A CBR App oach o Alloca e
Compu a ional Resou ces Wi hin a Cloud
Pla o m
Fe nando De la P ie a, Ja ie Bajo and Juan M. Co chado
Abs ac Cloud Compu ing pa adigm con inues g owing e y quickly. The
unde lying compu a ional in as uc u e has o cope wi h his inc ease on he
demand and he high numbe o end-use s. To do so, pla o ms usually use
ma hema ical models o alloca e he compu a ional esou ce among he o e ed
se ices o he end-use . Al hough hese ma hema ical models a e alid and hey a e
widely ex ended, hey can be imp o ed by means o use in elligen echniques.
Thus, his s udy p oposes an inno a i e app oach based on an agen -based sys em
ha in eg a ed a case-based easoning sys em. This sys em is able o dynamically
alloca e esou ces o e a Cloud Compu ing pla o m.
1 In oduc ion
The echnology indus y and he scien ific communi y ha e aken g ea s ides in
ecen yea s owa d implemen ing he Cloud Compu ing (CC) echnological pa a-
digm. This has esul ed in a apid g ow h o bo h p i a e and public pla o ms [12,
17,25,28] aimed o p o ide inno a i e solu ions ha can esol e he cu en needs
o he CC pa adigm.
The ma ke ing model used in he CC pa adigm is inno a i e, as i is based on a
pay-as-you-go concep [2], in which use s mus nego ia e and p e iously es ablish a
Se ice Le el Ag eemen (SLA) in o de o access se ices [1]. Once his con ac
F. De la P ie a (✉)⋅J.M. Co chado
Depa men o Compu e Science and Au oma ion Con ol,
Uni e si y o Salamanca, Plaza de la Me ced s/n, 37008 Salamanca, Spain
e-mail: [email p o ec ed]
J. Bajo
Depa men o A ificial In elligence, Technical, Uni e si y o Mad id,
Bloque 2, Despacho 2101, Campus Mon egancedo,
Boadilla del Mon e, Mad id 28660, Spain
e-mail: jbajo@fi.upm.es
© Sp inge In e na ional Publishing Swi ze land 2016
P. No ais e al. (eds.), In elligen Dis ibu ed Compu ing IX,
S udies in Compu a ional In elligence 616
DOI 10.1007/978-3-319-25017-5_7
75
o compu ing goods has been es ablished, bo h he use s ( h ough egula paymen s)
and he CC sys em (by main aining he se ice) a e obliga ed o ollow h ough wi h
hei ag eemen . In his ega d, no el y is de e mined by he inno a i e spec um o
unde lying echnology ( i ualiza ion, se ice a ms, web se ices, e c.), which ha e
ecen ly eached he poin o allowing he se ices o be o e ed wi h he same le el
o quali y, ega dless o exis ing use demand [16,26,31]. These new possibili ies a
a echnological le el lead o he bi h o a new concep , elas ici y [9], which is based
on he jus -in- ime p oduc ion me hod [13].
Exis ing esea ch in he s a e o he a is based on me hods ha use cen alized
algo i hms based on ma hema ical and heu is ic models [15,19,30], nei he o
which can ensu e he e ficiency o he sys em, o e en i s a ailabili y, in he e en o
a sys em ailu e.
Gi en hese sho comings, i is necessa y o s udy new echniques ha allow o
he e olu ion o exis ing models wi h ega d o elas ici y o se ices. This s udy
p oposes he use o models de i ed om A ificial In elligence (AI), since on an
in e nal poin o iew, a CC is cha ac e ized by i s massi e dis ibu ion, he e o-
genei y, and high le el o unce ain y, which is p ecisely whe e he applica ion AI
holds g ea po en ial. The inclusion o p oac i e, sel -adap a ion and lea ning
capabili ies, among o he s, is key o he e olu ion o hese elas ic managemen
algo i hms o compu a ional esou ces. As a esul , agen s and mul iagen sys ems
[29] (MAS) we e selec ed among all he a ailable AI echniques because o hei
dis ibu ed na u e and abili y o wo k in en i onmen s such as CC sys ems, whose
cha ac e is ics would clea ly iden i y hem as open sys ems.
