sequence and exchanged con en
o a nego ia ion p ocess be ween he in ol ed pa ies.
To gi e a sho ini ial o e iew on exis ing nego ia ion p o ocols, a axonomy is de i ed
in he ollowing. I is based on o he esea ch wo ks ound in he li e a u e, o igina ing
in so wa e agen , nego ia ion and elec onic ma ke esea ch (see o example Ba olini,
P eis , and Jennings 2005; Bichle and Kalagnanam 2006; Hude e al. 2009; Lomuscio,
Woold idge, and Jennings 2003; S ¨obel and Weinha d 2003; Wu man, Wellman, and
Walsh 1998, 2001).
Essen ially, nego ia ion p o ocols can be e y oughly ca ego ized acco ding o hei
o e all configu a ion, e.g. he numbe and dis ibu ion o nego ia ing agen s. Such a
ca ego iza ion will esul in he ollowing classes o p o ocols:
•In 1:N nego ia ions one selle agen and an a bi a y se o buye agen s ake pa .
The buye s pos bids o he one selle agen which in u n chooses he bes pos ed
offe and engages in an ag eemen wi h he sende o his bid. Such 1:N si ua ions
a e called auc ions.
•N:1 nego ia ions ep esen he class o e e se auc ions. Re e se auc ions consis
o one buye and a se o selle s pos ing bids. The buye hen chooses he bes bid
analogously o egula auc ions, excep ha he bes bid in a e e se auc ion is
he one wi h he lowes s a ed p ice whe eas in egula auc ions he highes pos ed
p ice wins.
•In N:M nego ia ion se ings bo h, selle and buye se s, can consis o an a bi a y
numbe o agen s. Each agen is allowed o pos bids o a cen al ma ke ins ance
which ma ches offe and demand.
•1:1 nego ia ions comp ise wo agen s exchanging bids in o de o each an ag ee-
men . 1:1 nego ia ions a e also called ba gaining scena ios.
When ying o u he subca ego ize hese classes a mo e de ailed se o cha ac e is ics
can be employed:
•Roles: The oles a ibu e specifies which oles a e p esen in a nego ia ion. In
gene al wo oles a e inhe en ly pa icipa ing in a nego ia ion: SP and SC. Addi-
ionally, an op ional media o ole can be p esen , allowing o he implemen a ion
o cen al ma ke ins ances (o en called ma ke b oke s), auc ionee s o us ed
hi d pa ies.
•Agen s: This pa ame e defines he minimum and maximum numbe o agen s
allowed o join a nego ia ion in a gi en ole (1:1, 1:N, N:1, N:M).
•Sides: The sides pa ame e specifies which o he in ol ed oles a e allowed o pos
bids.
47
2. Objec i es and Founda ions
•S a us: The s a us flag indica es whe he he nego ia ion s a us, e.g. he cu en ly
winning bid, can be accessed by he nego ia ing agen s o no . This allows o
defini ion o sealed-bid nego ia ions.
Based on he shown axonomy each nego ia ion p o ocol ype can be defined as a
uple o he ollowing o m: <SP(a, b, c),SC(a, b, c),Med(a, b),S >
Fo each o he oles p esen in a nego ia ion p o ocol (SP, SC and media o ) he
minimum (a) and maximum (b) numbe o agen s allowed is p esen ed along wi h a flag
indica ing whe he his / hese agen (s) a e allowed o pos bids (0 indica ing hey a e
no , 1 indica ing hey a e allowed o). Pos ing bids is only possible o se ice p o ide s
and / o se ice consume s and he e o e he hi d pa ame e (c) is only p esen o
hese wo oles. Finally ano he flag is gi en defining whe he he cu en s a us o he
nego ia ion (S) is accessible o no , e.g. whe he he agen s can access he cu en bid
o no .
English Auc ion
<SP(1,1,0),SC(1,N,1),Med(0,1),1>
In an English Auc ion (EA) he p ice o he good unde nego ia ion s eadily inc eases
un il no bidde is willing o pay he nex p ice s ep. The bidde ha ing bid he las , and
he e o e highes , alid bid will ge he ag eemen and win he auc ion.
An EA always has exac ly one selle (SP) and an a bi a y numbe o buye s (SC).
Depending on he implemen a ion he e can be a mos one media o p esen , i no he
SP also ac s as auc ionee and no media o is needed. In he EA he cu en ly winning
bid is always known o he nego ia o s, so he s a us flag is se o 1.
Du ch Auc ion
<SP(1,1,0),SC(1,N,1),Med(0,1),1>
Du ing a Du ch Auc ion (DA) he p ice o he nego ia ed good is se qui e high a
he beginning and hen lowe ed du ing he auc ion p ocess. Whene e an agen decides
o buy he p oduc o he cu en ly gi en p ice i pos s he bid and engages in he
ag eemen .
Analogous o he EA he e is one SP, possibly an indefini e numbe o SCs and an
op ional media o .
Vick ey / Fi s -p ice-sealed-bid (FPSB) Auc ion
<SP(1,1,0),SC(1,N,1),Med(0,1),0>
In a Vick ey o FPSB Auc ion he nego ia o s only pos one bid each. They do so in a
sealed way, e.g. he o he nego ia o s a e no able o access he con en o he pos ed
bids. A e a ce ain e en occu s he auc ion clea s and he bid offe ing he (second)
highes p ice wins19. Conce ning he in ol ed oles and numbe s o agen s in each ole
he Vick ey and FPSB Auc ions do no diffe om EA / DA p o ocols. Howe e , in a
Vick ey / FPSB Auc ion he s a us is no isible o he agen s.
19In a FPSB Auc ion he bid offe ing he highes p ice and in a Vick ey Auc ion he one offe ing he
second highes bid wins.
48
2.2. Concep ual Founda ions
Re e se English Auc ion
<SP(1,N,1),SC(1,1,0),Med(0,1),1>
ARe e se English Auc ion (REA) is basically he same p o ocol as a egula EA excep
ha he e is only one buye and many selle s. This way he p ice o he nego ia ed
good is s eadily dec easing om a s a ing p ice. In he end he selle willing o accep
he ag eemen o he lowes p ice wins he auc ion.
Re e se Du ch Auc ion
<SP(1,N,1),SC(1,1,0),Med(0,1),1>
In a Re e se Du ch Auc ion (RDA) he e is only one buye and possibly many selle s as
in a REA. Now he p oduc p ice is simply inc eased by a ce ain s ep in each ound.
The fi s selle accep ing he cu en p ice s ep pos s a bid and wins he nego ia ion by
doing so.
Re e se Vick ey / FPSB Auc ion
<SP(1,N,1),SC(1,1,0),Med(0,1),0>
Du ing a Re e se Vick ey / FPSB Auc ion p o ocol each o he N selle s pos s one sealed
offe . A e a ce ain e en occu s, he bes offe , e.g. he one s a ing he (second) lowes
p ice will esul in an ag eemen . This diffe s om a egula Vick ey / FPSB Auc ion
in ha ing only one buye and possibly many selle s, espec i ely.
Call Ma ke
<SP(1,N,1),SC(1,M,1),Med(1,1),0>
ACall Ma ke (CM) is cha ac e ized as a scena io, consis ing o a se o selle s and
buye s pos ing bids o sell o buy o a cen al ma ke b oke (media o ). This b oke
hen ma ches offe and demand messages acco ding o some defined sol e algo i hm.
This ma ching egula ly akes place a e ce ain in e als o bidding. Each offe sen o
he ma ke ins ance is sen in a sealed manne , so ha he o he agen s on he ma ke
canno adap o pos ed bids hemsel es.
Con inuous Double Auc ion (CDA)
<SP(1,N,1),SC(1,M,1),Med(1,1),1>
A CDA exhibi s he same cha ac e is ics as a Call Ma ke excep ha he pos ed bids
a e submi ed isible o all ma ke pa icipan s unlike wi h a Call Ma ke .
Ba gaining
<SP(1,1,1),SC(1,1,1),Med(0,0),1>
In one-on-one ba gaining nego ia ions wo agen s nego ia e by exchanging offe s and
coun e offe s. The agen s ake u ns in pos ing he bids o one ano he un il one o he
agen s is offe ed an ag eemen i can accep . Then he espec i e agen jus sends an
accep message back and he ag eemen is in place. I no ag eemen can be eached, o
example i he accep in e als o he p ice o bo h agen s do no o e lap, he nego ia ion
49
2. Objec i es and Founda ions
Table 2.1.: Nego ia ion P o ocol Types
P o ocol A ibu e Values
English Auc ion. <SP(1,1,0),SC(1,N,1),Med(0,1),1>
Du ch Auc ion <SP(1,1,0),SC(1,N,1),Med(0,1),1>
Vick ey/FPSB Auc ion <SP(1,1,0),SC(1,N,1),Med(0,1),0>
Re e se English Auc ion. <SP(1,N,1),SC(1,1,0),Med(0,1),1>
Re e se Du ch Auc ion <SP(1,N,1),SC(1,1,0),Med(0,1),1>
Re e se Vick ey/FPSB <SP(1,N,1),SC(1,1,0),Med(0,1),0>
Call Ma ke <SP(1,N,1),SC(1,M,1),Med(1,1),0>
CDA <SP(1,N,1),SC(1,M,1),Med(1,1),1>
One-On-One Ba g. <SP(1,1,1),SC(1,1,1),Med(0,0),1>
is canceled a e a ce ain condi ion occu s (an example would be a ce ain numbe o
offe s sen ).
As i is he mos commonly used app oach o SLA defini ion and nego ia ion I wan
o gi e a sho ema k on he p o ocol p oposed in he WS-Ag eemen (And ieux e al.
2007) and WS-Ag eemen Nego ia ion (Waeld ich e al. 2010) specifica ions espec i ely.
The o me offe s only a e y simple p o ocol in which an ini ia o can que y he a ailable
SLA empla es om a esponde , chose he one fi ing bes and c ea e an ag eemen
offe om i . This is subsequen ly p oposed o he esponde node, which in u n can
accep o ejec i . No coun e offe possibili ies a e p esen in his p o ocol. WS-
Ag eemen Nego ia ion ex ends his model by in oducing a new Nego ia ion Laye ,
abo e he Se ice and Ag eemen Laye s. This laye comp ises Nego ia ion Fac o y
and ac ual nego ia ion componen s, allowing he in ol ed pa ies o c ea e a nego ia ion
(implemen ed as a se ice ins ance om he ac o y) and pos offe s o each o he . Gi en
bo h pa ies offe he men ioned nego ia ion componen s an ac ual ba gaining p o ocol,
consis ing o offe s and coun e -offe s is possible.
Table 2.1 shows he o mal desc ip ion o he p o ocol ypes jus iden ified.
Fo all o hese ypes a a ie y o mo e o less complex sub- ypes ha e been de el-
oped, each a ying in e ms o he sold p oduc , allowed messages o message sequences,
ma ching algo i hms e c. (see o example Bichle 2001; E molaye and Kebe le 2006; Li,
Giampapa, and Syca a 2003; Smi h 1980; S ¨obel and Weinha d 2003; Walsh, Wellman,
50
2.2. Concep ual Founda ions
and Ygge 2000; Wu man, Walsh, and Wellman 1998a)20.
Nego ia ion (P o ocol) Desc ip ion Languages In he pas yea s qui e a lo o wo k
has been done in defining languages and axonomies o he o mal desc ip ion o nego-
ia ion p o ocols. Since my nego ia ion amewo k also builds on a s uc u ed p o ocol
desc ip ion in o de o achie e p o ocol-gene i y (see chap e 3), I now gi e a sho
o e iew on such languages.
On ology-based languages mos ly desc ibe no only he nego ia ion p o ocol i sel , bu
also model he applied s a egies (E molaye and Kebe le 2006) and o he po en ially
p i a e aspec s o he nego ia o s. In doing so, a se o commonly known on ologies
along wi h seman ic easoning mechanisms is applied.
Languages based on pa ame e s aim a desc ibing a nego ia ion p o ocol by a se o
p ocess a ibu es and espec i e alues. Due o his e y simple app oach i is no
su p ising ha many o he languages ound ollow ha pa adigm.
Lomuscio e al. p oposed a “Classifica ion Scheme o Nego ia ion in Elec onic Com-
me ce” (Lomuscio, Woold idge, and Jennings 2003). This app oach, howe e , ocuses on
p e-nego ia ion phases and assumes high human in e ac ion a es, con adic ing wi h
he IoS scena io.
Wu man e al. p esen ed a se o auc ion pa ame e s while de eloping an In e ne -
based “pla o m o p ice-based nego ia ion - he Michigan In e ne Auc ionBo ” (Wu -
man, Wellman, and Walsh 1998, p. 1) . This sys em was designed o se e as an auc ion
se e o humans as well as o so wa e agen s. Un o una ely, he ocus is on one sin-
gle nego ia ed issue: he p ice. Thus, only auc ion p o ocols a e suppo ed. The same
au ho s ex ended his axonomy o also co e mul idimensional auc ions in a ollow-up,
much mo e comp ehensi e pape (Wu man, Wellman, and Walsh 2001).
The mos comp ehensi e pa ame e -based desc ip ion language so a was published
by S ¨obel and Weinha d (2003). The classifica ion scheme p esen ed he e co e s a e y
comp ehensi e se o nego ia ion p o ocols, bo h om he human-cen ic and au oma ed
pe spec i e, wi hou s essing echnology- ela ed issues o ocusing jus on a subse o
possible nego ia ion p o ocols.
In gene al, pa ame e -based app oaches a e o en deemed insufficien o desc ibing
empo al pa e ns o nego ia ions, such as he sequence o possible e en s o ac ions
(Lochne and Wellman 2004).
Rule-based languages on he o he hand, y o ackle his p oblem by speci ying ac ion
ules ha p esc ibe wha a nego ia ing agen has o do unde ce ain condi ions. By also
inco po a ing use -defined a iables, such as coun e s e c., ule-based sys ems can offe
a high flexibili y o he auc ion designe (Lochne and Wellman 2004). A good example
o such an app oach is (Wu man, Wellman, and Walsh 2002).
Finally, s a e au oma a can be used o define he beha io o a gi en se ice o a
whole sys em. This concep allows o he defini ion o in e nal s a es and possible s a e
20A subse o hese p o ocols has also been published by he Founda ion o In elligen Physical Agen s
(FIPA) , as pa o hei p o ocol lib a y o agen -based communica ions, which can be ound unde
h p://www.fipa.o g/.
51

2. Objec i es and Founda ions
changes along wi h op ional condi ions o such changes as well as igge ed e en s when
en e ing o lea ing a espec i e s a e (Ma in 2002).
In his hesis, a desc ip ion language building on pa ame e s, op ional ule-elemen s
as well as s a e machines is employed. Fo mo e de ails see subsec ion 3.2.2.
Nego ia ion Objec s
Defini ion 2.15 The good unde nego ia ion (in he IoS ep esen ed by he offe ed elec-
onic se ices and hus SLAs) is commonly e e ed o as he nego ia ion objec .
Depending on he numbe o nego iable cha ac e is ics, nego ia ion objec s can be
dis inguished on a fi s le el in o single-andmul i-a ibu i e ones (a dis inc ion also
some imes used o he nego ia ion p ocesses conce ning such p oduc s) (Lai e al. 2004).
Single-a ibu i e nego ia ion objec s expose only one nego iable a ibu e, he p ice.
E en nego ia ion objec s ha a e mo e complex, bu only offe one nego iable a ibu e
a e conside ed single-a ibu i e (wi h espec o he nego ia ion hey a e in ol ed in).
Wi hin he IoS se ing such mechanisms could be applied in si ua ions whe e commodi y
se ices in a s anda d configu a ion a e aded, only offe ing he p ice a ibu e o be
nego iable.
Acco dingly, mul i-a ibu i e nego ia ion objec s expose mul iple diffe en nego iable
a ibu es. Apa om he p ice, such a ibu es could comp ise he da e o deli e y o
maximum esponse ime e c. Fo he IoS scena io mul i-a ibu i e nego ia ion p ocesses
deem app op ia e, as a SLA egula ly comp ises se e al SLOs o which mos ly mo e han
one a e nego iable.
In a second dimension, nego ia ion objec s can be dis inguished acco ding o whe he
hey ep esen a single good (ins ance o a se ice in he IoS se ing) ha is unde ne-
go ia ion o a whole bundle o such goods. T adi ional nego ia ion axonomies assume
a single nego ia ion objec (which is single- o mul i-a ibu i e espec i ely). Recen
de elopmen s in nego ia ion esea ch iden ified wo dis inc classes o nego ia ions con-
ce ning bundled nego ia ion objec s: combina o ial and mul i-uni nego ia ions.
In combina o ial nego ia ions bidde s “place bids on combina ions o i ems, called
“packages” a he han jus indi idual i ems” (C am on, Shoham, and S einbe g 2006,
p. 1). Each nego ia ion objec he e o e ep esen s an offe desc ibing he desi ed bundle
o (po en ially qui e diffe en ) i ems and he p ice one offe s o pay o i . This is
especially use ul when complemen a ies a e p esen among he i ems o be sold21.
In mul i-uni nego ia ions on he o he hand he nego ia ion objec comp ises se e al
uni s o he same p oduc o be sold wi hin as a bundle. They diffe om combina o ial
nego ia ions in ha he indi idual uni s do no diffe significan ly and a e ea ed in e -
changeably h oughou he nego ia ion. The esul ing ag eemen hus conce ns a se o
iden ical goods ha a e sold.
As each se ice in he IoS is e e enced indi idually i he e o e defines an indi idual
i em, ea ed in an own nego ia ion p ocess. I a combina ion o indi idual se ices o
se ice uni s a e o be sold as a bundle his will egula ly esul in hem being combined
21I ems a e complemen a y i he u ili y o a se o i ems succeeds he sum o he indi idual u ili ies.
52
2.2. Concep ual Founda ions
En i onmen
So wa e Agen
e ec o s
senso s
Agen
Reasoning
Figu e 2.7.: Abs ac A chi ec u e o a So wa e Agen (Russell and No ig 1995, p. 45)
o a highe -o de , complex se ice. Combina o ial and mul i-uni nego ia ions a e hus
no app op ia e in he an icipa ed se ing and will no be in he p ima y ocus o my
wo k.
Nego ia ion S a egies / Pa icipan s
When desc ibing he ac o s wi hin a nego ia ion p ocess, h ee ypes o nego ia ions can
be dis inguished wi h ega d o hei le el o au oma ion: unsuppo ed,suppo ed and
au oma ed (elec onic) nego ia ions (Bichle , Ke s en, and S ecke 2003).
