Managing he E olu ion o an En e p ise A chi ec u e using
a
MAS-P oduc -Line App oach
Joaquin Pe ia Michael G. Hinchey Manuel Resinas
Uni e si y o Se ille NASA Godda d Space Fligh Cen e Uni e si y o Se ille
Spain USA Spain
[email p o ec ed]
[email p o ec ed]
[email p o ec ed]
Roy S e i James
L.
Rash
Uni e si y o Uls e NASA Godda d Space Fligh Cen e
No he n I eland USA
.s eni @uls e .ac.uk James.L.Rash @nasa.go
Abs ac
We iew an e olu iona y sys em ns being n so wa e
p oduc line. The co e a chi ec u e is he ~inchnngingpn o
he sys em, and each e sion o he sys em may be iewed as
a p oduc om he p oduc line. Each "p oduc
"
may be de-
sc ibed as he co e a chi ec u e wi h son e agen -based nd-
di ions. The esul is a mul iagen sys em so wa e p oduc
line. We desc ibe an app oach o such n So wa e P oduc
Line-based app oach using he MaCMAS Agen -O ien ed
nze hoclology. The app oach scales o en e p ise n chi ec-
u es as a mul iagen sys em is an app op in e means o
ep esen ing a changing en e p ise n chi ec cl e nnd he in-
e ac ion be ween componen s in i .
Keywo ds: Mul iagen Sys ems P oduc Lines, En e -
p ise a chi ec u e e olu ion.
1
In oduc ion and Mo i a ion
is ixed (i.e., he subs an ial pa o he sys em ha does no
change), and each e sion o he e ol ing sys em may be
iewed as a pa icula p oduc om he p oduc line.
Simila ly, an en e p ise a chi ec u e may be iewed as
he co e a chi ec u e ha is unchanging, and a ious spe-
cializa ions o he a chi ec u e (as he
en e p ise
e ol es)
implemen a ious p oduc s o he p oduc line.
I we conside he unchanging pa o a so wa e sys em
o o an en e p ise o be he co e a chi ec u e, he special-
iza ion o a ious p oduc s ( e sions o he sys em) can be
iewed as agen -based addi ions. The esul is ha an e ol -
ing sys em can be iewed as a So wa e P oduc Line
o
mul iagen sys ems (MAS).
Ou app oach scales o en e p ise a chi ec u es and so -
wa e a chi ec u e o wo easons. Fi s ly, a mul iagen sys-
em (MAS) is a e y app op ia e means o ep esen ing an
en e p ise and he in e ac ions wi hin i , hanks o he o -
ganiza ional me apho ha a chi ec s he sys em mimicking
he eal en e p ise o ganiza ion. Secondly, he gap be ween
he en e p ise a chi ec u e and he so wa e a chi ec u e is
mi iga ed h ough he addi ion o a chi ec u al concep s a
When
dealing
wi h
sys ems.
and
in
pa icu- he unning Tha is o say, MAS a e
la 'ys ems exhibi ing any
o
au onomy
O
au onomic
able o
manage
a chi ec u al e olu ions and
p ope ies, i is un ealis ic o assume ha he sys em will be
ec u al
concep s
a
he
implemen a ion
le el,
s a ic. Complex sys ems e ol e o e ime, and he a chi ec- In his pape , we p opose a se o modeling echniques
u e o an e ol ing sys em will change e en a un ime, as based on an agen -o ien ed me hodology called
Me hodol-
he sys em implemen s sel -con igu a ion, sel -adap a ion,
ogy o Analyzing Complex Mul iagen Sys ems
(MaCMAS)
and mee s he challenges o i s en i onmen . ha is designed o deal wi h complex unp edic able sys ems
An e ol ing sys em can be iewed as mul lple e sions [ll]'. Speci ically, he app oach we use is based on an ex-
o he same sys em. Tha is, as he sys em e ol es i es- ension o MaCMAS ha allows o model MAS P oduc
sen ially ep esen s mul iple ins ances o he same sys em, Lines (MAS-PL)
[14,
131. This allows us o manage he
each wi h i s own a ia ions and sueci ic changes. Tha is
..
o say, an e ol ing sys em may be iewed as a p oduc line
'See
www. dg-se ille.in olmembe sljoaquinpl nacmasl
o
de ails
and
o sys ems, whe e he co e a chi ec u e o he p oduc line
case
s udies
using his
me hodology
modeling o he e olu ion o he sys em in a sys ema ic way.