Using his MAS-based app oach, he amewo k o his s udy p oposes a
dynamic and sel -adap ing model o he dis ibu ion o compu a ional esou ces in
a CC en i onmen . This model is based on he lea ning capabili ies p o ided by a
case-based easoning (CBR) [10] sys em, an app oach which has no p e iously
been used in his ype o dis ibu ed en i onmen . These easoning sys ems de elop
a easoning model simila o ha o humans, using pas expe iences o sol e a
specific p oblem.
This wo k is o ganized as ollows: he ollowing sec ion p o ides a desc ip ion
o he con ex o and ela ed app oaches, Sec . 3p oposes a solu ion based on
mul iagen sys ems, while he e alua ion and alida ion o hese sys ems a e p e-
sen ed in Sec . 4. Finally, he las sec ion p esen s he conclusions o he esea ch.
2 Resou ce Alloca ion in Cloud Compu ing Pla o ms
In a CC en i onmen , he ha dwa e in as uc u e is i ualized [7,8], which means
ha he e is an abs ac ion laye be ween he eal ha dwa e in as uc u e and he
compu ing nodes. Each o he se ices is ac ually deployed in he compu ing nodes
o his abs ac ion laye ( e e ed o as i ual machines). In u n, he se ices a e
gene ally dis ibu ed among a ious compu a ional nodes, which is why hei needs
76 F. De la P ie a e al.
o be a wo k balance sys em ha can dis ibu e he eques s among he a ious
compu a ional nodes a ending he se ices.
The use o i ualiza ion g ea ly simplifies he managemen o compu a ional
esou ces a he in as uc u e le el, making is possible o dynamically c ea e o
elimina e i ual machines on demand o e en mig a e a i ual machine om one
physical se e o ano he in execu ion ime, wi hou needing o s op o pause he
machine. The e o e, and gi en he capabili ies o e ed by i ualiza ion echnology,
he p oblem, while complex, is ac ually simple in i sel , since i is only based on he
e ficien edis ibu ion o physical ( eal) esou ces among he di e en compu a-
ional ( i ual) nodes.
In cu en li e a u e, he dis ibu ion o esou ces is iewed om wo poin s o
iew [5]:
•QoS-awa e based, o ma ke o ien ed [5]. This fi s g oup is associa ed wi h a
clien -o ien ed dis ibu ion o esou ces model which a emp s o minimize
compu a ional isks in o de o dis ibu e he compu a ional esou ces acco ding
o he SLA eached, and ollowing he pay-pe -use economic model. Acco ding
o his model, he managemen echniques o he compu a ional esou ces aim
o adhe e o hese ag eemen s a all ime, hus p o iding he quali y o se ice
ha was eques ed and consequen ly expec ed by he end use . The s a e o he
a includes s udies in line wi h his app oach by means o ma hema ical models
[18,23,27].
•Ene gy-awa e based [5]. In his second app oach, he dis ibu ion o esou ces
akes place by aking in o accoun bo h he p e-es ablished SLA and he ene gy
consump ion, which assumes compliance wi h bo h. The e a e ewe s udies in
he s a e o he a wi h his app oach as compa ed o he fi s , al hough hey a e
mo e no el. This includes a a ie y o echniques a e also based on ma he-
ma ical models [3,15,19].
In ligh o hese s udies in he cu en s a e o he a , i is necessa y o p opose a
model o he dis ibu ion o compu a ional esou ces which would ake ene gy
consump ion in o accoun . The aim is o educe he ene gy consump ion equi ed o
sa is y he SLA ha ha e been es ablished wi h he pla o m use s. The p esen
s udy ollows a comple ely di e en app oach based on op imiza ion echniques and
AI, which allows o he dis ibu ion o esou ces by ollowing a dis ibu ed and
scalable model, hus allowing he sys em o lea n o e an ex ended pe iod o ime.