In an unsuppo ed nego ia ion p ocess humans nego ia e abou a gi en good wi hou
any help o an IT sys em. Du ing a suppo ed nego ia ion he in ol ed people delega e
some o he nego ia ion- ela ed ask o so wa e componen s (e.g. c ea ion o a coun e -
offe o weighing o diffe en offe al e na i es). Finally, a ( ully) au oma ed nego ia ion
depic s he p ocess “in ol [ing] so wa e agen s ha make decisions and con ol he en i e
p ocess, including he specifica ion o offe s and concessions, and he final decision abou
ag eemen o disag eemen ” (Bichle , Ke s en, and S ecke 2003, p. 322).
The IoS scena io is cha ac e ized by a huge amoun o economic ansac ions as well
as e y high in e ac ion a es. Mos o he pu chased se ice in oca ions happen wi hin
he con ex o a business wo kflow, mos o which a e ime-c i ical and he e o e show a
endency o high au oma ion. Wo kflow engines will need o eques , buy and in eg a e
ex e nal se ices much mo e apidly han would e e be possible wi h human nego ia o s.
E en accompanied wi h nego ia ion suppo sys ems, inc easing he efficiency o nego-
ia ion p ocesses d ama ically, such an app oach would s ill no be app op ia e wi hin a
se ing whe e possibly housands o nego ia ions happen in pa allel a a gi en ime and
up o a couple o hund eds would ha e o be conduc ed by one pa icula use . Thus, I
53
2. Objec i es and Founda ions
deem he applica ion o so wa e agen s o nego ia ions in he IoS manda o y and ocus
on hese o he emainde o his hesis.
One o he mos well-known defini ions o a so wa e agen was gi en in (Jennings
2000, p. 280):
Defini ion 2.16 “An agen is an encapsula ed compu e sys em ha is si ua ed in some
en i onmen and ha is capable o flexible, au onomous ac ion in ha en i onmen in
o de o mee i s design objec i es” (Jennings 2000).
The defining a chi ec u al cha ac e is ics o an agen a e hus ha i is si ua ed in
an en i onmen , some aspec s o which i can obse e wi h some senso mechanism and
wi hin which i can unde ake some ac ions, using some effec o s a i s disposal (Russell
and No ig 1995, p. 31). Figu e 2.7 illus a es his abs ac a chi ec u e o a so wa e
agen .
In con as o simple eac i e de ices, such as passi e so wa e demons, agen s a e
acc edi ed o ha e he ollowing defining cha ac e is ics (see o example (Woold idge
1997):
•Reac i i y
•P oac i eness
•Au onomy
•Social-Abili y
Reac i i y The claim o eac i i y s a es ha a so wa e agen mus be capable o
obse ing i s en i onmen (consis ing o human use s, o he simple so wa e componen s
o e en sophis ica ed pee agen s) and “ espond in a imely ashion o changes ha
occu in i in o de o sa is y [i s] goals” (Woold idge 2005, p. 23). In o de o do so,
an agen mus no only be able o obse e and subsequen ly ac wi hin i s en i onmen ,
bu also mus be equipped wi h an in e nal p ocessing mechanism choosing he ac ion
o be unde aken a e a gi en s imulus occu ed.
P oac i eness So wa e agen s a e gene ally supposed o “exhibi goal-di ec ed beha -
io ” (Woold idge 2005, p. 23). They a e assumed o p oac i ely pu sue he goals ecei ed
om a human use o o he so wa e de ice. In con as o pu ely eac i e sys ems a
so wa e agen he e o e has o an icipa e esul s o i s own ac ions and assess i s op ions
in he ligh o i s goals. I will hus choose ac ions ha will b ing i close o his goal
o e o he s. Again such a beha io demands some cogni i e capabili ies (commonly e-
e ed o as s a egy) and expe ience wi h ac ions chosen in he pas and esul s ecei ed
hen.
54
2.3. Rela ed Wo k
Au onomy Au onomy e e s o he way an agen eac s o a s imulus ecei ed h ough
one o i s senso s. When being capable o au onomous ac ions an agen is assumed o no
simply eac de e minis ically o ex e nal s imuli, bu a he o choose i s ac ions wi h
espec o he expe iences i has made o e ime (Russell and No ig 1995). This implies
ha agen s ha e a means o emembe ing pas decisions, and espec i ely aken ac ions,
and he so induced en i onmen al changes and a e capable o inco po a e hese “lea ned
cohe encies” in o hei cu en beha io . Au onomy hus adhe es o adap i i y o some
poin , as agen s a e supposed o “be able o ope a e success ully in a wide a ie y o en-
i onmen s, gi en sufficien ime o adap ” (Russell and No ig 1995, p. 35). Commonly,
such an adap i e beha io is achie ed using au oma ed lea ning algo i hms building on
o example gene ic algo i hms (Goldbe g 1989) o neu onal ne wo ks (Haykin 1994).
Social-Abili y The las aspec pays ibu e o he me e ision unde lying he so wa e
agen pa adigm: decen aliza ion. So wa e agen s a e indi idual nodes, in e connec ed
wi h each o he and he en i onmen hey eside in. Each is ying o each i s indi idual
goals, which canno be achie ed unila e ally. Consequen ly he agen s in e ac in o de
o join ly each goals hey would no be able o achie e alone.
Wi hou s essing ac ual implemen a ion echniques o so wa e agen s, hese ou
cha ac e is ics al eady show he po en ial so wa e agen s offe o decen alized se -
ings such as he IoS. Thei abili y o decen ally coo dina e, coope a e and nego ia e
(Woold idge 2005, p. 3) fi s excellen wi h he demanding p oblem o c oss-o ganiza ional
se ice managemen . Ini ial wo ks ha e al eady been done o po so wa e agen ech-
nologies o his p oblem domain especially when conce ned wi h he nego ia ion o se ice
ag eemen s (see o example Buyya, Ab amson, and Venugopal 2005; Eymann, S ei -
be ge , and Hude 2007; G adwell and Padge 2005).
This concludes he concep ual ounda ions o my wo k om bo h a SLA o se ice
cen ic iew and a p ocessual iew ega ding he disco e y and nego ia ion mechanisms,
egula ly employed in dis ibu ed IS. Building on hese undamen al mechanisms, a se o
esea ch p ojec al eady aims a sol ing he p oblem o dis ibu ed SLA managemen and
pa icula ly he elec onic disco e y and nego ia ion o such documen s. The solu ions
de i ed wi hin hese p ojec s he e o e ep esen al e na i e solu ions o a e y simila
(o e en he same) esea ch p oblem, o which my app oach has o be compa ed in he
end.
2.3. Rela ed Wo k
In his sec ion, ela ed esea ch wo k conce ning SLA nego ia ions in dis ibu ed IS is
p esen ed. Each o he iden ified in as uc u e p oposals is desc ibed in e ms o i s
o e all aims, abs ac a chi ec u e and, specifically, i s SLA disco e y and nego ia ion
mechanisms. Finally, e e y app oach is assessed on he basis o he equi emen s iden-
ified in sec ion 2.1; a summa y o his assessmen can be ound in able 2.2.
Fo cla i y pu poses, I ocus on he ones mos closely ela ed o my wo k (e.g. in
e ms o simila i y o he an icipa ed scena io o my IoS model) and mos p ominen ly
55
2. Objec i es and Founda ions
No e: The T us COM p ojec ( unning om Feb ua y 2004 h ough Janua y 2007)
esul ed in a “ amewo k o us , secu i y, and con ac managemen o secu e, col-
labo a i e business p ocessing and esou ce sha ing in dynamically-e ol ing i ual o -
ganiza ions” (Wilson 2007, p. 4). Conce ning he equi emen s used in his hesis his
p ojec is basically equi alen o Nex GRID, so i is again no conside ed in all de ail
he e.
2.3.3. P ojec s Focusing on Economic Aspec s
CATNETS
As opposed o he a o emen ioned p ojec s, CATNETS (“Ca allaxy pa adigm o decen-
alized ope a ion o dynamic applica ion ne wo ks“)32 did no ocus on he managemen
o SLA documen s pe se, bu a he on he applica ion o economic “ma ke -based e-
sou ce alloca ion mechanisms in applica ion laye ne wo ks such as g ids” (Vei e al.
2007, p. 3). I was again unded unde he EC’s six h amewo k p og amme unde he
con ac numbe 003769 and las ed om Sep embe 2004 un il Augus 2007.
The main goal o CATNETS can be summa ized as he assessmen o ma ke -based
esou ce alloca ion mechanisms in massi ely dis ibu ed compu ing in as uc u es. Fo
his in es iga ion a wo-laye ed ma ke se ing was assumed, consis ing o a highe -
le el Se ice Ma ke and lowe -le el Resou ce Ma ke . The idea behind his ision was
ha Basic Se ices o espec i e Agen s a e cons uc ed on he Resou ce Ma ke by
in eg a ing a dis inc se o aw elec onic Resou ces, such as s o age o compu a ion.
These Basic Se ices a e subsequen ly combined on he Se ice Ma ke o mo e complex
se ices o wo kflows (Schnizle e al. 2005b). Wi hin his scena io se ing, wo compe ing
app oaches o economic esou ce managemen we e compa ed: a cen al, auc ion-based
and a decen alized Ca allaxy-based (A daiz e al. 2006) mechanism.
Du ing he p ojec bo h a simula ion en i onmen o assessing hese wo app oaches
as well as a p oo -o -concep p o o ype (A daiz e al. 2007) (howe e only inco po a ing
he decen alized mechanism) was de eloped (Schnizle e al. 2005b). Addi ionally, a lo
o heo e ical esea ch in he a ea o esou ce alloca ion mechanisms, ma ke design and
e alua ion o such concep s was done.
Al hough he agen s wi hin he CATNETS sys em nego ia e in a simplified p op i-
e a y nego ia ion language he p ojec conso ium could show how his language can
be mapped o SLA languages such as WS-Ag eemen (Schnizle e al. 2005b). Since he
p ima y ocus was on he economic mechanisms his decision is e y comp ehensible. On
he o he hand he CATNETS conso ium e y ho oughly in es iga ed how an economic
middlewa e as he CATNETS p o o ype could be in eg a ed wi h exis ing middlewa e
sys ems, subsequen ly ac ing as a sophis ica ed esou ce alloca ion ool. This effo
p ima ily buil on WS-Ag eemen , hus eflec ing he impo ance o SLA-based se ice
p o isioning in CATNETS (Joi a e al. 2007).
Rega ding he iden ified equi emen s i can be s a ed ha his p ojec ac ually in e-
g a es so wa e agen s o disco e y and nego ia ion o he aded se ices. This is he
32h p://www.ca ne s.uni-bay eu h.de/
62

2.3. Rela ed Wo k
case on bo h ma ke s. Rega ding he applied p o ocol one mus dis inguish be ween he
cen alized and he decen alized mechanism. The o me applied a CDA on he Se ice
Ma ke (single-a ibu i e goods a e aded he e) and a mul i-a ibu i e combina o ial
exchange on he Resou ce Ma ke (Schnizle e al. 2005a). The la e always employs he
Al e na e Offe s P o ocol (AO), a one-on-one nego ia ion app oach (Vei e al. 2007).
Simila ly, he disco e y p ocess also diffe s in he wo ma ke ypes: In he cen alized
app oach no ac ual disco e y is men ioned. I assume he agen s a e pa ame e ized wi h
he loca ion o he b oke , as his se ing was only ealized in a simula ion en i onmen .
The decen alized se ing on he o he hand employs a b oadcas disco e y p o ocol
(Vei e al. 2007).
No e: In e ms o he iden ified equi emen s he in as uc u e p oposal desc ibed in
(Ouelhadj e al. 2005) is qui e simila o he CATNETS sys em; he only diffe ence is,
ha he au ho s do no men ion an ac ually implemen ed p o o ype o hei sys em.
SORMA
SORMA (“Sel -O ganizing ICT Resou ce Managemen ”)33, being a ollow-up p ojec o
CATNETS, u he de eloped he idea o ma ke -based esou ce alloca ion in dis ibu ed
in as uc u es. I was unded unde he six h amewo k p og am, con ac numbe
034286, and las ed om Augus 2006 un il July 2009.
Acco ding o one o he fi s p ojec deli e ables, SORMA aimed a “ he de elopmen
o me hods and ools o es ablishing an efficien ma ke -based alloca ion o esou ces in
a mo e efficien way in o de o enable esou ce accessibili y o all use s and o inc ease
use ’s sa is ac ion, p ofi and p oduc i i y” (Ma os e al. 2008, p. 8). Fo his, he p ojec
conso ium buil on so wa e agen s o enable a ully au oma ed and economically sound
se ice ading p ocess (Ma os e al. 2008).
The esul ing a chi ec u e consis s o six diffe en laye s (SORMA 2007):
•G id Applica ion Laye : applica ions o be un on he esou ces, pu chased o e
he SORMA ma ke .
•In elligen Tool Laye : smalle so wa e componen s suppo ing he SCs and SPs
in he ading p ocess ( o example he Bid Gene a o ).
•Open G id Ma ke Laye : mainly economic se ices, esponsible o offe ma ching
(T ading Managemen ) and SLA handling (Con ac Managemen ,SLA En o ce-
men and Billing)e c.
•Co e Ma ke Se ices: ex ension se ices, needed o allow ma ke -based esou ce
alloca ion in adi ional G id middlewa es ( o example ansac ion logging, ma ke
di ec o y e c.).
•Economically Enhanced Vi ualiza ion Middlewa e: esou ce-cen ic in e ace o
s anda dized G id implemen a ions, allowing o esou ce co-alloca ion o SLA
managemen on he basis o an indi idual esou ce.
33h p://www.so ma-p ojec .o g/
63
2. Objec i es and Founda ions
•G id Resou ce Laye : se o se ices o be aded o e he SORMA ma ke .
This logical a chi ec u e was subsequen ly implemen ed in a p oo -o -concep p o o-
ype employing a CDA-based ading sys em (SORMA 2009).
Gi en i s economic se ing, SORMA explici ly builds on SLAs as a undamen al con-
cep , bo h defining he ela ionship be ween SC and SP as a esul o he ma ching
p ocess, bu also as an inpu o he SLA en o cemen and billing sys em (SORMA
2008). This componen subsequen ly “pe o m[s] a con inuous un ime SLA iola ion
p edic ion, de ec ion, and eac ion ha can help p e en SLA iola ion om occu ing”
(SORMA 2008, p. 58).
Following he ideas al eady p oposed du ing CATNETS, SORMA aims a an eco-
nomically sound esou ce-alloca ion based on ma ke s. Al hough he need o diffe en
nego ia ion p o ocols is explici ly s a ed in (SORMA 2007, p. 9), only a single p o ocol
(a CDA) was ac ually implemen ed in he final p o o ype sys em (SORMA 2009, p. 57).
The men ioned possibili y o in eg a ing a ious nego ia ion p o ocols ia he Ma ke
Exchange Se ice was appa en ly no o no sufficien ly implemen ed wi hin he p ojec
(SORMA 2009, p. 50).
On he o he hand, SORMA builds on a dis ibu ed se o ma ke egis ies, assis -
ing he SC and SP agen s du ing he disco e y and publica ion p ocesses espec i ely
(SORMA 2008). O igina ing in he economic usage scena io unde lying SORMA, an
ex ended SLA desc ip ion language was c ea ed allowing o he defini ion o echnical
and also economic aspec s o a se ice in oca ion (Bo isso e al. 2009). Finally, so -
wa e agen s a e hea ily used du ing bo h he disco e y and he nego ia ion o SLAs in
SORMA (Bo isso , Neumann, and Weinha d 2009).
2.3.4. Ini ial App oaches Towa ds P o ocol-Gene i y in SLA Nego ia ions
NegoFAST
Resinas e al. (Resinas, Fe nandez, and Co chuelo 2010) e y ecen ly defined a com-
p ehensi e amewo k o he design and implemen a ion o gene ic SLA nego ia ion
componen s, he NegoFAST amewo k. Fo his he au ho s conduc ed a comp ehen-
si e su ey on elec onic nego ia ion sys ems and subsequen ly deduc ed a se o equi e-
men s, simila o hose p esen ed in his hesis. Based on hese equi emen s, a e e ence
a chi ec u e o gene ic se ice nego ia o s was designed, ollowed by a p oo -o -concep
implemen a ion.
The au ho s claim ha hei main goal is “ o unde s and he equi emen s o au o-
ma ed nego ia ion sys ems o se ice ag eemen s in open en i onmen s and o p o ide
he ounda ions o de eloping such sys ems” (Resinas, Fe nandez, and Co chuelo 2010,
p. 2).
The esul ing e e ence a chi ec u e comp ises ou diffe en modules: P o ocol Man-
agemen (“selec ion and execu ion o nego ia ion p o ocols” (Resinas, Fe nandez, and
Co chuelo 2010, p. 13)), Decision Making (“de e min[ing] he beha io o he [...] sys em]
(Resinas, Fe nandez, and Co chuelo 2010, p. 14)), Wo ld Modeling (used o “ga he [ing],
analys[ing] and manag[ing] use ul in o ma ion o make decisions du ing a nego ia ion”
64
2.3. Rela ed Wo k
(Resinas, Fe nandez, and Co chuelo 2010, p. 13)) and Coo dina ion (o ches a ing he
o he modules du ing he nego ia ion p ocess).
These modules subsequen ly ha e o be ins an ia ed when implemen ing a gi en ne-
go ia ion sys em o a pa icula se ing. By igo ously building on common in e aces
and a gene ic da a model34 he NegoFAST a chi ec u e allows o implemen ing a i-
ous decision making s a egies and nego ia ion p o ocols, om which sys em designe s
can chose du ing his ins an ia ion p ocess. The au ho s e en claim hei sys em allows
changing he applied nego ia ion p o ocol o decision making componen a un ime
(Resinas, Fe nandez, and Co chuelo 2010).
I is clea ly obse able ha he NegoFAST amewo k pa icula ly ocuses on he
nego ia ion phase o he se ice li e cycle. No asse ions on se ice disco e y and he
applica ion o so wa e agen s he ein a e made. On he o he hand a a ie y o di -
e en SLA ep esen a ion schemes is suppo ed, basically all SLA languages ha can
be mapped o he defined gene ic da a model. Simila ly, all nego ia ion p o ocols ha
can be buil using he defined nego ia ion p imi i es a e eligible in he amewo k. The
au ho s men ion in (Resinas, Fe nandez, and Co chuelo 2010) ha he nego ia ion p o-
ocol ac ually applied is again nego ia ed be o ehand, howe e no asse ion on how his
is done (o mos impo an ly how he espec i e p o ocol is desc ibed / communica ed
o he o he pa y) could be ound.
Aneka and G idBus
Ano he wo k, e y simila o my app oach was published by B andic e al. (2008a). I
was done in he con ex o he Aneka35 and G idBus36 p ojec s o he Cloud Compu ing
and Dis ibu ed Sys ems Labo a o y, Uni e si y o Melbou ne.