To he bes o ou knowledge. his is he i s app oach
ha deals wi h a chi ec u al changes o MASS based on
MAS-PL.
2
Backg ound and Rela ed Wo k
The so wa e p oduc line pa adigm (he ea e , SPL) au-
gu s he po en ial o de eloping a co e a chi ec u e om
which cus omized p oduc s can be apidly gene a ed, e-
ducing ime- o-ma ke , cos s, e c.
[I],
while simul aneously
imp o ing quall y, by making g ea e e o in design, im-
plemen a ion and es mo e inancially iable, as his e o
can be amo ized o e se e al p oduc s. The easibili y o
building MAS p oduc lines is p esen ed in
[13].
In [14],
we discuss he de ails o how o build he co e a chi ec u e.
In a MAS-PL, we can obse e he en e p ise a chi ec u e
o he sys em om wo di e en poin s o iew. This dis-
inc ion s ems om he o ganiza ional me apho 19.
10.
181.
These wo iews a e he ollowing:
Acquain ance poin o iew:
shows he o ganiza ~on as
he se o in e ac ion ela ionships be ween he oles
played by agen s in models called ole models. I o-
cuses on he in e ac ions wi hin he sys em and also on
ep esen ing how a unc ionali y designa ed by a sys-
em goal is achie ed.
S uc u al poin o iew:
shows agen s as a i ac s ha
belong o sub-o ganiza ions, g oups, eams. In his
iew agen s a e s uc u ed in o hie a chical cons uc-
ions showing he social s uc u e o he sys em. I
shows which agen s a e playing he oles in he Ac-
quain ance O ganiza ion, and hus, shows how sys em
goals a e achie ed by means o agen s in e ac ing o
ul ill he sys em goals.
As shown in
[7],
he acquain ance o ganiza ion can be
modeled o hogonally o i s s uc u al o ganiza ion. This
allows us o change he sys em goals ha a e enabled in he
sys em by changing he pa s o he acquain ance o ganiza-
ion p esen in he s uc u al o ganiza ion. This in ac . is
he basis o MAS-PLs.
The so wa e p ocess o MAS-PLs is di ided in wo main
s ages: Domain Enginee ing and Applica ion Enginee ing.
The o me is esponsible o p o iding he eusable co e as-
se s ha a e exploi ed du ing applica ion enginee ing when
assembling o cus omizing indi idual applica ~ons [4]. En-
e ing in o de ails, we migh say ha , gene ally, bo h s ages
can be u he di ided in o equi emen s, analysis. design.
and implemen a ion (a ypical so wa e de elopmen li ecy-
cle).
domain equi emen s:
This phase desc ibes he equi e-
men s o he comple e amily o p oduc s. highligh ing
bo h he common and a iable ea u es ac oss he am-
ily. In his phase, commonal~ y analysis is o g ea im-
po ance o aiding in de e mining which a e he com-
monal~ ies and a iabili ies. The models used in his
phase o speci ying ea u es show when a ea u e is
op ional, manda o y o al e na i e in he amily. The
models used a e called ea u e models
[2,
131. A ea-
u e is a cha ac e is ic o he sys em ha is obse able
by he end use . which in essence ep esen s he same
concep han a sys em goal as shown p e iously
[6].
domain analysis:
This phase p oduces a chi ec u e-
independen models. i.e. acquain ance o ganiza ion
models, ha de ine he ea u es o he amily and he
domain o applica ion. MAS-PLs use ole models
o ep esen he in e aces and in e ac ions needed o
co e ce ain unc ionali y independen ly (a ea u e
o a se o ea u es)[l4]. The mos ep esen a i e
e e ences in he non-MAS-PL ield a e
[5,
171.