As no ed abo e, he CC compu a ional pa adigm has g own s ongly in ecen
yea s; i s de elopmen has led o he ad ancemen o a la ge numbe o pla o ms,
bo h public and p i a e. A MAS amewo k based on VO has been selec ed o deal
wi h hese obs acles. Al hough one may ini ially conside hese wo dis ibu ed
sys ems (MAS and CC) o be incompa ible, a de ailed analysis demons a es ha
hey a e in ac no only complemen a y, bu sha e conside able syne gy be ween
hem. Fi s o all, CC en i onmen s can co e he compu a ional needs o pe -
sis ence o in o ma ion and he compu ing po en ial ha MAS equi e o di e en
applica ions such as da a mining, managemen o complex se ices, e c. Addi-
ionally, MAS can be used o c ea e a much mo e e ficien , scalable and adap able
A CBR App oach o Alloca e Compu a ional Resou ces …77
design o he CC en i onmen han wha is cu en ly a ailable. Finally, he use o
MAS in he amewo k o he design o CC sys ems p o ides his pa adigm wi h
new cha ac e is ics such as lea ning o in elligence, which makes i possible o
de elop much mo e ad anced compu a ional en i onmen s in all aspec s (in elligen
se ices, in e ope abili y among pla o ms, e ficien dis ibu ion o esou ces, e c.).
The numbe o s udies ha can be ound on he s a e o he a ela ing CC wi h
agen echnology is ac ually qui e low. Howe e , his endency is changing and i is
becoming inc easingly common o find s udies and applica ions ocused on his
field. Despi e he limi ed numbe o s udies on he ma e , Agen -based Cloud
compu ing, o he Agen -based Cloud pla o m, is becoming a common concep ,
men ioned by a ious au ho s in ecen yea s [4,6,14,20–22,24].
3 P oposed A chi ec u e
Taking in o accoun he needs and sho comings de ec ed in he e iew o he s a e
o he a , his s udy p oposes a new model o alloca ing esou ces based on a CBR
app oach and guided by a mul iagen a chi ec u e especially designed o he
managemen o CC en i onmen s. This sec ion will desc ibe he key componen s
ha allow ex ending he ope a ion o he elas ic algo i hms o he dis ibu ion o
esou ces p oposed wi hin his wo k.
To begin, we would like o no e ha since he p oposed MAS is a dis ibu ed
sys em by na u e, each o he agen s ha wo k in he dis ibu ion o esou ces can
be loca ed h oughou he en i e CC en i onmen . Tha is, he CC sys em is
moni o ed and con olled in a dis ibu ed manne . This dis ibu ed moni o ing
model makes i possible o ins an ly adap exis ing esou ces o he CC en i onmen
acco ding o demand o each se ice, which in u n mee s he dual objec i e o
complying wi h he es ablished SLA ag eemen s and educing ene gy consump ion.
Figu e 1p esen s an o e iew o he agen -based a chi ec u e The ollowing
agen s a e di ec ly ela ed o he moni o ing and con ol o he ha dwa e:
•Local Moni o . In cha ge o ga he ing da a ela ed o he s a e o he local
esou ces o each physical se e , including he physical machine as well as he
di e en i ual machine i hos s.
•Local Manage . In cha ge o con olling he compu a ional esou ces o he
physical machine. In o he wo ds, esponsible o ini ia ing o u ning o i ual
machines acco ding o he p e iously configu ed se ice empla es.
•Global Manage . The p ima y agen in cha ge o decision making wi h ega d
o he dis ibu ion o compu a ional esou ces. In o de o pe o m his ask, he
agen uses a CBR-BDI model, which will be explained in de ail in he ollowing
sec ion. As a means o suppo o making decisions ega ding he dis ibu ion
o esou ces, his agen uses a pa ial knowledge base (p o ided by he Local
Moni o ) and he abili y o modi y local esou ces in each machine (p o ided by
he Local Manage ).