The au ho s s a e he p oblem ha mos SLA nego ia ion in as uc u es demand a
p e-defined configu a ion o he pa icipan s in e ms o applied p o ocol o nego ia ed
SLA e ms and p opose a solu ion based on a comp ehensi e me a-nego ia ion a chi ec-
u e.
This me a-nego ia ion sys em builds on a se o me a-nego ia ion documen s, “whe e
pa icipa ing pa ies may exp ess: he p e- equisi es o be sa isfied o a nego ia ion [...],
he nego ia ion p o ocols and documen languages o he specifica ion o SLAs ha hey
suppo and condi ions o he es ablishmen o an ag eemen ” (Ke esz, Kecskeme i, and
B andic 2009, p. 29).
B andic e al. no only defined he o e all p ocess o me a-nego ia ion bu also
p esen ed a XML-based language desc ip ion o be used when cons uc ing espec i e
documen s, as well as a ole-based a chi ec u e o hei sys em. I comp ises a se o
egis ies, whe e he documen s a e published as well as some middlewa e componen s
on bo h SC and SP sides used o pa se hese documen s and load he espec i e s a egy
34Wi hin his da a model a se o nego ia ion p imi i es ( o example accep , ejec Nego ia ion, ejec P o-
posal,commi ,p opose o in o m) is used as a common basis o all suppo ed nego ia ion p o ocols
(Resinas, Fe nandez, and Co chuelo 2006).
35h p://www.manj aso .com/p oduc s.h ml
36h p://www.g idbus.o g/
65
2. Objec i es and Founda ions
modules necessa y o he nego ia ion. A p oo -o -concep p o o ype based on he Aneka
GC sys em and he G idBus esou ce b oke was implemen ed as an ex ension o he
in as uc u e p esen ed in (Venugopal, Chu, and Buyya 2008).
The Aneka me a-nego ia ion sys em builds on a egis y-based disco e y mechanism
and allows a dis inc se o diffe en nego ia ion p o ocols o be used subsequen ly. How-
e e , hese p o ocols a e simply e e enced by hei name, so a common unde s anding o
he espec i e p o ocol is necessa y o success ul execu ion. In my hesis, I y o use a
mo e gene ic app oach, in which only some defined p o ocol p imi i es ha e o be known
(simila o he NegoFAST model) and mo e complex p o ocols can be cons uc ed om
hese. B andic e al. also do no men ion he usage o so wa e agen echnology o he
achie ed deg ee o au oma ion wi hin hei wo k. Also, no ma ke in e media ies a e
in oduced, as needed o example in a b oke ed ma ke s uc u e.
SLA@SOI
The mos ecen p ojec dealing wi h elec onic SLA managemen in a business con ex is
SLA@SOI 37(“Empowe ing he se ice indus y wi h SLA-awa e in as uc u es”). I is
unded unde he EC’s se en h amewo k p og amme unde he con ac numbe FP7-
216556, s a ed on 31s o May 2009 and will un h ough 38 mon hs. The conso ium
consis s o ele en pa ne s om bo h indus y and academia, he o me co e ing many
significan fields in ol ed in he IoS ision: elecommunica ions, so wa e de elopmen ,
business se ices e c.
Acco ding o one o hei ini ial p ojec deli e ables, he main goal o SLA@SOI is o
design and implemen “a business- eady se ice-o ien ed in as uc u e empowe ing he
se ice economy in a flexible and dependable way” (Theilmann e al. 2009, p. 7). In ha ,
hey de elop a amewo k a chi ec u e, suppo ing he managemen o bo h SLAs and
he unde lying se ice in as uc u e on ei he he business,so wa e and in as uc u e
le el (Theilmann e al. 2010). I comp ises so wa e componen s suppo ing he o e all
se ice li e cycle (see (Theilmann e al. 2010, p. 19) o de ails), such as SLA Manage s
esponsible o managing he a ailable SLA empla es and nego ia ion p ocesses building
he eupon, Se ice E alua o s handling p edic ion asks o Se ice Manage s o un
ime managemen o he se ice ins ances.
Acco ding o (Ko sokalis e al. 2010) and (Lambea e al. 2010), a ule-based p o ocol
engine is used wi hin each SLA Manage node, allowing o p o ocol-gene ic nego ia ions.
Fo his, i is configu ed wi h a espec i e desc ip ion file, o which all in ol ed nego-
ia o s ag ee and subsequen ly en o ces he p o ocol ules by in e cep ing all message
exchanges (one SLA Manage is loca ed a bo h he SP and he SC ealm).
The undamen al concep s, applied in SLA@SOI, a e e y simila o my app oach: a
gene ic nego ia ion componen is employed, and is configu ed wi h a p o ocol desc ip ion
documen as pe nego ia ion i engages in, a egis y-based disco e y phase is assumed
and he nego ia ion p o ocols hemsel es a e desc ibed, a leas pa ially, on he basis
o s a e machines. This ac unde lines no only he mo i a ion o my wo k, bu also
he gene al design idea applied he ein (see 3.1 o mo e de ails).
37h p://sla-a -soi.eu/
66
2.3. Rela ed Wo k
Howe e , mos o he concep s p esen ed in he cu en ly a ailable p ojec deli e ables
a e in a e y p elimina y s age and some claims a e e en s ill con adic o y a his poin
in ime. The conso ium o example s a es, ha he cu en p o o ype suppo s all
“single nego ia ions (one- o-one, one- o-many and many- o-one)” (Lambea e al. 2009,
p. 51) bu s a e in he e y same documen ha “a he momen , he p o ocol [ hey]
decided o use in SLA@SOI is a one- o-one p o ocol (mul i-a ibu e bila e al p o ocol)”
(Lambea e al. 2009, p. 53). A he da e o publica ion o his hesis, he ac ual deg ee
o which amoun he SLA@SOI sys em ulfills he s a ed equi emen s could hus no be
de e mined. The e o e i is no in eg a ed in o able 2.2 and only men ioned he e sho ly.
When wo king on dis ibu ed SLA managemen in he u u e, he esul s achie ed he ein
can none heless be o eseen o be highly ele an and a e ecommended o u he
in es iga ion.
67

2. Objec i es and Founda ions
Table 2.2.: Rela ed Wo k
Rela ed
P ojec s
AssessG id /
HPC4U Akog imo BEinGRID
BREIN
/ASG-
ASAPM
Nex GRID /
T us Com CATNETS SORMA NegoFAST Aneka /
G idBus
Requi emen s
Regis y-
based Dis-
co e y Phase
[R1]
?, no asse -
ion Yes No Yes Yes No Yes ?, no asse -
ion Yes
Se ice
Manage-
men based
on SLAs
o a ying
Complexi y
[R2]
Yes Yes Yes Yes Yes Yes Yes Yes Yes
Res ic able
SLA Offe s
incl. non-
nego iable
Te ms [R3]
No No No No Yes No No ?, no asse -
ion
Pa ly, usage
o s anda d
SLA lan-
guages wi h
a possibili y
o define
he nego-
iable e ms
be o ehand
P o ocol-
gene ic SLA
nego ia ions
[R4]
No No No No No No No Yes Yes
Minimal Re-
s ic ions on
he used Se
o P o ocols
[R5]
No No No No No No No No No
Au oma ed,
agen -based
se ice man-
agemen
[R6]
No ?, no asse -
ion
?, no asse -
ion Yes No Yes Yes ?, no asse -
ion
?, no asse -
ion
Ma ke In-
e media ies
[R7]
Pa ly (se -
ice b oke
a ailable,
al hough
no ac ually
b oke ed
nego ia ion
p o ocol is
employed)
Pa ly (QoS
/ Vi ual
O ganiza-
ion b oke s
a ailable,
al hough
no ac ually
b oke ed
nego ia ion
p o ocol is
employed)
No
Pa ly (in-
e media ies
a e no
di ec ly in-
ol ed in he
nego ia ion)
Pa ly (no
ac ually
b oke ed
nego ia ion
p o ocol is
employed)
Pa ly (b o-
ke a ailable
in he cen-
alized case)
Yes No No
68
2.3. Rela ed Wo k
The equi emen s de i ed in sec ion 2.1 allow o a s uc u ed assessmen and com-
pa ison o he iden ified ela ed esea ch p ojec s. As can be seen in able 2.2 none
o hese p ojec s is cu en ly capable o ulfilling all o he s a ed equi emen s. The
BabelNEG sys em, he design o which is p esen ed in he nex chap e , aims a closing
ha gap. A de ailed in es iga ion o i s capabili y o ulfill all posed equi emen s (and
hus indi ec ly i s compa ison wi h he p ojec s p esen ed abo e) is gi en in sec ion 4.2.
A e ha ing p esen ed ei he he concep ual and echnical ounda ions and al e na i e
app oaches o he s a ed esea ch p oblem along wi h a concep ual amewo k o hei
assessmen (in e ms o he s a ed equi emen s) he nex wo chap e s ocus on he
ac ual design and e alua ion o he BabelNEG sys em; my p oposed solu ion o he
s a ed esea ch p oblem.
69
3. Design and De elopmen
In his chap e , he cen al deli e able o his disse a ion p ojec , he in as uc u e
design, is p esen ed. Due o i s complexi y i is u he s uc u ed acco ding o h ee
well- ecognized pe spec i es on IT-a i ac s, which, when used in combina ion, esul in
a comp ehensi e iew on he in es iga ed sys em: da a,in e ac ion and unc ional pe -
spec i e (Fe s l and Sinz 2008, p. 137)1. This icho omy also co esponds o he h ee
al eady in oduced pe spec i es on (elec onic) nego ia ions: nego ia ion objec (da a
pe spec i e, ocusing on he documen s in ol ed in he nego ia ion), nego ia ion p o o-
col (in e ac ion pe spec i e, desc ibing he nego ia ion p ocesses) and decision making
s a egy ( unc ional pe spec i e, dealing wi h he p ocessing o messages and e en s).
Example ins an ia ions o he de eloped da a s uc u es and agen oles as well as
po en ial s a -up and usage pa ame e s a e p esen ed along wi h he demons a ion
and e alua ion s eps in chap e 4.
3.1. Abs ac Design Idea
The basic design idea, unde lying his wo k, is o offe a gi en good (SLA o an elec onic
se ice in he IoS se ing) independen ly om he way an ag eemen conce ning his
p oduc can be a ained (nego ia ion p o ocol).
Such an app oach has many analogies in eal-wo ld se ings. Many p oduc s sold in
e e yday li e a e sold wi h diffe en nego ia ion p o ocols in diffe en si ua ions. Fo
example a TV se , displayed a an elec onic e aile , implici ly s a es ha he only way
o nego ia e abou i is o accep he s a ed p ice. This p o ocol hus co esponds o
a classic ca alogue p icing model, also called Fixed P ice Auc ion (FPA)2. Then again,
he same TV se , offe ed o e an online auc ion pla o m such as eBay3, implies ha he
consume s ha e o ou bid each o he un il a ce ain deadline occu s. This co esponds
o an EA p o ocol. eBay could e en ac as an example o bo h o hese p o ocols,
since apa om he s anda d auc ions also buy i now-offe s a e a ailable, which a e
equi alen o FPAs. Al hough he p oduc sold in bo h cases is exac ly he same (a new
TV se o a gi en ype), he nego ia ion p o ocols applied a e qui e diffe en .
1Regula ly a ou h, dynamic iew conce ning he ac ual p ocesses is used. In his hesis his iew is
in eg a ed wi h he in e ac ion pe spec i e, allowing o he conjoin desc ip ion o communica ion
channels, i. e. messages, and he ac ual p ocesses building on hem.
2A FPA is basically a e y simple auc ion p o ocol in which he p ice is fixed and each bidde can only
accep he s a ed alue o lea e he auc ion. The fi s bidde sending he accep message wins he
nego ia ion, gi en only one uni o he sold p oduc is a ailable (o he wise he fi s n bidde s ecei e
one i em each, gi en n uni s a e a ailable).
3h p://www.ebay.com/
71
3. Design and De elopmen
<xsd:simpleType name="ValuesType">
<xsd: es ic ion base="xsd:s ing">
<xsd:enume a ion alue="single"/>
<xsd:enume a ion alue="mul iple"/>
</xsd: es ic ion>
</xsd:simpleType>
<xsd:complexType name="Nego iableSLOType">
<xsd:sequence>
<xsd:elemen name=" alues" ype="ValuesType" />
</xsd:sequence>
<xsd:a ibu e name="p ope yID" ype="xsd:anyURI"/>
</xsd:complexType>
<xsd:complexType name="Nego ia ionObjec Type">
<xsd:sequence>
<xsd:elemen name="nego iableSLO"
ype="Nego iableSLOType" minOccu s="0"
maxOccu s=“unbounded“ />
</xsd:sequence>
</xsd:complexType>
Figu e 3.5.: Type Decla a ion: Nego ia ion Objec Elemen
The op ional maximumNumbe O Agen s and minimumNumbe O Agen s elemen s de-
fine he maximum and minimum amoun o pe mi ed pa icipa ing agen s o ha
pa icula ole. The desc ip ion o a one-on-one ba gaining p o ocol o example, would
hus s a e a minimum and maximum numbe o agen s o bo h oles o be 1.
On he o he hand an auc ion p o ocol, would es ic one side (SPs o SCs) o a
maximum numbe o one and he o he side would possibly no expose any uppe bound
o pe mi ed agen s.
Some nego ia ion se ings equi e he joining agen s o sa is y some c i e ia o be
admi ed. Such admission es ic ions can be specified wi hin he admissionRes ic ion
elemen . I consis s o an op ional elemen o xsd:anyType (admissionRes ic ionRule),
ep esen ing a placeholde o a bi a y admission es ic ion ules exp essed in some
ex e nal ule language, and one a ibu e defining whe he admission es ic ions do
exis a all (admissionRes ic ionFo m). I open admission is defined he e, no admission
es ic ion ule is specified ( he espec i e elemen would be omi ed). In he es ic ed
case an admission ule would ha e o be p esen .
Nego ia ion Objec A e speci ying he in ol ed oles and hei pe missions, he ac-
ual nego ia ion objec has o be defined in e ms o he SLOs unde nego ia ion (see
figu e 3.5).
Following he common unde s anding o elec onic SLA documen s and hei in e nal
s uc u e, his elemen consis s o a (po en ially emp y) lis o nego iableSLO elemen s14.
Each o hese e e s o a espec i e se ice p ope y as defined in he ST; his link is
c ea ed ia he p ope y’s ID. Addi ionally, a child elemen is p esen , defining whe he
mul iple alues a e allowed o be offe ed o his SLO o no . In ui i ely no all possible
14An emp y lis would indica e, ha no SLO is nego iable, hus defining a FPA p o ocol.
78

3.2. Se ice Desc ip ion Documen s
Domain
O de ed No O de ed
Possible ope a o s o
- single alues:
=
- mul iple alues:
! = (no equal)
[enume a ion o alues]
< ; < = ; > ; > =
Possible ope a o s o
- single alues:
=
- mul iple alues:
! = (no equal)
[enume a ion o alues]
Figu e 3.6.: O de ed s. No O de ed Domains
ela ional ope a o s used o exp ess mul iple alues a e applicable in e e y ype o domain
(as defined o he espec i e SLO wi hin he ST). No -o de ed domains only allow =
o simple enume a ions o exp ess mul iple possible alues. In o de ed domains on he
o he hand, ela ions like e.g. <,≤,>and ≥a e possible.
Offe Res ic ions Each nego ia ion desc ip ion may op ionally define an a bi a y
numbe o a ibu eRes ic ion and / o gene alRes ic ion elemen s. Each o hese
ep esen s a specific es ic ion on how one pa icula SLO is o be ea ed wi hin he
nego ia ion. I , o example, a nego ia ion designe wan s o speci y ha o some
pa icula a ibu e a new offe always has o succeed he las offe , like in an EA, his
would be achie ed wi h an a ibu eRes ic ion ega ding his SLO.
Each a ibu eRes ic ion elemen hus con ains a p ope yID a ibu e e e encing he
SLO i applies o. Addi ionally, one o h ee possible es ic ion classes (modeled as child
elemen s) can be chosen:
•The p og ess elemen defines he di ec ion o a nego ia ion ega ding he e e enced
a ibu e. Wi h his elemen one can speci y whe he new offe s ha e o s a e
highe o lowe alues o his SLO, compa ed o he p e ious offe s. In ui i ely
he p og ess-elemen is only applicable o o de ed domains. In addi ion o he
di ec ion, specified in he p og essFo m child elemen , some minimum inc emen
o dec emen may be defined wi h he del a child elemen .
•The h eshold elemen defines wha is called a ese e alue in nego ia ion heo y.
A ese e alue is one nego ia o ’s uppe o lowe bound o accep able alues o
a pa icula nego ia ion a ibu e. No mally, his cons uc is only used o he
p ice, bu in his mo e gene al app oach i can also be applied o all o he se ice
a ibu es o o de ed domains. A nego ia ion designe could o example speci y
79
3. Design and De elopmen
<xsd:simpleType name="P og essFo mType">
<xsd: es ic ion base="xsd:s ing">
<xsd:enume a ion alue="ascending"/>
<xsd:enume a ion alue="descending"/>
</xsd: es ic ion>
</xsd:simpleType>
<xsd:complexType name="P og essType" mixed=“ ue“>
<xsd:sequence>
<xsd:elemen name="p og essFo m"
ype="P og essFo mType"/>
<xsd:elemen name="del a" ype="xsd:anySimpleType"
minOccu s="0" maxOccu s="1"/>
</xsd:sequence>
</xsd:complexType>
<xsd:complexType name="Th esholdType">
<xsd:choice>
<xsd:elemen name="lowe Bound"
ype="xsd:anySimpleType"/>
<xsd:elemen name="uppe Bound"
ype="xsd:anySimpleType"/>
</xsd:choice>
</xsd:complexType>
(a)
<xsd:complexType name="A ibu eRes ic ionType"
mixed=“ ue“>
<xsd:choice>
<xsd:elemen name="p og ess"
ype="P og essType"/>
<xsd:elemen name=" h eshold"
ype="Th esholdType"/>
<xsd:elemen name=" es ic ionRule"
ype="xsd:anyType"/>
</xsd:choice>
<xsd:a ibu e name="p ope yID" ype="xsd:anyURI" />
</xsd:complexType>
(b)
Figu e 3.7.: Type Decla a ion: A ibu e Res ic ion Elemen
ha in ei he case a leas 2 GB o memo y ha e o be a ailable o some se ice;
any alue lowe han ha would no be accep able.
•Finally, a es ic ionRule elemen o xsd:anyType can be used o exp ess any
a ibu e- ela ed cons ain in addi ion o h esholds and p og ess es ic ions. In
his elemen ex e nal ule languages may be employed.