Simila app oaches ha e appea ed also in he
00
ield. o example
[3.
161,
bu all o hese app oaches
use ole models wi h he same pu pose, namely.
ep esen ing ea u es o he sys em in isola ion om
he inal en e p ise a chi ec u e.
domain design:
In his phase, a co e a chi ec u e o he
amily is p oduced. and is e med he co e s uc u al
o ganiza ion o he sys em. The co e a chi ec u e is
o med as a composi ion o he ole models co e-
sponding o he mo e s able ea u es in he sys em
[13].
applica ion enginee ing:
This phase has he esponsibili y
o building conc e e p oduc s. As ou pu pose in his
pape is he un ime e olu ion o he sys em, we do no
illus a e i he e.
3
A
NASA
case s udy
The case s udy we use is a swam o pico-spacec a s
ha a e used o p ospec he as e oid bel . The en e p ise
a chi ec u e o he sys em changes a un- ime depending
on he en i onmen and he s a e o he swa m. F om all he
possible e olu ions we show only wo s a es o he sys em.
in he i s one he swa m is o bi ing an as e oid in o de o
analyze i ; in he second, a sola s o m occu s in he en i-
onmen and he sys em changes i s s a e o p o ec i sel .
We
will show he ole models o bo h s a es and an ex-
ample o composi ion o bo h o hem since bo h ea u es o
he sys em a e no comple ely o hogonal: o p o ec om
a sola s o m he spacec a mus ake wo basic s eps: (a)
o ien ~ s sola sails o minimize he a ea exposed o he so-
la s o m pa icles ( im sails) and (b) powe -o all pos-
sible elec onic componen s. S ep (a) minimizes he o ces
om impinging sola -s o m pa icles, which could a ec he
spacec a 's o bi . Bo h s eps (a) and (b) minimize po en ial
damage om he cha ged pa icles in he s o m (which can
deg ade senso s, de ec o s, elec onic ci cui s, and sola en-
e gy collec o s).
4
Modeling an E olu iona y MASS
As we ha e shown. each p oduc in a MAS-PL is de ined
as a se o ea u es. Gi en ha all he p oduc s p esen a
se o ea u es ha emain unchanged, he co e a chi ec u e
is de ined as he pa o all o he p oduc s ha implemen
hese common ea u es[l4]. Thus, a sys em can e ol e by
changing. o e ol ~ng, he se o non-co e ea u es.
A p oduc o a s a e in ou e olu iona y sys em can be
de ined as a se o ea u es. Le
F
=
{ i..
,)
be he se
o all ea u es o a MAS-PL. Le
cF
c
F
be he se o
co e ea u es and
ncF
=
F
CF
be he se o non-co e
ea u es. We de ine a alid s a e o he sys em as he se o
co e ea u es and a se o non-co e ea u es, ha is o say,
S
=
cF
U
SF,
whe e
SF
c
ncF
is a subse o non-co e
ea u es.
Gi en ha . he e olu ion om one s a e o ano he
S,
is de ined as:
whe e
nF,
,-I
c
ncF
1s he se o new ea u es and
dF,
,-I
C
7x9
is he se o dele ed ea u es.
Finally,
Ai,,-1
desc ibes he a ia ion be ween he p od-
uc o he s a e
i
-
1
and he p oduc o he s a e
i,
ha is o
say,
nF,,,-1
dF,,,-1.
In
[13],
we show ha a ea u e co ela es wi h a ole
model. Thus, o a sys em o e ol e om one s a e o an-
o he , we mus compose o decompose he ole models in
nF
and
dF.
Speci ically, we mus compose he ole models
co esponding o he ea u es in
nF
wi h he ole models
co esponding o he ea u es ha emain unchanged om
he ini ial s a e
STPI,
ha is o say
S,
dF,.;-l.
Decompo-
si ion is used o ole models ha mus be elimina ed.
In he ollowing subsec ions, we desc ibe ole models,
and he ope a ions o composi ion and decomposi ion.
5
Models
MaCMAS is he AOSE me hodology ha we use o ou
app oach. I is specially ailo ed o model complex acquain-
ance o ganiza ions
[15].