78 F. De la P ie a e al.
The edis ibu ion o esou ces a a mac o le el is pe o med by he Global
Manage agen s, which ha e g ea e au ho i y han he Local Manage agen and
can in o m hem o he need o s a up a new machine wi h a specific se ice and
specific cha ac e is ics. A he end o his p ocess, a new i ual machine (VM), wi h
specific cha ac e is ics o Memo y and i ual cpus will be ins an ia ed in o de o
mee he cu en demand.
The Global Manage is a highly specialized agen ha implemen s a CBR-BDI
[10,11] delibe a i e a chi ec u e. As a esul , he easoning p ocess in each physical
node is based on pas expe ience gained om s o ing simila cases. The case
memo y is cen al o he en i e CC sys em; he sys em’s global knowledge can be
sha ed by each o i s membe s, in his case he Global Manage agen . Gi en ha
his memo y can g ow exponen ially as a main enance s a egy, a high-speed
schema-less da abase is used o p o ide as access o he s o ed da a, based on
MongoDB.
1
The Global Manage ini ia es he p ocess by defining he concep o case
C={P,S(P), E} whe e:
•Pco esponds o he p oblem desc ip ion, which has a ma ix-ma ched ep e-
sen a ion associa ed o he ins an ia ion o he use o esou ces.
•S(P) is associa ed wi h he solu ion o he p oblem: S(P)={M, cpu} in e ms o
memo y and cpu.
•The e ficiency (E) is measu ed om wo pe spec i es:
• he deg ee o e ficiency o he p oposed solu ion wi hin he physical se e
whe e he i ual machine has been ins an ia ed. This deg ee o e ficiency is
Global
Supe iso
Physical Resou ce M
. . .
Physical Resou ce 1
Local
Manage
Global
Regula o
Global
Regula o
Local
Resou ce
Moni o
VM 1 VM i
Se ice Demand
Moni o
Se ice
Se ice
Supe iso
. . .
VM j
. . .
Ne wo k
Se ice Demand
Moni o
Se ice
Se ice
Supe iso
. . .
VM 2 VM 1 VM 2
Use
Consume
O ganiza ion
Resou ce
o ganiza ion
Managemen
O ganiza ion
Ha dwa e
Manage
Ha dwa e
Manage
Fig. 1 Agen -based a chi ec u e o a Cloud Compu ing Pla o m
1
h p://www.mongodb.o g.
A CBR App oach o Alloca e Compu a ional Resou ces …79
p oposed by he Local Moni o agen acco ding o he usage a es o he
p ocesso and he alloca ed memo y.
•The deg ee o e ficiency om he poin o iew o he se ice. The deg ee o
e ficiency measu es he numbe o addi ional nodes equi ed by he se ice.
The CBR (Case-Based Reasoning) p ocess is ini ia ed and e ie es simila cases
om he case memo y. The mos simila cases a e selec ed acco ding o he ol-
lowing s eps: (i) Selec he cases om he physical machines wi h simila cha ac-
e is ics; (ii) a ec o is configu ed o each case ha con ains he same numbe o
i ual machines ha a e in he case; and (iii) he cases selec ed om his subse a e
hose ha p e iously used he same se ice ha is now eques ing esou ces, and
du ing a pe iod o ime simila o he cu en case.
A solu ion o he p oblem, which is based on he e ie ed cases, will be p epa ed
du ing he euse phase:
•I he case base does no con ain a p e ious simila case, he solu ion o he
p oblem will be associa ed o he minimum esou ces de e mined a he le el
whe e he se ice is ins an ia ed.
•I , on he o he hand, simila cases a e e ie ed, he solu ion o he p oblem will
be he closes case mul iplied by he case e ficiency:
•I he alues assigned o he p e ious solu ion a e g ea e han he alues
assumed by he machine, due o he ac ha he e a e no many esou ces
a ailable, he esul o he case will be he maximum amoun o esou ces
a ailable in he machine.