Simila ly o such ee-hand es ic ion ules ela ing o exac ly one SLO each, mo e
gene al es ic ions ela ing o mo e han one se ice a ibu e can be exp essed wi h he
gene alRes ic ion elemen . This could be used o example, i a nego ia ion designe
wan s o speci y ha in a new offe a leas o one o wo diffe en a ibu es has o be
offe ed a highe alue han in he cu en offe .
Offe Alloca ion Policy The offe Alloca ion elemen defines he way he clea ing o he
nego ia ion is conduc ed, ha is how he winning offe s a e iden ified and ans o med
in o a alid SLA.
Offe ma ching can be o o wa ded o defined o m. In ei he case a alid ag eemen
is c ea ed by one side ecei ing an offe and accep ing i 15. The only diffe ence lies in
whe he he o he pa icipan s know how he offe alloca ion is conduc ed o no . In he
o wa ded case he ma ching algo i hm is no explici ly desc ibed. The accep ing agen
chooses one offe o win wi hou le ing he o he pa icipan s know acco ding o which
ules. In he defined case hese ules a e gi en wi hin he p o ocol desc ip ion and a e
he e o e openly a ailable. This way e e y agen can p edic he winning offe be o e
15Depending on he p o ocol only one side could be allowed o ac i ely accep an offe . This is defined
in he p ocess elemen o he p o ocol desc ip ion.
80
3.2. Se ice Desc ip ion Documen s
he ac ual ag eemen is pos ed o he in ol ed pa ies ( his is o example he case in an
EA, whe e he winning bid is always he one s a ing he highes p ice).
The offe Alloca ion elemen hus consis s o a ma chingFo m child elemen speci ying
whe he defined o o wa ded offe ma ching is applied and an op ional child elemen ,
called ma chingRule, o he defini ion o he ma ching algo i hm.
In o ma ion P ocessing Policy The in o ma ion a ailable du ing a nego ia ion can be
di ided in o wo da a se s: he nego ia ionS a us and he pas Offe s. The nego ia ion-
S a us is ep esen ed by all cu en offe s o all pa icipan s allowed o pos offe s16,
whe eas he pas Offe s a e all offe s pos ed in his nego ia ion un il now (some imes also
called nego ia ion his o y).
To allow o diffe en ia ed defini ions o accessible nego ia ion da a he in o ma ion-
P ocessing elemen con ains a nego ia ionT anspa ency and a s a usT anspa ency ele-
men o he pas Offe s and he nego ia ionS a us espec i ely. These wo elemen s a e
bo h es ic ed o one o h ee possible alues: public,p o ec ed and none.I public is
s a ed, he nego ia ion s a us as well as he pas offe s may be que ied by all agen s; no
es ic ion is applied o such in o ma ion eques s. P o ec ed anspa ency deno es ha
only SPs and SCs in ol ed in he ac ual nego ia ion can do so, whe eas none is used i
no da a can be que ied a all. This way, nego ia ion designe s can speci y, o example,
sealed-bid auc ions. Addi ionally, wo op ional elemen s defining he espec i e con en s
o pas offe s and s a us da a (nego ia ionCon en and s a usCon en ) can be p esen .
He e, an ex e nal es ic ion language, defining exac ly which elemen s o pas o cu -
en offe s a e isible, would ha e o be employed (in XML en i onmen s his could o
example be XQue y (Boag e al. 2007)).
Nego ia ion P ocess A e ha ing defined he s a ic aspec s o a nego ia ion, he las
elemen is dedica ed o he ac ual nego ia ion p ocess. (Nego ia ion) p o ocols a e mos ly
defined in e ms o s a e machines desc ibing how pa icipan s (o a pa icula ole) ac
du ing he ac ual communica ion. A s a e machine is defined as a se o e en s,ase o
ac ions, a se o s a es ( om which one is he ini ial and one o mo e a e final s a es) and
a ansi ion unc ion defining he ac ion and / o s a e change a pa icula pa icipan
unde akes in a gi en s a e a e a gi en e en occu ed (Ma in 2002, p. 79).
In es iga ion o a significan amoun o nego ia ion p o ocols has shown ha while
s ill ac ually nego ia ing he in ol ed pa ies no mally don’ change hei s a es ega ded
om he abs ac ion le el o he nego ia ion p o ocol17. A e ha ing s a ed a nego ia-
ion hey s ay in he NEGOTIATING s a e jus un il an ag eemen is eached (sending
o ecei ing o an accep ance message), he nego ia ion is abo ed o he pa icipan
didn’ win he nego ia ion ( ecei ing o a ejec ion message). Hence, gi en his language
16I is assumed ha each pa icipan can only ha e one alid offe a a ime and newly pos ed offe s
eplace olde ones.
17Once nego ia ing hey con inuously send and accep nego ia ion messages. In e nal sub-s a es o
he high-le el NEGOTIATING s a e and he change be ween such do no ha e any effec s on his
beha io wi hin he nego ia ion p o ocol.
81
3. Design and De elopmen
is in ended only o desc ibe nego ia ion p o ocols, no s a e changes a e an icipa ed be-
o e he final accep ance phase. The p ocess elemen as desc ibed in he ollowing will
hus build on e en s and ac ions only. This is a guable, howe e he elemen defini ion
is designed in a way ha i can easily be ex ended o inco po a e he exis ence o se e al
diffe en s a es wi hin he nego ia ion phase as well, i i becomes necessa y.
All agen s aking pa in a nego ia ion adop he SC, he SP o he Nego ia ion
Coo dina o (NC) ole. A NC is a hi d pa y go e ning he ac ual bidding p ocess,
such as o example as an auc ionee . Employing hese h ee oles, basically all he
nego ia ion p o ocols ha we e in es iga ed as pa o his wo k can be desc ibed.
Nex , he e en s and ac ions in elec onic nego ia ions a e iden ified. The e do exis
in e nal and ex e nal e en s o ac ions. As he EST documen is in ended o desc ibe
a p o ocol in a way ha ano he agen can adap o i , only ex e nal e en s o ac ion
a e conside ed. Only hose po en ially conce n he o he nego ia o , and a e hus o
in e es du ing he nego ia ion p ocess. Fo all nego ia ion p o ocols in es iga ed, he
se s o possible ex e nal e en s and ac ions could be iden ified o be equi alen ; hey a e
defined as he se o nego ia ion messages:
•Call o Bids: This message implemen s a pull mode wi hin he ac ual bidding
p ocess. An auc ionee uses his message o ask he nego ia o s o hei cu en
offe s.
•Offe : An offe is used by a nego ia o o exp ess i s cu en ly accep able ag ee-
men .
•No ifica ion Rejec : This message indica es ha he ecei e did no win he e-
spec i e nego ia ion (o he nego ia ion was abo ed)18.
•No ifica ion Accep : Vice e sa, an accep message is sen o he winne .
•S ill In e es ed: This message is used whene e a gi en SLA p oposal is offe ed o
a se o agen s and hese wan o indica e ha hey a e s ill in e es ed. This is
used in many auc ion p o ocols, whe e he indi idual bidde s a e no allowed o
accep an offe , bu only o indica e hei con inuing in e es , while he auc ionee
con inuously in- / dec eases a SLO alue un il only one in e es ed bidde is le ,
he winne o he nego ia ion.
•Admission: An admission message (only pe mi ed as an e en ) is used o define
he fi s p o ocol s ep o a p o-ac i e nego ia ion. He e, he SC does no eac on
incoming messages, i a he sends he e y fi s nego ia ion message, modeled as
a eac ion o he admission acknowledgemen message.
Employing hese messages, he beha io o a SC du ing he p o ocol can be defined.
Hence, he p ocess consis s o an un es ic ed numbe o p o ocolS ep child elemen s.
18In he cu en BabelNEG e sion his message is assumed o always be a possible ac ion and can hus
also be used by SCs o explici ly lea ing a nego ia ion.
82
3.2. Se ice Desc ip ion Documen s
<xsd:simpleType name="RoleNameType">
<xsd: es ic ion base="xsd:s ing">
<xsd:enume a ion alue="se iceP o ide "/>
<xsd:enume a ion alue="se iceConsume "/>
<xsd:enume a ion alue="nego ia ionCoo dina o "/>
</xsd: es ic ion>
</xsd:simpleType>
<xsd:simpleType name="MessageType">
<xsd: es ic ion base="xsd:s ing">
<xsd:enume a ion alue="callFo Bids"/>
<xsd:enume a ion alue="o e "/>
<xsd:enume a ion alue="s ill_in e es ed"/>
<xsd:enume a ion alue="no i ica ion_accep "/>
<xsd:enume a ion alue="no i ica ion_ ejec "/>
<xsd:enume a ion alue="admission"/>
</xsd: es ic ion>
</xsd:simpleType>
<xsd:complexType name="E en Ac ionType" >
<xsd:sequence>
<xsd:elemen name="messageType"
ype="MessageType" />
</xsd:sequence>
<xsd:a ibu e name=" om" ype="RoleNameType" />
<xsd:a ibu e name=" o" ype="RoleNameType" />
</xsd:complexType>
(a)
<xsd:complexType name="P o ocolS epsType">
<xsd:sequence>
<xsd:elemen name="p o ocolS ep"
maxOccu s="unbounded">
<xsd:complexType>
<xsd:sequence>
<xsd:elemen name="e en " ype="E en Ac ionType“ />
<xsd:elemen name="possibleAc ion"
ype ="E en Ac ionType" maxOccu s="unbounded"/>
</xsd:sequence>
</xsd:complexType>
</xsd:elemen >
</xsd:sequence>
</xsd:complexType>
<xsd:complexType name="P ocessType">
<xsd:sequence>
<xsd:elemen name="se iceConsume "
ype="P o ocolS epsType"
minOccu s="0" maxOccu s="unbounded" />
</xsd:sequence>
</xsd:complexType>
(b)
Figu e 3.8.: Type Decla a ion: P ocess Elemen
Each o hese defines an e en and a lis o possibleAc ion elemen s, ha a e possible
o he SC in case he espec i e e en occu s 19.
Bo h, e en s and ac ions, s a e a pa icula message (as specified in he MessagesType
elemen ) o be sen o ecei ed. Fo he espec i e messageType child elemen , one o
he fi e diffe en message ypes lis ed abo e can be chosen. As sende (in case an e en
is defined) o ecipien s (in case an ac ion is defined) one o he h ee possible oles SC,
SP o NC mus be s a ed.
Figu e 3.8 shows he espec i e XML Schema defini ions o he p ocess and p o ocol-
S ep elemen s20.
3.2.3. Se ice Iden ifica o
Each indi idual se ice ins ance is defined wi hin a espec i e SI. Such a documen
desc ibes whe e exac ly his se ice can be ound (impo an o he ac ual binding
p ocess), wha i s ype is and which EST is offe ed o i . I hus consis s o he ollowing
elemen s:
19A desc ip ion o he p o ocol om he NCs and he SPs iew is assumed no app op ia e since only
SCs a e po en ially en e ing new, o me ly unknown p o ocols a un ime. A SP, jus as a NC is
always implemen ed o a defined se o p o ocols and will hus no ha e he need o adap a un
ime. I would, i necessa y, jus be e-deployed wi h a new p o ocol. E en i an adap ion o he
SP is desi able his would p obably be implemen ed by jus loading ano he nego ia ion s a egy
componen (s a egy pa e n (Gamma e al. 1995, p. 315)). This again would no include any gene ic
p o ocol adap ion as in ended o he SC wi hin his hesis.
20Due o hei s uc u al simila i y one common ype defini ion was c ea ed o bo h e en and ac ion
elemen s.
83

3. Design and De elopmen
•se iceID: ID o his pa icula SI. This elemen should again be o he ype URI
in o de o ensu e uniqueness.
•se iceTypeID: link o he desc ip ion o his se ice’s ype.
•slaTempla eID: link o he EST used o his se ice.
•wsdlFile: e e ence o he WSDL file, desc ibing he ac ual se ice in e ace in
e ms o ope a ions wi h inpu , ou pu pa ame e s and e o ypes as well as i s
endpoin e e ence (EPR), defining whe e he ac ual se ice ins ance is loca ed.
•nego ia ionCoo dina o and se iceP o ide : hese wo elemen s ep esen ole
bindings o his se ice.
3.2.4. Final SLA
An elec onic SLA consis s o some in o ma ion on he SP and he SC, he ac ual se ice
and i s desc ip ion as well as o a se o gua an ees (SLOs) on how he se ice is o
be deli e ed21. Fo his, he scope o such SLOs is conside ed o be known be o e a
nego ia ion and / o can be ex ac ed and in e p e ed by he nego ia o s based on each
SLO’s decla a ion in he ST documen .
In he con ex o his hesis he WS-Ag eemen s anda d (And ieux e al. 2007) has
been used as a e e ence s uc u e o SLA documen s. Thus, he ollowing elemen s o
an elec onic SLA ha e been iden ified22:
•slaID: ID o his SLA as used o moni o ing pu poses. This elemen can also be
ound in he WS-Ag eemen specifica ion.
•con ex : con ex in o ma ion o his SLA, such as he in ol ed pa ies (SP, SC),
he IDs o he associa ed ST and EST documen s and he WSDL file. WS-
Ag eemen also defines a con ex elemen wi h e y simila con en . Howe e ,
he defini ion o he ac ual se ice (ST elemen in my app oach) is only gi en im-
plici ly in he o m o SDT and SPT elemen s. Simila ly he e e ence o he ac ual
se ice EPR is gi en in e ms o he SRTs in WS-Ag eemen (And ieux e al. 2007,
p. 20).
•Lis [SLO]: his elemen finally con ains all he nego ia ed SLOs making up his
SLA documen . In WS-Ag eemen hese elemen s a e p esen ed as GTs.
Gi en he abo emen ioned se ice documen s, a conc e e SLA documen is c ea ed by
combining elemen s o he ini ial EST, he ST and he SI documen . The final SLOs a e
based on hei defini ion (ST), he nego ia ed alues du ing he nego ia ion and / o hei
21I is assumed, ha a SLA always conce ns only one pa icula se ice ins ance, o mo e p ecisely one
pa icula configu a ion o one se ice ins ance. Con ac ual gua an ees conce ning mo e han one
se ice a e modeled as a SLA o e a composi e se ice.
22Fo each o his elemen s a sho hin on whe e his can be ound in he WS-Ag eemen specifica ion
is gi en.
84
3.3. P o ocol Design
ini ial alues as s a ed in he EST. The WSDL file, he in ol ed SP and he employed
EST and ST is de i ed om he espec i e elemen s o he SI. The SC, ep esen ing
he second con ac ual pa y is also de i ed du ing he nego ia ion o e en du ing he
admission p ocess.
3.3. P o ocol Design
In o de o suppo bo h, he disco e y and he nego ia ion o SLAs, a se o p o ocol
p imi i es has been de eloped, building on he abo emen ioned da a s uc u es. They
a e sho ly desc ibed in he ollowing.
3.3.1. Disco e y Phase
The disco e y phase basically ep esen s he se o ac i i ies ul ima ely leading o a
si ua ion whe e po en ial ansac ion pa ne s ( hus SCs and SPs) know one ano he
and can s a a nego ia ion p ocess. This means he disco e y phase is supposed o
suppo a gi en SC o find one o mo e SI documen s fi ing he sea ch c i e ia i ecei ed
om a espec i e use . To his end, he SPs should be able o publish he se ices hey
offe ( hus he espec i e SI documen s) in a way ha hey can be ound by po en ially
in e es ed SCs.
As a qui e common p oblem in SOC, a mechanism is hus needed ha suppo s he
publica ion and disco e y o elec onic documen s in a dis ibu ed se ing. A se o
he mos p ominen app oaches in ha a ea has been discussed in sec ion 2.2.2. As
al eady elabo a ed he e, he mos p omising pa adigms o he IoS a e p obably cen-
al (bu po en ially eplica ed) egis ies in combina ion wi h mul i- o b oadcas -based
messaging p o ocols o P2P-based in as uc u es. The o me offe he ad an age o
high-pe o mance o he indi idual egis y nodes and he possibili y o implemen com-
plex e ie al algo i hms he e, whe eas he la e ha e p o en o be e y esilien o
node ailu es.
Fo my p oo -o -concep implemen a ion, a e y simple single egis y node, which can
be ound ia a b oacas -based disco e y p o ocol, has been chosen. Due o he na u e o
such a p o o ype, echnical quali y me ics such as obus ness ( a o ing P2P solu ions)
a e no o p ima y in e es . The e o e, a simple a chi ec u e based on a cen al egis y
is mo e han sufficien o in es iga ing he esea ch ques ion a hand. I in u u e
e sions mo e elabo a e mechanisms become necessa y, i would none heless be possible
o in eg a e hem espec i ely, as sho ly ske ched in sec ion 5.4.
Consequen ly, a se o p o ocol p imi i es along wi h he espec i e messages o such
a disco e y phase ha e been de eloped: The fi s s ep o bo h SP and SC, when pub-
lishing o disco e ing a gi en se ice, is o find a espec i e egis y node. Fo ha
a egis yDisco e y message has been defined, con aining he ID o he sende agen .
This message is sen ia a b oadcas in o de o find one o mo e egis ies o he nex
s eps. This basically co esponds wi h he b oadcas -based mechanisms desc ibed in
sec ion 2.2.2. Subsequen ly, all egis y nodes ha ecei ed his message answe wi h an
acknowledgemen (ACK) message s a ing he EPR, whe e hey a e loca ed.
85
3. Design and De elopmen
Se ice Consume
Reques Se ice
Que y Se iceIden i ica o s
Que y unknown Se iceTypes
and SLATempla es
Choose SI
Reques admission
NEGOTIATION PHASE
Ins an ia e Gene icNego ia o
Regis y
Re ie e SIs and send
esul s
Re ie e Se iceTypes / SLA
empla es and send esul s
S o e documen s
Se ice P o ide
Publish Se ice
Publish SI and i no known
Se iceType and SLATempla e
Wai o admission
eques s
NEGOTIATION PHASE
Nego ia ion Coo dina o
Admissi
on
possible
Send admission ACK
Yes
No
Yes
No
Figu e 3.9.: Disco e y Phase O e iew
F om now on SP and SC p ocesses diffe . A SP, wan ing o publish a gi en SI,
fi s que ies whe he he espec i e ST and EST documen s associa ed wi h he SI a e
al eady a ailable a he egis y (que ySe iceTypes and que ySLATempla es messages).
In case ei he o hem is s ill unknown, he SP egis e s he espec i e documen wi h he
egis y ( egis e Se iceType o egis e SLATempla e messages)23. Once bo h documen s
a e co ec ly egis e ed (as confi med again wi h ACK messages) he ac ual SI can be
published ( egis e Se iceIden ifica o message).