We use his me hodology since i
is he only ha p o ides explici suppo o MAS-PLs.
Fo he pu poses o his pape , we only need o know
a ew ea u es o MaCMAS, mainly some o he models i
uses. Al hough a p ocess o building hese models is also
needed, we do no add ess his in his pape , and e e he
A)
Plan
Model
--
B)
Role Model
Figu e
2.
Sel -p o ec ion om sola s o ms
au onomic p ope y model
in e es ed eade o he li e a u e on his me hodology. F om
he models i p o ides, we a e in e es ed in he ollowing:
a) S a ic Acquain ance O ganiza ion View:
This shows
he s a ic in e ac ion ela ionships be ween oles in he
sys em and he knowledge p ocessed by hem. In his
ca ego y, we can ind models o ep esen ing he on-
ology managed by agen s, models o ep esen ing
hei dependencies, and ole models. Fo he pu poses
o his pape we only need o de ail ole models:
Role Models:
show
an
acquain ance sub-o ganiza ion
as a se o oles collabo a ing by means o se e al
n~~l i-Role
In e ac ion
(mRI). mRIs a e used o
abs ac he acquain ance ela ionships amongs
oles in he sys em. As mRIs allow abs ac
ep esen a ion o in e ac ions, we can use hese
models a wha e e le el o abs ac ion we de-
si e.
In Flgu e
I-B
and
2-B,
we show he ole model
ha ep esen s how he swa m o bi s an as e oid
and he one ep esen ing he p o ec ion om a
sola s o ms. In he igu es, in e aces, ep e-
sen ed as boxes, ep esen he s a ic ea u es o
oles showing hei goals, he knowledge man-
aged. and he se ices p o ided. mRIs, ep e-
sen ed as dashed ellipses, ep esen he in e ac-
ions be ween he oles linked o hem, show-
ing hei goal when collabo a ing, hei pa e n
o collabo a ion, and he knowledge consumed,
used, and ob ained om he collabo a ion.
b)
Beha io o Acquain ance O ganiza ion View:
The
beha io al aspec o an o ganiza ion shows he se-
quencing o mRIs in a pa icula ole model. I is
A)
Plan Model
<dn , onmen >>
O bi e
ds" eP0a
Poo
Oib Mcddlw
--
RdeG&
Ca iulaLe
ahb
o bilM O ,W mRlG *ls
s d
06,
mod&%
OC1,0+#<.
~m~-,l,Om~
~~IBB~JPOB POB
aQe aORelaldb3
PO$
asDa a
As cela
*slModei
mode(
1
n
Calaula n bqiisle adDa a,
OlbllMcdei
.
me.
as nbu e
Rum:
Collaboiai an
Ou :
..
-
-
.
-__--
B)
Role Model
Figu e
1.
O bi ing and measu ing an as e oid au onomous p ope y
ep esen ed by wo equi alen models:
Plan o
a
ole:
sepa a ely ep esen s he plan o each
ole in a ole model showing how he mRIs o
he ole sequence. I is ep esen ed using UML
2.0
P o ocolS a eMachines. I is used o ocus on
a ce ain ole, while igno ing o he s.
Plan o a ole model:
ep esen s he o de o mRIs in
a ole model wi h a cen alized desc ip ion. I is
ep esen ed using UML
2.0
S a eMachines. I is
used o acili a e easy unde s anding o he whole
beha io o a sub-o ganiza ion.
In Figu e
I-A
and
2-A,
we show he plan o he
ole models o ou example.
We mus add a new model o MaCMAS in o de o ep-
esen he e olu ions o he sys em. This model is called he
e olu ion plan.
E olu ion Plan:
Is ep esen ed using a UML s a e machine
whe e each s a e ep esen s a p oduc , and each an-
si ion ep esen s he addi ion o elimina ion o a se o
ea u es, ha is o say,
A.
In addi ion, he condi ions in
he ansi ions ep esen he p ope ies ha mus hold
in he en i onmen and in he sys em in o de o e ol e
o he new p oduc .