Once he solu ion o he case has been calcula ed, he new node will be
ins an ia ed and i s use e alua ed om a mic o and mac o pe spec i e, hus p o-
iding he alue o e ficiency o he solu ion. Finally, du ing he final s a e o he
p oposed CBR cycle, he case and i s co esponding e ficiency will be s o ed o
u u e use.
4 E alua ion
The e alua ion and alida ion o he model o his s udy will be done h ough a CC
pla o m de eloped wi hin he scope o he esea ch ca ied ou by he BISITE
esea ch g oup,
2
and will include di e en compu a ional se ices a he ha dwa e
and so wa e le el. This CC pla o m was deployed in he HPC en i onmen o he
BISITE esea ch g oup and composed o 15 la es gene a ion machines ha suppo
i ualiza ion in he ha dwa e wi h he use o In el-VT echnology and he KVM
i ualiza ion sys em.
Du ing he expe imen , 10 h eads ha que y o specific me hods o he se ice
(Ge Size and Ge Folde Con en ) a e launched e e y h ee seconds, o a maximum
2
h p://bisi e.usal.es.
80 F. De la P ie a e al.
o 40 h eads. The p ocess s a s once he agen -based a chi ec u e de ec s a
dec ease in pe o mance, a which ime i di ec ly execu es he adap a ion p ocess.
The Global Manage agen o each o he physical machines ha hos he se ice
nodes. We should ecall ha he Global Manage agen is a specialized agen ha
uses a CBR-BDI easoning p ocess [10] in cha ge o he dis ibu ion o esou ces a
a mac o le el. Once hey ecei e he ini ial ale , hese agen s esend he ale
message o he emaining Global Manage agen s in he CC sys em.
Each indi idual Global Manage hos ed by each physical machine ca y ou in
pa allel he p ocess desc ibed in he p e ious sec ion. Thus nsolu ions a e p o-
posed. The agen -based a chi ec u e eac i ely selec s he node ha o e s he mos
esou ces a he i ual machine le el.
The esul s in e ms o QoS can be seen in Fig. 2, which also show an inc ease in
he quali y le el a e he adap a ion has been comple ed.
The case s udy was epea ed nume ous imes, which made i possible o s o e a
good numbe o pas expe iences in he case memo y. Howe e , as p esen ed in
Fig. 3when he e a e many cases in he memo y and a numbe o pas expe iences
simila o he cu en p oblem, he adap a ion esul s a e ac ually be e because he
QoS le el is lowe .
0,7
0,6
0,5
0,4
0,3
0,2
0,1
Response ime (sec.)
Elapsed ime (sec.) Elapsed ime (sec.)
0
0,7
0,6
0,5
0,4
0,3
0,2
0,1
Response ime (sec.)
0
Fig. 2 Expe imen 1: eadjus men o he in as uc u e esou ces o me hod (le no adap a ion;
igh adap a ion)
Elapsed ime (sec.) Elapsed ime (sec.)
0,7
0,6
0,5
0,4
0,3
0,2
0,1
Response ime (sec.)
0
0,9
0,8
0,7
0,6
0,5
0,4
0,3
0,2
0,1
0
Response ime (sec.)
Fig. 3 Expe imen 1: eadjus men o he in as uc u e esou ces o me hod (le expe imen 1;
igh in o med adap a ion)
A CBR App oach o Alloca e Compu a ional Resou ces …81
5 Conclusions
This s udy ini ially se o h o be one o he fi s MAS app oaches o all wi hin he
amewo k o con ol and moni o ing sys ems in a CC en i onmen . The s udy
p oposed a new a chi ec u al model based on a MAS wi h a clea ly in eg a i e
cha ac e . A se ies o algo i hms o he dis ibu ion o compu a ional esou ces in a
CC en i onmen we e de eloped, e alua ed and alida ed. I s bigges inno a ion
cen e s on he sys em’s dynamic abili y o au oma ically adap acco ding o demand
and lea n om p e ious expe iences.