The SC que ies he SI documen s a ailable a he egis y (o egis ies in case mo e
han one was ound) using a que ySe iceIden ifica o message24. A e po en ially ha -
ing ound se e al fi ing SIs he espec i e EST documen s used he ein a e que ied
(que ySLATempla es message), i no al eady known.
Gi en a lis o adequa e SIs we e ound ( o which also bo h ST and EST documen s
could be e ie ed), he agen chooses one o hem and ies o s a / join a nego ia ion.
In o de o do so, he SC mus pass h ough an explici admission p ocess, o which a
cen al en y poin is assumed, he NC. The esul ing p ocess consis s o a SC eques ing
admission o a se ice nego ia ion a he NC node, which in u n, answe s wi h an ACK
(SC was admi ed) o NACK (SC was ejec ed) message25. I is he e o e equi ed ha
he SC inse s i s c eden ials in he joinNego ia ion message, so he NC agen can make a
23This is necessa y, since espec i e SCs need bo h documen s when deciding on a SI o nego ia e o
and when adap ing o he s a ed p o ocol.
24In he cu en p o o ype sys em only he ID o a dis inc ST can be used as a sea ch c i e ia. The
que y message was howe e implemen ed in a way ha o he , mo e powe ul que ies, o example
conce ning indi idual elemen s o a ST, can be in eg a ed. In ha case, he SC would p obably fi s
que y he ST documen s o ones fi ing i s sea ch c i e ia and subsequen ly que y he SIs fi ing any
o he ound STs.
25NACK is commonly used o messages s a ing ha some eques was “no acknowledged”.
86
3.3. P o ocol Design
alid decision on whe he o no admission can be g an ed. Figu e 3.9 gi es an o e iew
on his disco e y app oach26.
The a ailabili y o a se ice ins ance is ad e ised implici ly in ha o a busy se ice
( he se ice is unning o he SP is cu en ly nego ia ing o i and joining his nego ia ion
is no possible) he NC jus ejec s new admission eques s27.
In case o a double auc ion p o ocol (CM o CDA), he gi en SI is pos ed om he SP
o he NC ia an offe ToSell message and can subsequen ly be ma ched wi h incoming
offe s o buy. A e his ma ching, he se ice is busy and will no be conside ed o
u u e ma chings un il he offe ToSell o his SI is e-pos ed by he SP agen .
In case o a success ul admission he SC and SP now engage in he ac ual nego ia ion
p ocess.
3.3.2. Nego ia ion Phase
Since he main goal o his hesis is o define an in as uc u e o p o ocol-gene ic
nego ia ions, no single one nego ia ion p o ocol can be iden ified o his phase. Ra he ,
a se o nego ia ion message ypes along wi h hei espec i e con en s has been defined.
These messages can hen be used in a gi en nego ia ion p ocess, o ches a ed acco ding
o he p o ocol desc ip ion in he EST documen .
To make su e ha as many diffe en nego ia ion p o ocols as possible can be mapped
o he defined message se , a ho ough li e a u e e iew has been conduc ed, basically
c ea ing he minimally necessa y supe se o messages used in he ound p o ocols defi-
ni ions (as lis ed o example in he FIPA p o ocol lib a y28).
The esul eflec s he undamen al cha ac e is ic o any nego ia ion p o ocol, he ac
ha i mainly consis s o a p ocess o exchanging offe s ollowed by a final accep ance
o ejec ion o one o he in ol ed pa ne s. This al eady implies an offe ,anaccep
and a ejec message ype. Addi ionally, some o he p o ocols ound inco po a e pull-
pa adigms wi hin he bidding p ocess in ha hey dis ibu e callFo Bids messages o
he bidde s, which in u n answe wi h an offe acco dingly. Finally, in some cases a
cen al auc ionee jus asks he bidde s whe he o no hey a e s ill in e es ed in he
cu en ly offe ed ag eemen , while i cons an ly changes i s offe , jus un il only one
in e es ed bidde is le , he winne o he nego ia ion. The iden ified message se hus
esul s in he message ypes al eady in oduced on page 82, excep o he admission
message, which is only used he e o model p o-ac i e p o ocols.
Using hese message ypes, a as a ie y o diffe en p o ocols can be designed and
subsequen ly desc ibed in an EST documen .
26This figu e desc ibes he disco e y p ocess om an abs ac ole-cen ic iew; no asse ion on ole
adop ions by indi idual agen s a e made. A de ailed desc ip ion o his aspec is gi en on page 89.
27In o de o gi e a ejec ed SC agen a meaning ul eason o i s ejec ion (i could be ejec ed on he
basis o i s c eden ials o because he se ice is una ailable a he momen ) a comp ehensi e aul
model should be used, building on a se o s anda dized aul ypes.
28www.fipa.o g
87
3. Design and De elopmen
INITIATION NEGOTIATION s a e and s a o e om he e.
Du ing he EXECUTING s a e his leads he SC o be able o change o he INITI-
ATION NEGOTIATION, he DISCOVERY o he IDLE s a e, depending on whe he
he in oca ion was success ul o no , whe he o he SIs we e ound he agen can s a
a new nego ia ion wi h o , i no , whe he o he se ice eques s ha e been ecei ed in
he mean ime.
Simila ly o he SP, his diag am does no illus a e he ac ual nego ia ion phase. The
in e nal ou ines enabling a SC o adap o a gi en p o ocol a e desc ibed sepa a ely in
subsec ion 3.4.6.
3.4.4. In e nal Beha io : Regis y
The RA node is a placeholde elemen o any disco e y mechanism used in u u e
e sions. I was designed o simply ecei e egis e messages and s o e he co esponding
da a in o an in e nal da a s uc u e. This is o example used when a SP publishes a
new SI documen . Suppo ing he disco e y o se ice documen s, i also offe s he
possibili y o que y he s o ed SI, EST and ST documen s. No complex in e nal s a es
a e main ained and changed h oughou i s li e ime. I simply p ocesses eques s o
egis e o que y se ice documen s o answe s egis yDisco e y messages.
This e y basic egis y se ice was conside ed app op ia e due o he ac ha he
de eloped sys em only ac s as a p oo -o -concep . Fu u e e sion will p obably eplace
i wi h a mo e sophis ica ed egis y implemen a ion o e en use mo e decen alized
disco e y mechanisms.
3.4.5. In e nal Beha io : Nego ia ion Coo dina o
The main ask o a NC agen is o handle he admission o SCs o a gi en nego ia ion. Fo
his, i offe s he possibili y o submi espec i e messages along wi h a se o c eden ials,
as needed o he admission decision.
Addi ionally, i he nego ia ion phase is also assigned o he NC (configu a ion C), i
mus also be able o p ocess incoming nego ia ion messages. Jus wi h configu a ions
A and B (see page 89) hese a e simply o wa ded o a nego ia ion s a egy componen ,
which in u n p o ides he p o ocol-specific unc ionali y.
Simila ly o he RA nodes, a NC hus does no expose complex in e nal s a es and
s a e changes. Admission eques s a e e alua ed and answe ed based on he ecei ed
da a and he implici se ice a ailabili y in o ma ion, and nego ia ion ela ed messages
a e simply o wa ded o he s a egy componen .
3.4.6. The p o ocol-gene ic SC S a egy Componen
The s a egy subcomponen o an SC agen ep esen s one o he cen al deli e ables o
his hesis, as basically all ou ines needed o adap ion o a new nego ia ion p o ocol
a e implemen ed he ein.
Du ing he ins an ia ion p ocess o a espec i e Gene ic Nego ia o (GN) componen ,
i ecei es om he SC:
94

3.4. A chi ec u e Design
• he se ice desc ip ion documen s,
• he ID o he cu en nego ia ion,
•a link back o he SC,
• he cons ain s he use posed on he SLA o be nego ia ed (e.g. he ese a ion
alues) and
•some addi ional nego ia ion pa ame e s, such as imeou o concession alues.
As a fi s s ep, he GN pa i ions he s a ed SLOs in o hose ha a e fixed and hose
ha can be nego ia ed. The o me a e simply s o ed. Fo each o he la e a new
SLONeg objec is c ea ed, con aining bo h he ini ial SLO (p ope y ID and alue) and
all addi ional in o ma ion ha is associa ed wi h i (e.g. a ibu eRes ic ions o domain
desc ip ions). These objec s a e hen again sa ed and subsequen ly ea ed as he fi s
offe om he nego ia ion pa ne .
Nex , he SLOCons ain objec s ini ially ecei ed om he use a e p ocessed, each
o which basically desc ibes he use ’s demands on one pa icula SLO unde nego ia ion
(fixed o nego iable). Following he a ionale om abo e, wo ypes o such cons ain
classes a e p esen : Enume a ionSLOCons ain s o O de edSLOCons ain s, depending
on he alue domain o he unde lying p ope y37.
Bo h p o ide me hods o eques ing he s a ed es ic ions (e.g. uppe o lowe
ese a ion alues o concession s eps in case o an o de ed domain) and simple me hods
calcula ing whe he o no a gi en ( ecei ed) alue ulfills his cons ain o , con a ily,
is so a away om any accep able alue ha his offe can be ega ded as ejec able.
Addi ionally, hey also p o ide he GN wi h a me hod o c ea ing coun e offe s based
on a ecei ed alue.
An Enume a ionCons ain objec basically s a es a se o accep able alues, based
on which i can decide whe he o no a ecei ed SLO alue is accep able. In ui i ely
ecei ed alues applying o an Enume a ionCons ain a e ne e ejec able, as no as-
se ion can be made whe he o no he nego ia ion pa ne would accep one o he
demanded alues ( hese alues do no expose any ela ionship om which such an asse -
ion could be deduced). In he cu en p o o ype, when c ea ing a coun e offe , simply
one accep able alue is andomly chosen and e u ned38.
Con a ily, an O de edSLOCons ain defines a ange o accep able alues, egula ly
only bounded on one side (ei he a minimum o maximum ese a ion alue is p esen 39).
Consequen ly an offe is deemed accep able based on hese bounda y alues. Deciding
37Fo each cons ain ecei ed om he use an in e nal cons ain objec , ep esen ing i , is c ea ed
du ing ini ializa ion o he GN.
38In u u e e sions, po en ially a lis o all accep able alues will be sen in o de o assu e de e minism
o he nego ia ion.
39I is assumed ha a nego ia o has a clea p e e ence on a o de ed nego ia ion pa ame e . She will
he e o e accep all alues as long as hey a e highe / lowe han he ese a ion alue. A SC will
o example accep a p ice as long as i is lowe ha a gi en ese e alue bu will no ha e any lowe
bounda y o he accep able alues.
95
3. Design and De elopmen
whe he o no an offe is ejec able is mo e complica ed, as i diffe s om one use o
he o he .
Fo he c ea ion o a coun e offe , wo cases mus be dis inguished: an offe ed alue
o he espec ed SLO was al eady ecei ed o no . In he la e case, his agen pos s he
fi s offe o his SLO. He e i simply offe s he bes alue accep able o he SP (s a ed
as uppe o lowe ese a ion alues in he EST) o i such alues a e no specified i
offe s hal i s own uppe ese a ion alue / double i s lowe ese a ion alue40.In he
o me case i jus concedes om i s las ly pos ed offe by an amoun , specified by he
use o his pa icula SLO du ing ins an ia ion (concession s ep alue).
Once all his in o ma ion was ex ac ed he GN checks whe he i has o s a he
nego ia ion (p o-ac i e p o ocol). I so, wo possible ac ions can be defined: his agen
can be allowed o pos an offe o o simply accep all he alues s a ed in he EST
(ca alogue p icing model).
He e, a basic p inciple o he GN becomes ob ious: i always seeks o maximize i s
p ofi and hus will always choose some ac ions o e o he s (i bo h a e allowed). In his
case (p o-ac i e p o ocol) i will fi s check, whe he an offe is possible, which could
u he imp o e he cu en ly offe ed ag eemen . I so, an offe will be sen ; i no i will
simply accep he cu en alues, as no nego ia ion on hem is allowed.
The same p inciple applies h oughou he whole nego ia ion p ocess. Whene e a
nego ia ion message is dispa ched o he GN i checks i s op ions: In case o a ejec o
accep message he nego ia ion is o e . I simply p ocesses he esul in ha i passes
he espec i e in o ma ion / eached ag eemen o he SC agen . In case o a callFo Bid
message i c ea es an offe and sends i o he SP agen 41. All hese possibili ies a e
s aigh o wa d, as hey don’ gi e he GN any op ion o choose among a se o possible
ac ions.
When ecei ing an offe , his could po en ially change. I he offe is no comple ely
ejec able ( his is checked by i e a ing o e all in ol ed SLOCons ain objec s) and a
coun e offe is possible, he GN will always do so. A coun e offe is always he bes
op ion, as i po en ially inc eases alue o he ag eemen o be eached o he use .
I his is no possible, he GN will check whe he a s illIn e es ed message is allowed,
p o iding i wi h he possibili y o an ongoing nego ia ion, e en i i canno ac i ely
influence he changing o he nego ia ed alues. I e en his is no possible i will finally
check whe he he ecei ed offe can be accep ed o in he end ejec ed comple ely and
do so. This ou ine gi es he GN he possibili y o eac on incoming messages in a
way ha maximizes i s u he op ions du ing he nego ia ion and in he end po en ially
e en i s u ili y in e ms o he eached ag eemen .
Summa izing his chap e , a de ailed o e iew on he de eloped se ice desc ip ion
documen s, he employed disco e y and nego ia ion p o ocols and he designed so wa e
agen s employed he ein has been gi en. In he nex chap e I desc ibe how hese
concep s can be ins an ia ed and how he p o o ype sys em can be deployed and s a ed.
40These ac o s a e chosen andomly and would be p o ided by he use in u u e e sions.
41Offe messages a e simply c ea ed by i e a ing o e each no ye ulfilled cons ain objec and ig-
ge ing each o p o ide a coun e offe alue, as desc ibed be o e. These alues a e hen combined o
one bundled offe message.
96
3.4. A chi ec u e Design
Also, he esul s o a demons a ion o he p o o ype’s effec i eness (wi h ega d o he
s a ed equi emen s) as well as i s efficiency (e alua ion s ep) a e gi en.
97
4. Assessmen o he De eloped Sys em
A e ha ing p esen ed he in as uc u e design in he las chap e , he esul s o i s
assessmen a e p esen ed in he ollowing.
4.1. P o o ypical Sys em Implemen a ion
The concep s in oduced in chap e 3 ha e been implemen ed in a Ja a-based p oo -
o -concep p o o ype in as uc u e, building on he agen -based IoS simula ion oolki
SimIS (K¨onig, Hude , and Eymann 2010). This way a mul i ude o diffe en config-
u a ions, e en e y ex eme ma ke si ua ions, can be assessed only ocusing on he
nego ia ion and disco e y phases, be o e he componen s a e po ed o a p oduc i e IoS
pla o m dealing wi h he whole se ice li e cycle.
This oolki aims a p o iding esea che s wi h a comp ehensi e amewo k o in-
es iga ing dis ibu ed algo i hms o p o ocols wi hin he con ex o he IoS. Building
on he gene ic Recu si e Po ous Agen Simula ion Toolki (REPAST) (No h, Collie ,
and Vos 2006), i p oposes a wo- ie ed a chi ec u e di iding he o e all sys em in o an
Applica ion Laye (AL) and an In as uc u e Laye (IL) (see figu e 4.1).
The IL models opological se ings o he IoS, whe eas he AL ep esen s he ac ual
se o se ice managemen agen s and espec i e se ices. The undamen al idea is ha
all AL agen s a e linked o a single IL agen each, which ep esen s hei se e pla o m.
This in as uc u al node deals wi h message handling and ou ing issues.
In he AL he ac ual se ices o he IoS ision a e defined, communica ing ia he
offe ed messaging in e aces and ou ing unc ionali ies offe ed by he IL1. Each se ice
(as ep esen ed by an AL agen ) is implemen ed as a plain Ja a class and can he e o e
exploi he ull po en ial his p og amming language offe s.
The agen ypes and messages de eloped o his p o o ype ha e consequen ly been
implemen ed as specific AL agen s and Ja a message objec s wi hin SimIS. Each o he
managemen agen s (SCs and SPs) is addi ionally accompanied by a s a egy objec ,
he SCs wi h he p o ocol-gene ic GN componen and he SPs wi h a s a egy dis inc ly
fi ing o he offe ed p o ocol (as pa ame e ized du ing s a up).
1The message s uc u es and se ice / agen in e aces, building on hem, ha e been designed equi -
alen ly o eal-wo ld SOAP messages and WS in e aces. This eases po ing he p o o ype o a
p oduc i e in as uc u e building on WS echnologies ( he de ac o s anda d o such in as uc u es
as o oday) in he u u e.
99

4. Assessmen o he De eloped Sys em
Applica ion Laye
In as uc u e Laye
Se ice Managemen Agen s
In as uc u e Nodes
Figu e 4.1.: SimIS Toolki
4.2. Concep ual Demons a ion
As desc ibed in 1.3.2, his sec ion assesses he effec i eness o he de eloped in as uc-
u e. In his demons a ion s ep he capabili y o he designed mechanisms and da a
s uc u es o ulfill he posed equi emen s is in es iga ed.
4.2.1. Assessmen on he Basis o he S a ed Requi emen s
Recapi ula ing he equi emen s s a ed in 2.1, he ollowing asse ions can be made:
R1: Need o ma ke egis ies The disco e y phase in my app oach explici ly
builds on a egis y ole (RA) o publica ion and disco e y o he se ice desc ip ion
documen s (EST, ST and SI).
R2: Suppo o SLAs o diffe en complexi y as a concep ual basis o he
se ice managemen
The ST, EST and SLA documen s p o ide a e y gene ic se ice (SLA) desc ip ion
s uc u e. By offe ing ule-based ex elemen s along wi h he p e-defined and yped ele-
men s o quan i a i ely measu able se ice aspec s, a comp ehensi e se ice desc ip ion
can be c ea ed. The defined documen s uc u es also enable he usage o ex e nal, s an-
da d languages o desc ibing se ice cha ac e is ics ( o example WSDL when desc ibing
he se ice in e ace).
R3: Suppo o es ic able SLA offe ings, including non-nego iable e ms.
The sepa a ion o offe ed se ice pa ame e s (defined in he ST documen ) and he
asse ions on hei nego iabili y as well as hei s a ing alues (s a ed in he EST docu-
men ) di ec ly co esponds o ha claim. By defining e-usable ST and EST documen s,
a e y gene ic app oach o es ic ing indi idual SLA pa ame e s ( o p e-se alues) is
offe ed.
R4: Suppo o p o ocol-gene ic SLA nego ia ions
This ep esen s he mos undamen al claim made in his hesis. The main goal o my
wo k is o define mechanisms and da a s uc u es allowing o he dynamic c ea ion o
knowledge abou and subsequen adap ion o a p io i unknown p o ocols o nego ia ing
elec onic SLAs.