A)
Plan
Model
B)
Role
Model
Figu e
3.
Measu e s o ms model
In
Figu e
4,
we show pa o he e olu ion plan o ou
case s udy. The e we ep esen wo p oduc s, one ep-
esen ing he swa m when o bi ing an as e oid, and an-
o he ep esen ing he swa m when o bi ing and p o-
ec ing om a sola s o m.
As
can be seen, we add o
dele e he ea u e co esponding o p o ec om sola
s o m depending on whe he o no he
swa m
is unde
isk o sola s o m, which is measu ed by he ea u e
ep esen ed in he ole model o Figu e
3.
Figu e
5.
Composed Role Model
J
[STMeasu e Sola S o mRisk()c=K]
s~~~w.~h~~~~~a~~s~wm~
{
F a~ ~hm.sdu.s ms)
Figu e
4.
E olu ion plan o ou case s udy
6 E ol ing om one p oduc o ano he
6.1
Composing ole models
I is impo an o poin ou ha he composi ion o ole
models is used o map an acquain ance o ganiza ion on o
a se o agen s; ha is o say,
a
s uc u al o ganiza ion.
This mapping is no always o hogonal be ween all ole
models-applying wo ela ed ea u es o a p oduc may e-
qui e hei in eg a ion. The composi ion o ole model is he
p ocess equi ed o pe o m his in eg a ion. In he case o
ha ing o hogonal ea u es, and hus o hogonal ole mod-
els, we mus only assign he p esc ibed oles o he co e-
sponding agen s.
We ha e o ake in o accoun ha when composing se -
e al ole models ha a e no independen , we can ind:
eme gen oles and nRls,
a i ac s ha appea in he compo-
si ion ye hey do no belong o any o he ini ial ole mod-
els;
composed oles and mRls,
he oles and mRIs in he
esul an models ha ep esen se e al ini ial oles o mRIs
as a single elemen ; and,
unchanged oles and nRls,
hose
ha a e le unchanged and impo ed di ec ly om he ini-
ial ole models.
Once hose ole models o be used o he co e a chi ec-
u e ha e been de e mined, we mus comple e he co e a -
chi ec u e by composing ole models. In addi ion, o ob ain
a ce ain p oduc we pe o m he same p ocess. Impo ing
an mR1 o a ole equi es only i s addi ion o he composi e
ole model. The ollowing shows how o compose oles and
plans.
When se e al oles a e me ged in a composi e ole
model, hei elemen s mus be also me ged as ollows:
Goal o he ole:
The new goal o he ole is a new goal
ha abs ac s all he ole goals o he ole o be composed.
This in o ma ion can be ound in equi emen s hie a chical
goal diag ams o we can add i as he
and
(conjunc ion) o
he goals o be composed. In addi ion. he ole goal o
each nRI can be ob ained om he goal o he ini ial oles
o ha mRI.
Ca dinali y o he ole:
I is he same as in he ini ial
ole o he co esponding mR1.
Ini ia o s) ole(s):
I mRI composi ion is no pe -
o med, as in ou case, his ea u e does no change.
In e ace o a ole:
All elemen s in he in e aces o
oles o be me ged mus be added o he composi e in e -
ace. No ice ha he e may be common se ices and knowl-
edge in hese in e aces. When his happens, hey mus be
included only once in he composi e in e ace, o enamed,

depending on he composi ion o hen on ologies.
Gua d
o
a
ole/ nRI:
The new gua ds a e he
and
(con-
junc ion) o he co esponding gua ds in ini ial ole models
i oles composed pa icipa e in he same mRI. O he wise,
gua ds emain unchanged.
In ou case s udy, he e olu ion om he p oduc
o bi -
ing,
ha also ha e he ea u e
measu e s o ms,
o he p od-
uc
p o ec ing onz sola s o m
equi es he addi ion o he
ea u e o p o ec om a sola s o m. This is due o wo
easons: i s , he ea u es
o bi ing and measu e as e oid
and he
measu e s o ms
belongs o he co e a chi ec u e,
and second. he
p o ec ion om sola s o ms
can happen
in whiche e momen and we mus epo he las measu es
o he as e oid be o e powe ing-o subsys ems. Thus, as
hese ole models a e no o hogonal, we mus pe o m a
composi ion o hem. This composi ion, ep esen ed in Fig-
u e
5,
IS
done ollowing he ule p esc ibed abo e. As can
be obse ed, we ha e impo ed a11 he mRIs and mos oles.