This new model has demons a ed ha a con ol and moni o ing sys em in a CC
en i onmen can be designed wi h MAS. The inhe en ly dis ibu ed na u e o MAS
makes i possible o implemen elas ic algo i hms o se ices by ollowing a dis-
ibu ed s a egy. The dis ibu ion o esponsibili ies wi hin he scope o his ype o
algo i hm makes i possible no only o make decisions whe e he p oblems ac ually
a ise, bu o dis ibu e he compu ing capabili y equi ed o each a solu ion among
di e en ins ances o he CC en i onmen .
This app oach also ensu es independence o he decision-making p ocess in
so wa e laye s whe e he a ious ac ions a e execu ed. The e is no doub ha a
change in he capabili ies o e ed by he unde lying echnology will also equi e
changes o be made in he p oposed easoning models, as wi h any app oach wi h a
adi ional design. Gi en he defini ions o oles a a high le el, i he echnology
p oposes new capabili ies, he adap a ion in he p oposed a chi ec u e will consis o
modi ying he indi idual o indi iduals ha pe o m specific asks o ha e a ole
wi hin he MAS.
Finally, This app oach can maximize he deg ee o e ficiency o he p oposed
solu ions wi h ega d o p e ious solu ions, which in u n p og essi ely imp o es
he esponse and he sys em’s capabili y since i is capable o lea ning. Mo eo e ,
his lea ning abili y is impo an in an unce ain en i onmen such as he CC
sys em. I he con ex o en i onmen o he CC pla o m changes a any gi en ime,
he adap a ion model will e ol e in u n, adap ing he p oposed solu ions in o de o
maximize he e ficiency o he gi en solu ion.
Acknowledgmen s This wo k has been suppo ed by he MICINN p ojec
TIN2012-36586-C03-03.
Re e ences
1. Alhamad, M.; Dillon, T.S., Chang, E.: Concep ual SLA amewo k o cloud compu ing
(2010)
2. A mb us , M., e al.: A iew o cloud compu ing. Commun. ACM 53(4), 50–58 (2010)
3. Beloglazo , A. Abawajy, J. Buyya, R.: Ene gy-awa e esou ce alloca ion heu is ics o
e ficien managemen o da a cen e s o cloud compu ing. Fu u e Gene . Compu . Sys . 28(5),
755–768 (2012)
82 F. De la P ie a e al.
4. B aubach, L., Jande , K. Pokah , A.: A middlewa e o managing non- unc ional equi emen s
in cloud PaaS. In: IEEE In e na ional Con e ence on Cloud and Au onomic Compu ing
(ICCAC), pp. 83–92 (2014)
5. Buyya, R., Beloglazo , A., Abawajy, J.: Ene gy-e ficien managemen o da a cen e esou ces
o cloud compu ing: a ision, a chi ec u al elemen s, and open challenges. P ep in a Xi :
1006.0308 (2010)
6. Cao, B.-Q., Li, B. Xia, Q.-M.: A se ice-o ien ed QoS-assu ed and mul i-agen cloud
compu ing a chi ec u e. In: Cloud Compu ing, pp. 644–649. Sp inge , Be lin (2009)
7. Che, J., e al.: A syn he ical pe o mance e alua ion o OpenVZ, Xen and KVM. In: IEEE
2010 Asia-Pacific Se ices Compu ing Con e ence, pp. 587–594. IEEE (2010)
8. Chen, W., e al.: A no el ha dwa e assis ed ull i ualiza ion echnique. In: The 9 h
In e na ional Con e ence o Young Compu e Scien is s, pp. 1292–1297. IEEE (2008)