Each SC agen , p esen in he de eloped in as uc u e is accompanied by a p o ocol-
100
4.2. Concep ual Demons a ion
gene ic nego ia ion s a egy componen , he GN. This module is capable o pa sing
nego ia ion p o ocol desc ip ions coded wi hin EST documen s and subsequen ly aking
pa in he espec i e nego ia ion p ocess. No p io knowledge abou he p o ocol o be
execu ed is necessa y o his.
A mo e de ailed p oo o easibili y is gi en in subsec ion 4.3, unde lining he s a emen
jus made.
R5: A p io i un es ic ed se o possible nego ia ion p o ocols
The GN componen is capable o adap ing o e e y p o ocol ha can be desc ibed
wi hin an EST documen . The only es ic ion posed on he se o possible p o ocols
is hus he s uc u al es ic ion o he EST documen and he used on ology o SLO
pa ame e s ( his is assumed o be defined o he gi en indus y he in as uc u e is
used in).
Hence, gi en he EST documen offe s a high enough gene i y o desc ibing diffe en
p o ocols, his equi emen can be assumed ulfilled. In subsec ion 4.2.2 his exp essi e-
ness is assessed in mo e de ail.
R6: So wa e agen s ac ing as nego ia o s
The sys em a chi ec u e, as implemen ed in he p oo -o -concep p o o ype, hea ily
builds on so wa e agen s as ins an ia ions o he defined oles. Such componen s a e
he basic ac o s in he BabelNEG sys em.
R7: Need o in e media ies
The de eloped in as uc u e inco po a es an explici in e media y ole, used o he
defini ion o ma ke b oke s, he NC.
In summa y, he equi emen s can be conside ed ulfilled, once he exp essi eness o
he EST in desc ibing diffe en nego ia ion p o ocols and he GN’s abili y o adap o
such desc ip ions can be shown. Respec i e effo s a e made in he ollowing.
4.2.2. Concep ual Assessmen o he Se ice Desc ip ion Documen s
The o e all goal o he se ice desc ip ion documen s is o enable he defini ion o all
nego ia ion p o ocols and SLA cha ac e is ics commonly used in elec onic nego ia ion
se ings. Howe e , he e is no exhaus i e lis o such aspec s, upon which he exp es-
si eness o he language could be e alua ed. The e o e, o bo h he se ice / SLA and
p o ocol desc ip ion pa ame e s, ep esen a i e examples mus be iden ified and u he
used o demons a ion pu poses.
The desc ip ion o an elec onic se ice o SLA documen commonly comp ises a se o
unc ional and non- unc ional cha ac e is ics o he espec i e se ice ins ance. A SLA
documen addi ionally specifies some con ex elemen s, such as he in ol ed SC and SP.
This basic s uc u e can be ound in all ele an s anda d languages cu en ly in use (see
o example And ieux e al. 2007; Lamanna, Skene, and Emme ich 2003; Ludwig e al.
2003a).
The ein, he unc ional p ope ies o a se ice a e desc ibed using some in e ace de-
sc ip ion language, such as WSDL, some imes also augmen ed wi h seman ic anno a-
ions. On he o he hand, he non- unc ional aspec s a e defined as uples o QoS me ic
defini ions and espec i e alues.
101
4. Assessmen o he De eloped Sys em
<Se iceType>
<se iceTypeID>
h p://www.abc.com/demandFo ecas ing
</se iceTypeID>
<se iceDesc ip ion>
h p://www.abc.com/demandFo ecas ing/ unc ionali y
</se iceDesc ip ion>
<p ope y p ope yID="sla/p ice">
<domain>Double</domain>
<decla a ion>h p://www.sla.o g/on ology/
p ice</decla a ion>
</p ope y>
<p ope y p ope yID="scm/ o ecas ingAlgo i hm">
<domain>S ing</domain>
<decla a ion>h p://www.scm.o g/on ology/
o ecas ingAlgo i hm</decla a ion>
</p ope y>
<p ope y p ope yID="sla/alloca edMemo y">
<domain>Double</domain>
<decla a ion>h p://www.sla.o g/on ology/
alloca edMemo y</decla a ion>
</p ope y>
</Se iceType>
(a) ST Documen
<Se i ceIden i ica o >
<se iceID>
h p://www.xyz.com/demandFo ecas ing
</se iceID>
<se iceTypeID>
h p://www.scm.com/demandFo ecas ing
</se iceTypeID>
<slaTempla eID>
h p://www.slaNeg.com/s anda dDu chAuc ionTempla e
</slaTempla eID>
<wsdlFile>
h p://www.xyz.com/demandFo ecas ing/
o ecas ingSe ice.wsdl
</wsdlFile>
<nego ia ionCoo dina o >
h p://www.xyz.com/demandFo ecas ing/
du chAuc ionNego ia o
</nego ia ionCoo dina o >
<se iceP o ide >
h p://www.xyz.com/demandFo ecas ing/
du chAuc ionNego ia o
</se iceP o ide >
</Se i ceIden i ica o >
(b) SI Documen
Figu e 4.2.: Sample Se ice Desc ip ion Documen s
In he BableNEG sys em, he cha ac e is ics o a se ice a e defined in he ST, he SI
and, a e success ul nego ia ion, in he SLA documen as p esen ed in sec ion 3.2:
The unc ional aspec s o a se ice a e exp essed wi hin he ST and he WSDL file2,
which is e e enced in he SI documen . The in ol ed pa ies (be o e he SLA is signed
only SP and NC a e known) a e also s a ed he ein. Addi ionally, he non- unc ional
aspec s a se ice o a gi en ype exposes, and can hus offe gua an ees on, a e defined
in he ST documen . The ac ual alues o hese me ics a e no necessa ily defined
be o e he ac ual nego ia ion. In con as , a se ice mos ly only offe s he possibili y o
defining a gua an ee on a pa icula SLO be o e he ac ual SLA nego ia ion. The ac ual
alue o his gua an ee is egula ly ag eed upon du ing he nego ia ion p ocess.
I some o he me ics a e al eady associa ed wi h alues be o e a nego ia ion, his
indica es ei he a s a ing alue o be a gued o e subsequen ly o a fixed cha ac e is ic
ha canno be al e ed o he desc ibed se ice ins ance (SLO is non-nego iable). Such
ini ial SLO alues a e s a ed in he EST documen when defining he nego ia ion objec .
A sho demons a ion o hese aspec s is gi en in figu es 4.2(a), 4.2(b) and 4.3, show-
ing example ST, SI and SLA documen s, adhe ing o he schema defini ions p esen ed
in sec ion 3.2.
They desc ibe a demand o ecas ing se ice, offe ing h ee non- unc ional a ibu es:
he p ice, he alloca ed memo y (de e mining he du a ion o each o ecas ing un) and
he used o ecas ing algo i hm i sel . The fi s wo a e o an o de ed domain (Double),
ep esen ing US Dolla (USD3) and gigaby e (GB) alues espec i ely, as opposed o
2In adi ional SOS, only such a WSDL desc ip ion (po en ially wi h some seman ic ex ensions, as o
example used in (O e hage and Thomas 2005)) is used.
3The used measu e o a SLO is assumed o be p e-defined in he con ex o he a ge indus y o ,
al e na i ely, specified in he ST documen .
102
4.2. Concep ual Demons a ion
<SLA>
<slaID>
h p://sla.o g/14294
</slaID>
<con ex >
<se iceP o ide >
h p://www.xyz.com/demandFo ecas ing/
du chAuc ionNego ia o
</se iceP o ide >
<se iceConsume >
h p://www.abc.com/scm/nego ia o
</se iceConsume >
<se iceID>
h p://www.xyz.com/demandFo ecas ing
</se iceID>
<se iceType>
h p://www.scm.com/demandFo ecas ing
</se iceType>
<slaTempla e>
h p://www.slaNeg.com/s anda dDu chAuc ionTempla e
</slaTempla e>
<wsdlFile>
h p://www.xyz.com/demandFo ecas ing/
o ecas ingSe ice.wsdl
</wsdlFile>
</con ex >
<SLO p ope yID="sla/p ice">
< alue>1.50</ alue>
</SLO>
<SLO p ope yID="scm/ o ecas ingAlgo i hm">
< alue>bayesianFo ecas ing</ alue>
</SLO>
<SLO p ope yID="sla/alloca edMemo y">
< alue>4.0</ alue>
</SLO>
</SLA>
Figu e 4.3.: Example SLA Documen
he hi d, exposing an uno de ed domain (S ing, ep esen ing he algo i hms name).
Addi ionally, a WSDL file o his se ice can be ound a “h p://www.xyz.com/demand
Fo ecas ing/ o ecas ingSe ice.wsdl”.
A e he nego ia ion, a SLA documen was achie ed be ween he SP “h p://www.xyz
.com/demandFo ecas ing/du chAuc ionNego ia o ” and he SC “h p://www.abc.com/
scm/nego ia o ” defining a p ice o 1.50 USD. The “bayesianFo ecas ing” algo i hm is
o be used and an amoun o 4 GB o memo y is alloca ed o his se ice.
In he ollowing, he documen s’ capabili ies o desc ibe ac ual nego ia ion p o ocols
a e assessed. Fo his, a ep esen a i e se o mu ually as diffe en as possible p o ocols4
has been selec ed. Such an app oach canno claim comple eness, howe e since he e
is no comp ehensi e lis o possible p o ocols, he e is no app oach able o claim ha .
By choosing e y diffe en p o ocols a leas a sufficien ly gene alizable asse ion on he
language’s exp essi eness can be made.
The p o ocols chosen o his assessmen a e an EA, a DA (bo h 1:N p o ocols), a
FPA, a CM (M:N p o ocol)5and wo ypes o AO p o ocols (1:1 p o ocols), one being
4The chosen p o ocols a e mainly based on he FIPA in e ac ion p o ocol lib a y, being one o he mos
commonly used p o ocol axonomies o agen -based in e ac ions.
5This p o ocol ollows he M+1 clea ing algo i hm, p esen ed in (Wu man, Walsh, and Wellman 1998b).
103
4. Assessmen o he De eloped Sys em
Se ice Consume :Se ice P o ide :
loop
b eak
al
[p o ide accep s]
[consume accep s]
[bid is accep ed]
[1,*]
1: o e
1: no i ica ion_accep
1.1.1.1: o e
1.1.1: e alua e p oposal
1: e alua e p oposal
1.1: o e
1: no i ica ion_accep
1: o e
1: e alua e p oposal
1.1: o e
1.1.1: e alua e p oposal
1: no i ica ion_accep
1: no i ica ion_accep
1.1.1.1: o e
Figu e 4.10.: Sequence Diag am ep esen ing an AO / MAO P o ocol
ins ead o he in e nal s a egy module. The jus men ioned b oke s a e finally assumed
o suppo one dis inc auc ion p o ocol o one dis inc se ice ype each, allowing o
s uc u ed b oke disco e y p ocesses.
The quan i y and dis ibu ion o all hose agen s, hus defining he simula ion se ing,
a e specified in a configu a ion file ha is pa sed du ing simula ion s a up. Addi ionally,
a second file defines all a ailable ST and EST combina ions offe ed in his ma ke . Du ing
s a up he espec i e agen s a e ins an ia ed ( he SPs, ISPs and NCs a e addi ionally
pa ame e ized wi h he ST and EST ypes hey suppo ) and loca ed on he opology
as s a ed in he configu a ion files. In a second s ep he se ice desc ip ion documen s
a e gene a ed and passed o he SPs / ISPs o publica ion14.
In o de o allow o andom demand gene a ion a he SC agen s, all STs a ailable
in he sys em a e addi ionally s o ed a a DemandHelpe node. Du ing he ac ual sim-
ula ion, each SC, once igge ed in e nally by a imeou mechanism, eques s a andom
se ice eques om his helpe module. This esul s in a sys em whe e only se ices
14Due o he Ja a-based implemen a ion hese documen s a e in e nally ep esen ed as Ja a objec s.
Howe e , a pa se has been de eloped capable o c ea ing hese objec s om espec i e documen s
coded in XML.
110

4.3. Simula i e Demons a ion o he P o o ype’s Effec i eness
<Ex endedSLATempla e>
<slaTempla eID>al e na eO e sTempla e</slaTempla eID>
<slaTempla e>
...
</slaTempla e>
<nego ia ionP o ocol>
<con ex >
...
<se iceConsume >
<maximumNumbe O Agen s>
1
</maximumNumbe O Agen s>
<minimumNumbe O Agen s>
1
</minimumNumbe O Agen s>
<admissionRes ic ion admissionRes ic ionFo m="open"/>
</se iceConsume >
</con ex >
<nego ia ionObjec >
<nego iableSLO p ope yID="sla/p ice">
< alues>single</ alues>
</nego iableSLO>
</nego ia ionObjec >
(a)
...
<p ocess>
<se iceConsume >
<p o ocolS ep>
<e en om="se iceP o ide " o="se iceConsume ">
<messageType>o e </messageType>
</e en >
<possibleAc ion om="se iceConsume "
o="se iceP o ide ">
<messageType>o e </messageType>
</possibleAc ion>
<possibleAc ion om="se iceConsume "
o="se iceP o ide ">
<messageType>no i ica ion_accep </messageType>
</possibleAc ion>
</p o ocolS ep>
</se iceConsume >
</p ocess>
</nego ia ionP o ocol>
</Ex endedSLATempla e>
(b)
Figu e 4.11.: Sample EST Documen o he AO P o ocol
ac ually p esen on he ma ke a e demanded in he simula ed use eques s15.
Simula ion Pa ame e s and hei Usage wi hin he Agen s
Fo each simula ion un a se o pa ame e s ha e o be se : The one global pa ame e
is he amoun o icks he simula ions should un. This delimi s he du a ion o he
expe imen s. Addi ionally, he e exis s a a ie y o local pa ame e s, applied only wi hin
pa icula agen classes.
B oke Agen (ac ing as an independen NC) The b oke only has one pa ame e ,
he imeou ha defines he ime in e als, a e which a ma ching s ep is conduc ed
(ma chingTimeou )16.
SP / ISP Agen Bo h SPs and ISPs in e nally use he ollowing imeou ypes:
•findRegis yTimeou , used du ing egis y disco e y.
•publica ionTimeou , used du ing he publica ion p ocess.
•execu ionTimeou , used a e he nego ia ion when he se ice is ac ually in oked.
15Since he ocus o my wo k lies on he in es iga ion o he adap abili y o he de eloped agen s, eques s
o no a ailable se ice ypes would no p o ide any addi ional alue wi hin he simula ions. Hence,
hey we e omi ed.
16Due o he ime-disc e e simula ions all imeou s a e exp essed as an amoun o icks ha is discoun ed
e e y ick.
111
4. Assessmen o he De eloped Sys em
•auc ionTimeou , used o delimi he ime an agen wai s o he fi s joinNego ia-
ion message o a i e be o e decla ing his nego ia ion a emp o be ailed and
s a o e .
When accompanied wi h a EA, DA o (mul i-a ibu i e) AO s a egy, a SP also
needs a imeou delimi ing he ime i wai s o answe s om he SCs du ing a unning
nego ia ion (biddingTimeou ) be o e decla ing i o be abo ed o s a ing a new ound
(EA and DA).
Finally, he ISP employs wo addi ional imeou s, he findNCTimeou (employed in
he b oke disco e y) and he nego ia ionTimeou (delimi ing he maximum ime he
agen wai s o an answe message om he b oke ).
Apa om he imeou s, a se o pa ame e s defines he way he indi idual SLOs
offe ed should be ea ed du ing he nego ia ion.
The es ima edMa ke Value defines he alua ion an agen assigns o a gi en SLO. This
alue is also used o calcula ing he s a ing offe s. Depending on he alue p e e ence
(s a ed in he heHighe TheBe e boolean pa ame e , also gi en o each SLO) hese
s a ing alues a e se o double o hal he es ima edMa ke Value. In an EA o example,
gi en he es ima ed ma ke alue o he se ice o be sold is 10 USD, he s a ing alue
would be se o 5 USD.
The ma ke ValueAdop ion a e (only gi en o he p ice SLO17) defines how an agen
adap s i s ma ke alua ion o a gi en se ice o e ime. Whene e an agen was no able
o each an ag eemen (e.g. when a nego ia ion was abo ed due o elapsing imeou s
o i no mu ually accep able offe s could be ound), i educes i s alua ion o he
nego ia ed se ice. This will imp o e i s chances o winning he nex nego ia ion i
engages in. When an agen wins a nego ia ion, i ac s ice e sa, ying o ge an e en
be e deal he nex ime. This e y simple lea ning algo i hm p o ides he simula ion
uns wi h a dynamic elemen , also p e en ing single agen s om ne e being able o
each an ag eemen , jus because o hei w ong alua ion o a gi en se ice.
SPs a e also pa ame e ized wi h a concessionS ep alue (one pe SLO), used in ba -
gaining p o ocols. This defines he alue an agen is willing o concede om i s las offe
in o de o con e ge o a mu ually accep able ag eemen .
Finally, all SPs and ISPs offe a pa ame e ha delimi s he dis ibu ion o b oad-cas
messages, he maximumHops ac o .
SC Agen On he o he hand, he SC agen s define he ollowing imeou pa ame e s:
•disco e yTimeou , delimi ing he amoun o icks he SC can ake o find a SI
and e ie e he espec i e EST and ST documen s be o e his eques is decla ed
unsuccess ul o a fi ing SI is chosen o s a he nego ia ion phase.
•findRegis yTimeou , used du ing egis y disco e y.
17Lea ning is only in oduced o he p ice a ibu e, as his gene a es enough dynamic o he simula ion
uns and he o he SLOs a e assumed no o be as dynamically adap able as he ac ual alua ion o
ase ice.
112
4.3. Simula i e Demons a ion o he P o o ype’s Effec i eness
•joinNego ia ionTimeou , delimi ing he ime his SC wai s o an answe o a join-
Nego ia ion message.
•execu ionTimeou , used du ing he se ice in oca ion.
•wai Fo Execu ionTimeou , needed in case o a b oke ed nego ia ion ( ime he SC
wai s be o e ying o in oke he se ice; his gua an ees ha he ISP has ecei ed
and p ocessed he no ifica ion message om he b oke in he mean ime).
Finally, he SC is also gi en a ma ke ValueAdop ionRa e,concessionS ep and ese e
alues espec i ely18 as well as a maximumHops pa ame e .
Demons a ion Me ics
When assessing he esul s o a simula ion un, h ee dis inc me ics a e o p ima y
in e es :
• he absolu e numbe o success ully execu ed nego ia ions pe p o ocol.
• he minimum and maximum numbe o success ully execu ed nego ia ions pe p o-
ocol pe SC agen .