In addi ion, we ha e pe o med a composi ion o oles
Sel -
P o ecSC
and he es in he ole model
O bi and measu e
as e oicls.
The conlposi ion o plans consis s o se ing he o de o
execu ion o mRls in he composi e model, using he ole
model plan o ole plans. We p o ide se e al algo i hms o
assis in his ask: ex ac ion o a ole plan om he ole
model plan and lce e sa, and agg ega ion o se e al ole
plans; see
[
1
?]
o u he de ails o hese algo i hms.
Thanks o hese algo i hms, we can keep bo h plan iews
consis en au oma ~cally. Depending on he numbe o oles
ha ha e o be me ged we can base he composi ion o he
plan o he composi e ole model on he plan o oles o on
he plan o he ole model. Se e al ypes o plan composi-
ion can be used o ole plans and o ole model plans:
Sequen ial:
The plan is execu ed a omically in sequence
wi h o he s. The inal s a e o each s a e machine is supe -
imposed wi h he ini ial s a e o he s a e machine ha ep e-
sen s he plan ha mus be execu ed, excep he ini ial plan
ha main ains he ini ial s a e unchanged and he inal plan
ha main ains he inal s a e unchanged.
In e lea ing:
To in e lea e se e al plans, we mus build
a new s a e machine whe e all mRIs in all plans a e aken
in o accoun . No ice ha we mus usually p ese e he o de
o execu ion o each plan o be comp-s-d. We can use al-
go i hms o check beha io inhe i ance o ensu e ha his
cons ain is p ese ed, since o ensu e :his p ope y, he
composed plan mus inhe i om all he ini ial plans
[S].
The composi ion o ole model plans has o be pe o med
ollowing one o he plan composi ion echniques desc ibed
p e iously. La e , i we a e in e es ed in he plan o one o
he composed oles. as i is needed o assign he new plan o
he composed oles; we can ex ac i using he algo i hms
men ioned p e iously.
We can also pe om a composi ion o ole plans ollow-
Figu e
6.
Composed
plan
ing one o he echniques o compose plans desc ibed p e i-
ously. La e , i we a e in e es ed in he plan o he composi e
ole model, o example o es ing. we can ob ain i using
he algo i hms men ioned p e iously.
In Figu e
6.
we show he composed plan o ou case
s udy. This plan ollows an in e lea ing composi ion whe e
we include he mR1
epo nzeasu es
be o e s a ing he p o-
ec ion om he sola s o m. No ice ha when inishing
he sola s o m, he sys em will e ol e o he o he p oduc
dele ing he ea u e
sola s o m p o ec ion.
Then, he plan
o he ea u e
o bi ing and measu e
will s a om i s ini ial
s a e, hus e-s a ing he explo a ion o he as e oid.
6.2
Decomposi ion o ole models
The decomposi ion is simple han composi ion. When
he ole model o be elimina ed
IS
o hogonal o he es , we
only ha e o dele e he co esponding oles om he agen s
ha a e play~ng hem. In he case ha he ole model is de-
penden wi h o he s, we ha e o dele e he elemen s o ole
models and elimina e all he in e ac ions ha e e o hem.
Gi en ha he so wa e a chi ec u e whe e he sys em un
should suppo he ole concep and i s changes a un ime,
hese changes can be made easily wi h a lowe impac on
he sys em.
Howe e , when dependen , se e al ea u es may appea
whose ole models a e ela ed. In hese cases some oles
may ha e o be decomposed. These oles a e hose whose
mRls belong o he scope o he ole model(s) ha ha e o
be elimina ed. In hese cases, he ole has o be decomposed
in o se e al oles, in o de o isola e he pa o he ole we
wan o dele e.