9. Chiu, D.: Elas ici y in he cloud. C oss oads 16(3), 3–4 (2010)
10. Co chado, J.M., e al.: Replanning mechanism o delibe a i e agen s in dynamic changing
en i onmen s. Compu . In ell. 24(2), 77–107 (2008)
11. Co chado, J.M., Laza, R.: Cons uc ing delibe a i e agen s wi h case‐based easoning
echnology. In . J. In ell. Sys . 18(12), 1227–1241 (2003)
12. Fishe , P., Pan , R., Edbe g, J.: Cloud Compu ing: Assessing Azu e, Amazon EC2, Google
App Engine and Hadoop o i Decision Making and De elope Ca ee G ow h. Ap ess, New
Yo k (2010)
13. Hu chins, D.:Jus in ime. Gowe Publishing L d., London (1999)
14. Kang, J. Sim, K.M.: Cloudle: an on ology-enhanced cloud se ice sea ch engine. In: Web
In o ma ion Sys ems Enginee ing–WISE 2010 Wo kshops. Sp inge , Be lin, Heidelbe g,
pp. 416–427 (2011)
15. Kusic, D., e al.: Powe and pe o mance managemen o i ualized compu ing en i onmen s
ia lookahead con ol. Clus e Compu . 12(1), 1–15 (2009)
16. Liu, F., e al.: NIST Cloud Compu ing Re e ence A chi ec u e, ol. 500, p. 292. NIST Special
Publica ion (2011)
17. Luo, J.-Z., e al.: Cloud compu ing: a chi ec u e and key echnologies. J. China Ins . Commun.
32(7), 3–21 (2011)
18. an Nguyen H., Dang T an, F. Menaud, J.-M.: Au onomic i ual esou ce managemen o
se ice hos ing pla o ms. In: P oceedings o he 2009 ICSE Wo kshop on So wa e
Enginee ing Challenges o Cloud Compu ing, pp. 1–8. IEEE Compu e Socie y (2009)
19. Ragha end a, R., e al.: No powe s uggles: coo dina ed mul i-le el powe managemen o
he da a cen e . In: ACM SIGARCH Compu e A chi ec u e News, pp. 48–59. ACM, (2008)
20. Sim, K.M.: Agen -based cloud compu ing. IEEE T ans. Se . Compu . 5(4), 564–577 (2012)
21. Talia, D.: Cloud compu ing and so wa e agen s: owa ds cloud in elligen se ices. In: WOA
2011, pp. 2–6
22. Talia, D.: Clouds mee agen s: owa d in elligen cloud se ices. IEEE In e ne Compu . 16(2),
78–81 (2012)
23. Van Hien N., T an, F.D., Menaud, J.-M.: SLA-awa e i ual esou ce managemen o cloud
in as uc u es. In: Nin h IEEE In e na ional Con e ence on Compu e and In o ma ion
Technology, CIT’09, pp. 357–362. IEEE (2009)
24. Ven icinque, S., e al.: A cloud agency o SLA nego ia ion and managemen . In:Eu o-Pa
2010 Pa allel P ocessing Wo kshops, pp. 587–594. Sp inge , Be lin, Heidelbe g (2011)
25. Von Laszewski, G., e al.: Compa ison o mul iple cloud amewo ks. In: 2012 IEEE 5 h
In e na ional Con e ence on Cloud Compu ing (CLOUD), pp. 734–741. IEEE (2012)
26. Wang, L., e al.: Cloud compu ing: a pe spec i e s udy. New Gene . Compu . 28(2), 137–146
(2010)
27. Wei, G., e al.: A game- heo e ic me hod o ai esou ce alloca ion o cloud compu ing
se ices. J. Supe compu . 54(2), 252–269 (2010)
28. Wen, X., e al.: Compa ison o open-sou ce cloud managemen pla o ms: OpenS ack and
OpenNebula. In: 2012 9 h In e na ional Con e ence on Fuzzy Sys ems and Knowledge
Disco e y (FSKD), pp. 2457–2461. IEEE (2012)
A CBR App oach o Alloca e Compu a ional Resou ces …83