• he minimum numbe o diffe en p o ocols a SC could success ully ake pa in.
The fi s me ic desc ibes a global iew on he simula ion expe imen , s a ing how
many nego ia ion p ocesses could success ully be finished o each p o ocol ype, p esen
in he se ing (as deno ed by a dis inc EST).
The second in es iga es his esul on a pe -agen basis as i acks he minimum and
maximum amoun o success ul nego ia ions pe p o ocol any one SC in he sys em could
achie e. This gi es a hin on how he success ul nego ia ions we e sp ead o e he agen
popula ion.
The final me ic shows how di e se he used nego ia ion p o ocols we e in he iew o
he indi idual agen s, s a ing how many diffe en p o ocols he agen s ha e been in ol ed
in.
Expe imen s and In e p e a ion
Now he esul s achie ed in simula ing a ious ma ke configu a ions a e p esen ed.
Du ing all expe imen s a o al amoun o 10 IL agen s was employed in a ull meshed
ne wo k. Since I wan ed o in es iga e he easibili y o un ime p o ocol-gene i y wi h
as ew side effec s as possible, I ied o configu e he simula ion se ings in a way
hey esemble a pe ec ma ke . A ull-meshed opology especially suppo s he pe ec
18These alues a e p o ided by he DemandHelpe module pe eques (in o m o SLOCons ain
objec s). Howe e , all ecei ed ma ke alua ions a e in e nally o e w i en wi h he cu en ma ke
alua ion o ha se ice ype, as lea ned h oughou se e al nego ia ions (gi en he agen has al eady
accumula ed espec i e knowledge). This way he demand gene a ion is adap ed o he expe ience
he agen has al eady made.
113
4. Assessmen o he De eloped Sys em
dis ibu ion o ma ke in o ma ion (as coded wi hin he se ice desc ip ion documen s)
among he pa icipan s. Mo e complex opologies could in he wo s case esul in some
o he se ices a ailable no being ound and hus nego ia ed o by he SC agen s,
leading o a dec ease in he o al amoun o success ul nego ia ion. Hence, ull meshed
opologies we e conside ed app op ia e o he esea ch ques ion a hand.
Each o he connec ions was pa ame e ized wi h a bandwid h alue ensu ing ha a
comple e message o he size used in hese expe imen s a e ses one connec ion in one
ick. The simula ion i sel was delimi ed o 20.000 icks and o all expe imen s 50 SCs
we e p esen .
The maximumHop alue o all agen s was cons an ly se o 5 hops. The imeou s
employed in he espec i e agen classes we e se o he ollowing alues:
1. ma chingTimeou (NC): 40 icks
2. findRegis yTimeou (SP, ISP, SC): 10 icks
3. execu ionTimeou (SP, ISP, SC): 10 icks
4. publica ionTimeou (SP, ISP): 10 icks
5. findNCTimeou (ISP): 10 icks
6. nego ia ionTimeou (ISP): 50 icks
7. disco e yTimeou (SC): 30 icks
8. joinNego ia ionTimeou (SC): 10 icks
9. wai Fo Execu ionTimoue (SC): 5 icks
10. auc ionTimeou (SP): 60 icks
11. biddingTimeou (SP wi h EA, DA, AO o MAO): 30 icks
Th oughou he simula ion expe imen s, wo SLOs we e used, he p ice and he al-
loca ed memo y, ep esen ing a quali a i e se ice aspec . Mos nego ia ion p o ocols
employed a e only single-a ibu i e, hus he p ice would be sufficien . The mul i-
a ibu i e AO p o ocol is he only one also using he second SLO, which was chosen
because i exhibi s he con a y p e e ence di ec ion han p ice: whe eas SCs wan o
achie e a low p ice hey will po en ially y o nego ia e a highe memo y amoun hey
can use o hei in oca ions ( he p e e ence o SPs is exac ly ice e sa). This aspec
is eflec ed in he heHighe TheBe e pa ame e s se o hese cha ac e is ics a bo h
SPs / ISPs and SCs. An offe is conside ed ejec able by a nego ia o agen whene e
he alue o a espec i e SLO (p ice o alloca ed memo y) is smalle han 30% o he
lowe o highe han 170% o he uppe ese a ion alue.
The s a ing alues a e 10 USD o p ice and 20 GB o alloca edMemo y. Due o he
lea ning algo i hm he p ice alue will adap o e ime, as al eady shown. The ma ke -
ValueAdap ionRa e was se o 5% (meaning ha he alua ion will be in-/dec eased by
114
4.3. Simula i e Demons a ion o he P o o ype’s Effec i eness
5% o hei cu en alue). The concession s eps o bo h SLOs we e defined o be 1
USD / GB (SP) and 4 USD / GB (SC), espec i ely. This eflec s he po en ially highe
u gency on he SC and hus mo e hesi an concessions on he SP side.
All jus men ioned pa ame e s a e fixed h oughou all expe imen s. Gi en he ocus
on adap abili y o new p o ocols, such s a egic aspec s a e o mino in e es , once
a easonable configu a ion has ound. The pa ame e s, which a e changed om one
expe imen o he o he , a e now shown along wi h he achie ed esul s.
O e all, six simula ion expe imen s ha e been conduc ed. Each was configu ed wi h 50
SCs, 25 SPs / ISPs and 1 RA. Fo each simula ion un he h ee demons a ion me ics
in oduced abo e ha e been logged and a e gi en as (ba -)g aphs in he ollowing.
Only AO: Du ing he fi s expe imen all 25 SPs offe ed hei se ices o e he AO
p o ocol ske ched in subsec ion 4.2.219.
AO and EA: In a second s ep only 15 SPs used he AO p o ocol and he o he 10 used
he EA.
AO, EA and DA: Du ing he hi d expe imen 9 SPs offe ing he AO and ei he 8
offe ing he EA and DA we e p esen .
AO, EA, DA and CM: Fo he ou h un, one addi ional agen has been in oduced,
a b oke (NC ole). This agen in place, a new configu a ion o 8 SPs offe ing he AO,
ei he 6 offe ing he EA and DA and an addi ional 5 ISPs (using he b oke ed CM) was
se .
AO,EA,DA,CMandFPA: In he nex s ep, 5 SPs o ei he AO, EA, DA, CM (ISP)
and FPA we e p esen .
AO,EA,DA,CM,FPAandMAO: Du ing he final expe imen 5 ISPs and 4 SPs o
each o he o he p o ocols (AO, MAO, EA, DA and FPA) ha e been configu ed, again
summing up o 25 jus as wi h all o he expe imen al se ings.
The ac ual alues o he esul me ics a e o seconda y significance. Among o he s,
he specifics o he diffe en p o ocols ( o example he usage o bidding ounds wi h
p e-defined du a ions in con as o con inuous offe exchanges), he andom selec ion o
SI documen s o s a a nego ia ion wi h, o he dynamic ma ke alua ions de e mine
how many nego ia ions o a gi en ype a e ac ually finished.
The main s a emen o be p o en wi h hese expe imen s is ha he GN node is a all
able o adap o diffe en p o ocols, only based on hei desc ip ion in he EST documen s
(p oo -o -concep ). Based on he esul s achie ed, his asse ion can be app o ed. Wi h
he desc ibed escala ion o diffe en nego ia ion p o ocol ypes p esen h oughou he
expe imen s, each o he newly in oduced p o ocols has success ully been in eg a ed in
19Th oughou all simula ion expe imen s he six p o ocols in oduced in subsec ion 4.2.2 ha e been used.
115

4. Assessmen o he De eloped Sys em
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x 10 4
0
0.5
1
1.5
2
2.5
3x 10 4
# Fixedp ice Auc ions
# Al e na e O e s
# Mul i-a ibu e Al e na e O e s
# English Auc ions
# Du ch Auc ions
# Double Auc ions
(a) Amoun o success ul Nego ia ions
pe P o ocol
FPA AO MAO EA DA DoubleAuc ion
0
100
200
300
400
500
600
Max #
Min #
(b) Min / Max Amoun o success ul Ne-
go ia ions pe P o ocol
0 50 100 150 200 250 300 350 400 450 500
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Min # o Used P o ocol Types
Max # o Used P o ocol Types
(c) Range o used P o ocol Types
Figu e 4.12.: Resul s o using he ollowing P o ocols: AO
he ac ual ma ke beha io ; each p o ocol ype p esen has been execu ed by a significan
numbe o agen s.
Gi en he ela i ely small se ing, i is no su p ising, ha no all expe imen s esul ed
in e e y agen ha ing been able o win a nego ia ion in mo e han one ype o p o ocol
(again also affec ed by he andom selec ion o SIs and hus nego ia ion p o ocols o
each se ice eques ). Du ing he las expe imen (in es iga ing a se ing wi h all six
diffe en p o ocols p esen ) howe e , each agen could success ully ake pa in a leas
wo nego ia ion p o ocol ypes (see figu e 4.17(c)).
On he o he hand, in each simula ion un he e exis SC agen s ha ha e success ully
aken pa in all a ailable p o ocols. This ac is shown in figu es 4.12(c), 4.13(c),
4.14(c), 4.15(c), 4.16(c) and 4.17(c) espec i ely, whe e he maximum amoun o diffe en
p o ocol ypes a single agen has been able o ake pa in is always he amoun o
116
4.4. E alua ion based on Meffe ’s Theo y on Flexibili y
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x 10 4
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
# Fixedp ice Auc ions
# Al e na e O e s
# Mul i-a ibu e Al e na e O e s
# English Auc ions
# Du ch Auc ions
# Double Auc ions
(a) Amoun o success ul Nego ia ions
pe P o ocol
FPA AO MAO EA DA DoubleAuc ion
0
50
100
150
200
250
Max #
Min #
(b) Min / Max Amoun o success ul Ne-
go ia ions pe P o ocol
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
0.5
1
1.5
2
2.5
3
Min # o Used P o ocol Types
Max # o Used P o ocol Types
(c) Range o used P o ocol Types
Figu e 4.13.: Resul s o using he ollowing P o ocols: AO and EA
p o ocols a all p esen in he ma ke .
Ha ing now shown he adap abili y o he GN module o he in oduced nego ia ion
p o ocols (and espec i e desc ip ion documen s), he equi emen s s a ed o his hesis
can be conside ed ulfilled and he demons a ion s ep hus comple ed. The subsequen
e alua ion s ep is desc ibed in he nex sec ion.
4.4. E alua ion based on Meffe ’s Theo y on Flexibili y
Following he a ionale o He ne e al. (2004) and Peffe s e al. (2008) (see sec ion 1.3
o a de ailed discussion), he e alua ion o a scien ific a i ac comp ises i s assessmen
wi h ega d o a se o quali y aspec s.
To his end, He ne e al. do no dis inguish be ween a demons a ion and an e al-
117
4. Assessmen o he De eloped Sys em
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x 10 4
0
500
1000
1500
2000
2500
3000
3500
4000
4500
# Fixedp ice Auc ions
# Al e na e O e s
# Mul i-a ibu e Al e na e O e s
# English Auc ions
# Du ch Auc ions
# Double Auc ions
(a) Amoun o success ul Nego ia ions
pe P o ocol
FPA AO MAO EA DA DoubleAuc ion
0
20
40
60
80
100
120
140
160
Max #
Min #
(b) Min / Max Amoun o success ul Ne-
go ia ions pe P o ocol
0 500 1000 1500 2000 2500 3000 3500 4000
0
0.5
1
1.5
2
2.5
3
3.5
4
Min # o Used P o ocol Types
Max # o Used P o ocol Types
(c) Range o used P o ocol Types
Figu e 4.14.: Resul s o using he ollowing P o ocols: AO, EA and DA
ua ion s ep; hey basically claim he need o a s uc u ed assessmen o he designed
solu ion in e ms o i s effec i eness o sol ing he esea ch p oblem, he quali y o he
solu ion and he u ili y gain achie able by using he espec i e a i ac in he en isioned
p oblem se ing. Peffe s e al., on he o he hand, dis inguish be ween a basic demon-
s a ion phase, du ing which he effec i eness o he designed solu ion and an ac ual
e alua ion phase du ing which he quali y o his solu ion is assessed.
In his hesis, I mainly build on he a ionale o Peffe s e al., as i p o ides a e y
s uc u ed assessmen amewo k. Howe e , i also esul s in some inaccu acies when
ac ually execu ed: The au ho s claim ha demons a ing an a i ac simply means o
“use [ he a i ac ] in expe imen a ion, simula ion, a case s udy, p oo , o o he app o-
p ia e ac i i y” (Peffe s e al. 2008, p. 90). All assessmen s on he basis o measu able
quali y me ics as well as compa isons wi h he s a ed equi emen s a e assumed o be-
118
4.4. E alua ion based on Meffe ’s Theo y on Flexibili y
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x 10 4
0
100
200
300
400
500
600
700
800
900
# Fixedp ice Auc ions
# Al e na e O e s
# Mul i-a ibu e Al e na e O e s
# English Auc ions
# Du ch Auc ions
# Double Auc ions
(a) Amoun o success ul Nego ia ions
pe P o ocol
FPA AO MAO EA DA DoubleAuc ion
0
10
20
30
40
50
60
70
80
Max #
Min #
(b) Min / Max Amoun o success ul Ne-
go ia ions pe P o ocol
0 500 1000 1500 2000 2500 3000 3500 4000
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
Min # o Used P o ocol Types
Max # o Used P o ocol Types
(c) Range o used P o ocol Types
Figu e 4.15.: Resul s o using he ollowing P o ocols: AO, EA, DA and CM (double
auc ion)
long o he e alua ion phase. In my opinion his con adic s wi h hei ini ial s a emen
o e alua ion being he means o assess he quali y o a designed solu ion, a e ha ing
shown ha i sol es he s a ed p oblem a all (effec i eness) in he demons a ion s ep.
Th oughou his hesis, his asse ion has been used as a concep ual basis o dis in-
guishing which assessmen s eps a e shown in which sec ion (demons a ion o e alu-
a ion): Du ing he las sec ion he capabili y o my in as uc u e o sol e he s a ed
equi emen s has been demons a ed. In addi ion o using he p o o ype sys em in a
simula ed en i onmen , some quali y aspec s ha e also been assessed as a compa ison
was made wi h he s a ed equi emen s. All hese s eps aimed a p o ing he effec i e-
ness o he BabelNEG app oach. In he ollowing I will consequen ly elabo a e on “how
well [i ] suppo s a solu ion o he p oblem” (Peffe s e al. 2008, p. 92).
119
4. Assessmen o he De eloped Sys em
al eady be known o he eques ing SC), he communica ion o e head implied by my
sys em can be assumed ma ginal when compa ed o he d as ically inc eased ac i i y
flexibili y.
In his chap e he BabelNEG sys em (a p o o ype implemen a ion o he mechanisms
p esen ed in chap e 3) has been assessed wi h ega d o i s effec i eness and efficiency.
Fi s , i s Ja a-based implemen a ion in he con ex o he SimIS oolki has been shown.
A e ha he capabili y o he de eloped desc ip ion documen s o cap u e all ele an
se ice aspec s in a machine p ocessable way has been demons a ed. In o de o p o ide
asse ions on whe he BabelNEG is able o ulfill all s a ed equi emen s a comp ehensi e
concep ual and simula i e demons a ion has been conduc ed, he esul s o which ha e
been shown in sec ions 4.2 and 4.3. The chap e closes wi h an in es iga ion o he
efficiency o he p oposed solu ion, based on he limi a ions o he de eloped desc ip ion
documen s and he communica ion o e head in oduced by BabelNEG.
Summa izing he esul s, he de eloped p o o ype has been able o ulfill all equi e-
men s s a ed in sec ion 2.1. The quali y o he p oposed solu ion is also conside ably
high, as shown in he ligh o Meffe ’s flexibili y heo y. In he emaining chap e I
p o ide a summa y o he esul s achie ed and some ema ks on u u e wo k and ways
o implemen he de eloped ideas in o a comme cial so wa e p oduc .
126

5. Lessons Lea ned and Fu u e S eps
5.1. Summa y and high-le el In e p e a ion o Resul s
Du ing my disse a ion p ojec I de eloped a no el se ice in as uc u e o he s uc-
u ed disco e y and p o ocol-gene ic nego ia ion o elec onic SLA documen s.
The need o such an a i ac has been deduced om a de ailed scena io analysis,
ex apola ing pas and cu en de elopmen s in dis ibu ed compu ing. This s ep has
esul ed in a concep ual model o he u u e IoS, ac ing as he p oblem domain o
he emainde o my wo k. Based on economic heo y, he need o bo h nego ia ion
p ocesses as such and he possibili y o adap o diffe en p o ocols a un ime has been
in e ed.
A e ha ing mo i a ed my wo k ha way, he esea ch p ocess applied h oughou
his p ojec has been defined. A DS me hod has been chosen, due o he esea ch
ques ion a hand. In ha , I ollowed he me hodological guidelines p oposed by He ne
e al. (2004), Peffe s e al. (2008) as well as G ego and Jones (2007); hem being he
mos influen ial wo ks on DS esea ch me hodologies as o oday.
Following hese guidelines, he equi emen s o my p o o ype, being deduced om he
scena io model and unde lying heo ies, ha e been iden ified. These claims a e u he
used as c i e ia o he assessmen o my sys em.
I ha e subsequen ly discussed concep ual ounda ions o my wo k, comp ising he-
o e ical p inciples o he design, o mula ion, disco e y, nego ia ion and subsequen
usage o (elec onic) SLAs, especially ocusing on dis ibu ed in as uc u es o elec-
onic se ices. Rela ed esea ch effo s, a ge ing he esea ch p oblem, as s a ed o
my disse a ion, ha e been iden ified nex and subsequen ly desc ibed wi h special ocus
on hei po en ial o sol e he s a ed p oblem. This has been assessed by compa ing
each indi idual p ojec wi h he iden ified equi emen s.
In chap e 3, he ac ual in as uc u e design, hus he p oposed solu ion o he s a ed
esea ch p oblem, has been p esen ed. The unde lying idea is o decouple he good o
be sold ( he SLA) om he nego ia ion p o ocol, hus enabling a SP o apply diffe en
nego ia ion p o ocols o he same se ice o e ime (by simply c ea ing new combi-
na ions o SLA and p o ocol). On he o he hand, o he SC side a p o ocol-gene ic
nego ia ion componen has been designed, capable o adap ing o diffe en p o ocols, as
offe ed by he SPs. The concep ual copula be ween hese wo sides a e a se o s uc u ed
se ice desc ip ion documen s, defining no only he se ice- ele an unc ional and non-
unc ional pa ame e s, bu also he applied nego ia ion p o ocol in a machine- eadable
way.
Th oughou he las chap e , a ho ough assessmen o he in as uc u e design has
been done, including he ac ual implemen a ion o he de eloped mechanisms and da a
127
5. Lessons Lea ned and Fu u e S eps
s uc u es in a Ja a-based simula ion en i onmen , he concep ual demons a ion o he
sys em’s effec i eness (wi h ega d o he s a ed equi emen s) as well as a simula i e
demons a ion s ep p o ing he adap abili y o he SC agen s.
Based on hese assessmen s eps he ollowing esul s can be s a ed:
•The exp essi eness o he designed da a s uc u es could be shown on he basis o
a se o ep esen a i e nego ia ion p o ocols o igina ing in scien ific li e a u e.
•Du ing he simula ion uns he de eloped se ice agen s ha e been able o adap
o o me ly unknown nego ia ion p o ocols, jus by pa sing hese p o ocol desc ip-
ions.
Hence, he de eloped sys em is able o ulfill he s a ed equi emen s, hus p o ing i s
effec i eness o sol e he s a ed esea ch p oblem.
5.2. C i ical Reflec ion on he Applied Resea ch Me hod
As wi h e e y esea ch p ojec , my disse a ion has se e al poin s which could be c i -
icized me hodically. The fi s is p obably he ac ha I se ou o sol e a scien ific
p oblem no ac ually p esen cu en ly; he scena io an icipa ed is no ye exis en . This
ine i ably leads o unce ain ies ega ding he deduced esea ch p oblem and ul ima ely
he mo i a ion o my wo k in gene al.
Ne e heless, many inno a i e a i ac s ace his same p oblem: hey an icipa e u u e
scena ios owa ds which hey a e gea ed and aim a sol ing he p oblems occu ing
he ein. The designe s can only y o, as accu a ely as possible, make a p edic ion on
how he u u e will p obably look like and base hei wo k on his scena io. I ollowed his
exac p ocedu e. Wha I hope makes my p edic ion alid, is i s concep ual g ounding in
ongoing ends wi hin he dis ibu ed compu ing discipline, as can be wi nessed in bo h
academia and indus y oday.
Howe e , he e is s ill a chance ha he u u e In e ne will no look exac ly like he
IoS se ing ske ched in sec ion 1.1.1. In ha case, hope ully a leas some aspec s o my
scena io model will be ue (which is qui e easonable o assume), po en ially leading o
only mino losses in he u ili y gained by my in as uc u e o consequen ly mino needs
o changes o i .
The second poin , which could be c i icized is he assessmen o he de eloped a i-
ac . The demons a ion o he effec i eness o he de eloped desc ip ion language and
espec i e documen s canno claim comple eness. I was based on a se o diffe en
p o ocols ha we e chosen due o a mu ually as high as possible dissimila i y. Fo a
comple e demons a ion his assessmen would ha e o be done o all possible nego i-
a ion p o ocols used oday. Un o una ely, he e is no such exhaus i e lis , as al eady
s a ed. The concep ual demons a ion as p esen now could po en ially inc ease i s claim
on comple eness by ga he ing nego ia ion p o ocols om all cu en and u u e se ice
in as uc u es (o igina ing in indus y and esea ch) and in es iga e he language’s ex-
p essi eness wi h hose. In my wo k, a easonable amoun o ( esea ch) p ojec s and
128
5.3. F om P o o ype o P oduc
p o ocols p oposed in scien ific li e a u e has been used; paying ibu e o he ime
cons ain s a disse a ion p ojec poses. A mo e comp ehensi e e alua ion will ha e o
ake place o e he nex yea s, in which he de eloped in as uc u e is hope ully used
ex ensi ely.
On he o he hand, simula ion as a ool o he demons a ion o a p oo -o -concep
p o o ype is a guable. The unde lying model and simula ion ools always only “simula e”
some eal-wo ld se ing wi hou a p oo ha hey do i co ec ly. This is enough o
a p oo o easibili y, as aimed a wi h his hesis. Mo e sophis ica ed asse ions on
echnical quali y me ics, such as scalabili y o obus ness, demand an ac ual, physically
dis ibu ed p o o ype sys em. Only hen, a eliable assessmen o he echnical quali y
o he de eloped in as uc u e can be made.
Finally, my disse a ion p ojec suffe s om i being as inno a i e as i is. No o he
in as uc u e sol es he exac same p oblem s a ed o my wo k, hus esul ing in a
lack o al e na i es he BabelNEG sys em could be di ec ly compa ed o. As o oday
he one p omising candida e o such a di ec compe i o p ojec is SLA@SOI1,whichis
cu en ly wo k in p og ess. Once finished i would be in e es ing o compa e i wi h my
in as uc u e, especially wi h ega d o he exp essi eness o he p o ocol desc ip ions
and hus he esul ing ac i i y flexibili y achie ed by bo h.
5.3. F om P o o ype o P oduc
On an o ganiza ional le el, a necessa y p econdi ion o an applica ion o my ideas is
o a espec i e company’s IS o ollow he SO pa adigm. Only when s ic ly building
on loosely coupled, fine-g anula IT se ices, buying, nego ia ing abou , combining and
e-selling such compu a ional unc ionali ies is easonable.
Addi ionally, he economic s a egies applied in he ma ke beha io o he manage-
men agen s mus be defined. A de ailed specifica ion o concession s ep alues, ese e
p ices e c. is needed o each agen and po en ially e en each se ice o be sold / pu -
chased.
Wi h ega d o he binding o au oma ically nego ia ed “con ac s”, a ju is ical ame-
wo k (e.g. in e ms o a amewo k con ac go e ning a conso ium o in ol ed business
pa ne s) is needed as a common g ound o all ac o s on he elec onic ma ke . Gi en
such a ounda ion is p esen , he nego ia ion agen s can be ega ded as ep esen a i es o
hei hos o ganiza ions, enabling hem o bindingly ac in he esul ing se ice economy.
The ocabula y used in he SLAs and se ice desc ip ion documen s mus be un-
de s andable o all ma ke pa icipan s o an au oma ed economy o wo k. Thus, a
on ology o se ice concep s (such as quali y me ics o unc ional desc ip ions) mus
be defined, as al eady s a ed in he las chap e s. I should co e all ma ke - ele an
in o ma ion concep s along wi h he espec i e seman ics o such concep s.
On a echnical le el, some undamen al in as uc u e componen s a e needed wi hin
he company IS o success ully in eg a e wi h my sys em: A g aphical use in e ace
( o eques ing o selling se ices on he ma ke manually) o an equi alen ly ex ended
1h p://sla-a -soi.eu/
129
5. Lessons Lea ned and Fu u e S eps
o ches a ion engine ( o doing so au oma ically) a e needed as links be ween he adi-
ional se ice in as uc u es and he BabelNEG sys em.
Also, a dis ibu ed in as uc u e o ma ke b oke s and disco e y nodes is needed o
assu e a s able se ice economy. These nodes could ei he be main ained by he SC and
SP companies o by hi d pa y in e media ies. Simila ly, a echnical secu i y sys em is
indispensable, especially o ansac ions in a business con ex .
Gi en hese p econdi ions a e me wi hin he pa icipa ing companies and gene al
ma ke in as uc u e, he nego ia ion in as uc u e desc ibed in chap e 3, and he
esul ing p o o ypical implemen a ion as has been assessed in chap e 4, can be di ec ly
in eg a ed wi h he cu en ly used company ISs. Fo his, he defined managemen agen s
and espec i e s a egy modules mus be implemen ed and g an ed access o bo h he
in e nal se ice ins ances ( o gua an ee en o cemen and moni o ing pu poses) and he
ex e nal communica ion channels (including secu i y and disco e y sys em).
Fo a comp ehensi e se ice managemen in as uc u e, so wa e componen s sup-
po ing he o he phases o he li e cycle (apa om disco e y and nego ia ion) a e o
cou se also needed.
Gi en hese condi ions a e me he BabelNEG sys em can p o ide significan economic
ad an ages o bo h indi idual companies and he digi al se ice economy as a whole.
I d as ically educes en y ba ie s o he IoS, especially om he SC’s poin o iew,
as espec i e en e p ises can easily implemen SC agen s ins an ly capable o in e ac ing
wi h all possible SPs p esen in he se ice economy. The e is no need o implemen
diffe en nego ia o agen s o diffe en se ice ma ke s (and hus nego ia ion p o ocols).
In some cases his would be oo cos ly o inefficien , eques ing om he SC company o
choose he p o ocols espec i e agen s a e implemen ed o . Gi en he high dynamic o
he IoS se ing, he decision abou such a sel - es ic ion o a subse o he a ailable SPs
as po en ial ansac ion pa ne s is no only a highly complex ask, bu could also likely
ende i sel w ong o e ime. Wi h he BabelNEG sys em in place, no such decision
has o be made and SCs can flexibly adap o new SPs and nego ia ion p o ocols.
This is especially in e es ing o small and medium sized companies ac ing on niche
ma ke s. They gene ally do no ha e he unds o cons an ly e-e alua e hei decisions
on po en ial ma ke s o engage in and o e-design hei nego ia ion componen s. Ne -
e heless hey can delibe a ely p ofi om an IoS scena io as enabled by he BabelNEG
in as uc u e, as i allows hem o pu chase needed se ices (e en on e y sho no ice)
and concen a e he majo i y o hei in-house in es men s on hei co e business ins ead.
The BabelNEG sys em can also lead o comple ely new se ice p oduc s o be c ea ed
e y quickly by combining indi idual basic se ices pu chased o e he IoS. No isky
in es men sonin-houseITin as uc u esha e obemade o suchanewoffe ing,bu
he p o iding company can “ y ou ” he ma ke success a fi s wi h ex e nal se ices
be o e deciding o in e nally p o iding hem in case o success.
Gi en he BabelNEG sys em is in place, SPs can also easily adjus he employed
nego ia ion p o ocols o changes in he ma ke configu a ion o hei offe ed se ices.
Once in oduced and desc ibed wi hin a espec i e EST documen , hey can be su e o
assume SCs able o in e p e i and engage in espec i e nego ia ions subsequen ly.
On he economy’s poin o iew, he BabelNEG sys em enables a se o new business
130
5.4. Fu u e Wo k
models o eme ge wi hin he IoS, such as ma ke make s, us ed hi d pa ies o e en
p o ide s o bes p ac ice nego ia ion s a egies on demand. Gi en he much highe
amoun o compa ible agen s in a IoS se ing based on BabelNEG, he e is a much highe
possibili y o exploi ing economies o scale om such businesses. This is especially he
case in a digi al se ing, since a espec i e se ice, once implemen ed, p oduces almos
no inc emen al cos s in case he amoun o use s inc eases, bu con a ily a e y high
amoun o ma ginal gains.
5.4. Fu u e Wo k
Gi en he dynamic IoS se ing, an e e changing applica ion scena io can be o eseen o
he BabelNEG in as uc u e. In his las sec ion, I will sho ly ske ch wha adap ions
/ ex ensions o he cu en sys em design could be in oduced in o de o inc ease i s
o e all adap abili y o such new applica ion se ings.
The mos undamen al dynamics, which can be an icipa ed o he u u e IoS, a e a
changing se o se ices and se ice me ics; especially in case o an ongoing end o
au oma e indi idual business unc ionali ies and he eme gence o ju is ical ounda ions
o c oss-o ganiza ional au oma ed se ice ansac ions. This de elopmen can al eady
be cap u ed in he cu en sys em a chi ec u e, as i only affec s he se ice ocabula y
used wi hin he nego ia ions. The ac ual agen oles, messages and nego ia ion p o ocols
in gene al do no diffe jus because he nego ia ed me ics change.
Simila ly, op imiza ion o nego ia ion s a egies wi h ega d o a) indi idual nego-
ia ion p o ocols o e en b) when in oducing s a egies op imized o a whole se o
diffe en p o ocols (which would be a o ed by a p o ocol-gene ic app oach as p esen ed
in his hesis) does no affec he sys em a chi ec u e and communica ion mechanisms
designed he ein.
I mo e sophis ica ed nego ia ion p o ocols eme ge, an ex ension o he desc ip ion
documen s p oposed in his hesis could become necessa y. A po en ial adjus men , in-
c easing hei exp essi eness, could be he in oduc ion o sub-s a es wi hin he nego ia-
ion p ocess. This would esul in an al e ed p ocess-elemen wi hin he EST documen ;
no only he incoming e en will de e mine he possible ac ions, bu a he a uple o in-
coming e en and cu en s a e. This esul s in a di ec equi alence o he s a e machine
pa adigm.
Apa om hese dynamics o he unde lying scena io, some undamen al ex ensions
could be in oduced, inc easing he gene al efficiency o BabelNEG:
The fi s po en ial ex ension would be he in oduc ion o a dedica ed In o ma ion
Se ice ole, esponsible o dis ibu ing nego ia ion- ele an da a2. Which da a can
be accessed can al eady be desc ibed wi hin he EST documen and in he cu en
in as uc u e p oposal he NC is assumed o dis ibu e such in o ma ion. Howe e , a
comp ehensi e suppo o in o ma ion dis ibu ion, po en ially allowing o p o ocols,
2In he cu en p o o ype, his aspec was conside ed ou o scope, as he p ima y goal was o in es iga e
he adap abili y o he SCs o new p o ocols.
131

5. Lessons Lea ned and Fu u e S eps
in which he nego ia ing agen s ha e o ac i ely eques some in o ma ion, on e.g. he
cu en highes bid du ing hei nego ia ion beha io , is s ill o be designed.
Also, a publish / subsc ibe sys em could be ano he in e es ing ex ension o my sys em.
I could build on he al eady used que y da a s uc u e (used o que ying SIs, ESTs o
STs om he RA a he momen ) and would allow a SC agen o pos he need o a
se ice o a gi en ype (and / o offe ing a gi en p o ocol) whene e no such se ice could
be ound igh away. To his end, i would define he sea ch c i e ia, c ea e a espec i e
que y objec and pos his in e es o a RA. Whene e a SI fi ing he s a ed que y is
published a he RA, a no ifica ion message is sen o he espec i e SC, indica ing ha
an ins ance o he equi ed se ice is finally a ailable on he ma ke .
In o de o assu e imeliness o he s o ed in e es s a lease-based mechanism could be
employed. Each in e es , egis e ed wi h he RA, mus be enewed a e ce ain ime
in e als. Whene e a lease is no enewed in ime he espec i e in e es is dele ed om
he publish / subsc ibe in as uc u e.
A mino ex ension o he cu en sys em could also allow o he in eg a ion o e e se
nego ia ions (and hus e en ually o p o ocol-gene i y, o he need o such, a he SP
side). Ins ead o publishing SI documen s, ep esen ing a se ice offe ed by a SP, he
SC agen s could offe an ex ended SI documen ep esen ing a demand o a se ice, he
espec i e SC cu en ly needs. This demand documen would exhibi he same in e nal
s uc u e as an SI. Howe e , no link o a WSDL documen would be p esen , as he
ac ual se ice implemen a ion ulfilling his demand (i any o such is e e ound), is no
known a publica ion ime. Addi ionally, he EST o be applied o his demand is only
op ional. I exis en i deno es ha he se ice o be ound mus no only fi he s a ed
ST bu also adhe e o he s a ed EST (i.e. a dis inc p o ocol is eques ed), i no , only
he ype o he se ice is specified as a c i e ia in he demand.
Especially he disco e y a chi ec u e is e y basic in he cu en e sion o he sys-
em. This is due o he al eady men ioned p agma ic easons. Fu u e e sions should
inco po a e a mo e sophis ica ed and hus mo e obus and scalable disco e y mecha-
nism han a single egis y node. The mos p omising app oach o his is p obably a
P2P-based a chi ec u e, gi en such sys ems excel in scalabili y and esilience o node
ailu es (especially s uc u ed ones).
One idea o in eg a e he cu en p o o ype wi h a P2P-based disco e y a chi ec u e is
o in oduce a DHT which connec s all egis y nodes and in which all se ice desc ip ion
documen s a e s o ed. Whene e a se ice eques is submi ed o a egis y node
(node wi hin he DHT ing), he espec i e documen s a e e ie ed and e u ned o he
eques o .
A new RA can simply join his sys em by disco e ing a DHT node al eady p esen
(b oadcas disco e y) and hen join he DHT ing as a neighbo o his node. In e nally,
his esul s in he e-dis ibu ion o he s o ed da a and ou ing in o ma ion. In con as ,
he absence o a RA (which has le he DHT) is di ec ly no iced by i s neighbo , which
hen igge s he e-a angemen o he DHT da a acco dingly ( his ep esen s a s anda d
mechanism o DHTs).
All o hese ex ensions could help o inc ease he efficiency and o e all unc ionali y
o he de eloped sys em. Ne e heless, he p o o ype as cu en ly a ailable al eady
132
5.4. Fu u e Wo k
demons a es he easibili y o au oma ed p o ocol-gene i y in SLA nego ia ions, which
was he p ima y goal o his hesis.
133
A. Appendices
A.1. Se ice Desc ip ion Schema Documen s
A.1.1. Se ice Type Documen
1<?xml e sion=”1.0” encoding=”UTF−8”?>
2<?xml e sion=”1.0” encoding=”UTF−8”?>
3<xsd:schema xmlns:xsd=”h p://www.w3. o g/2001/XMLSchema”
4 a ge Namespace=”h p://xml.SLANeg. o g/schema/SLANeg”
5xmlns=”h p://xml.SLANeg. o g/schema/SLANeg”
6elemen Fo mDe aul =”quali ied”>
7
8<xsd:complexType name=”P ope iesType”>
9<xsd:sequence>
10 <!−− alue domain o he p ope y −−>
11 <xsd:elemen name=”domain” ype=”xsd:S ing” />
12
13 <!−− desc ip ion o he me ic ’s seman ics −−>
14 <xsd:elemen name=”decla a ion” ype=”xsd:S ing”
minOccu s=”0” maxOccu s=”1” />
15 </xsd:sequence>
16
17 <!−− me ic iden i ie as uniquely de ined in an indus y−
wide on ology −−>
18 <xsd:a ibu e name=”p ope yID” ype=”xsd:anyURI” />
19 </xsd:complexType>
20
21 <!−− de ines he unc ionali y o a class o se ices −−>
22 <xsd:elemen name=”Se iceType”>
23 <xsd:complexType>
24 <xsd:sequence>
25 <!−− quali ied name o he se ice ype −−>
26 <xsd:elemen name=”se iceTypeID” ype=”xsd:anyURI
”/>
27
28 <!−− UDDI−like desc ip ion o he se ice ype (
URL o i n −line desc ip ion) −−>
29 <xsd:elemen name=”se iceDesc ip ion” ype=”
xsd:anyType” />
30
31 <!−− non− unc ional se ice p ope ies −−>
32 <xsd:elemen name=”p ope y” ype=”P ope iesType”
maxOccu s=”unbounded” />
33 </xsd:sequence>
34 </xsd:complexType>
35 </xsd:elemen >
135