In addi ion, we ha e o elimina e he mRl(s) o he ole
model(s) o be elimina ed om he ole model plan o he
ole plans. This is done s a ing om he plan o he ini ial
dependen ole models. Each sepa a e ole model usually
main ains he o de o execu ion o mRIs de e mined in he
ini ial model bu execu ing only a subse o mRIs o he
ini ial ole models. The beha io o he ole model o be
dele ed can
be
ex ac ed au oma ically using he algo i hms [4]
M.
Ha su A su ey on domain enginee ing. Technical Re-
descnbed In
[12].
This algo i hm allows us o ex ac he po 3 1. Ins i u e o So wa e Sys ems, Tampe e Uni e si y o
plan o emaining ole models om he ini ial ones con- Technology, Decembe 2002.
s ain~ng his o he se o mRIs ha emains in he model. [5] A. Jansen. R. Smedinga. J. Gu p, and
J.
Bosch. Fi s class
ea u e abs ac ions o p oduc de i a ion.
IEE P oc. ~d ny.~
7
Conclusions and u u e wo k
We ha e desc ibed a no el app oach o desc ibing. un-
de s anding, and analyzing e ol ing sys ems.
Ou
app oach
is based on iewing di e en ins ances o a sys em as i
e ol es as di e en "p oduc s" in a So wa e P oduc Line.
Tha So wa e P oduc Line is in u n de eloped wi h an
agen -o ien ed so wa e enginee ing app oach and iews he
sys em as a Mul iagen Sys em P oduc Line. The use o
such an app oach is pa icula ly app op ia e as i allows us
o scale ou iew o add ess en e p ise a chi ec u es whe e
a ious en i ies in he en e p ise a e modeled as so wa e
agen s.
The main ad an age o he app oach esides in he ac
ha i allows us o de i e a o mal model o he sys em and
o each s a e ha i may each. This allows us o clea ly
speci y he di e ences om one s a e o he a chi ec u e
and any subsequen s a es o ha e ol ing sys em. This sig-
ni ican ly imp o es ou capabili ies o unde s and, analyze
and es e ol ing sys ems. Addi ionally. hanks o he use
o MaCMAS which allows o he desc ip ion o he same
ea u e a di e en le els o abs ac ion. we can also spec-
i y and es he a chi ec u al changes a di e en le els o
abs ac ion.
Finally. such an app oach p o ides suppo a un ime
o he addi ion and dele ion o oles in he a chi ec u e.
I p o ides e lec ion mechanisms ha enable unde s and-
ing o he ea u es, oles, and agen s in an en e p ise a -
chi ec u e a di e en le els o
abs ac ion,
p o iding ca-
pabili ies o ensu ing quali y o se ice by means o sel -
o ganiza ion. sel -p o ec ion, sel -healing and o he sel -"
p ope ies iden i ied by he Au onomic Compu ing ini la-
i e. Fu he mo e, i dec eases he dis ance be ween en e -
p ise a chi ec u es and so wa e a chi ec u es. enabling us
o model en e p ise a chi ec u es as so wa e a chi ec u es
and exploi all o he ad an ages o so wa e a chi ec u es
app oaches.
Re e ences
!ll
P.
Clemen s and
L.
No h op.
So wa P oduc L ne, P ac.-
i es and Pa e ns.
SEI Se ies in So wa e Engineenng.
Addison-Wesley. Aug. 2001.
121
K.
Czamecki and U. Eisenecke .
G n -a ~'e P og amming:
M lzods, Tools, and Applica ions.
Addison-Wesley. 2000
[31
D.
D'Souza and A. Wills.
objec ^,
Componen .c.
i zd
F ame-
wo k. wi h UML: T ze Ca alysi.~ App oach.
Addison-Wesley,
Reading, Mass., 1999.
-
So m a e.
15 l(4): 187-1 98. 2004.
161
K.
Kang,
S.
Cohen,
J
Hess, W No ak. and
A.
Pe e -
son Fea u e-o ien ed domain analysis ( oda) easibili y s udy.
Technical Repo CMU SEI-90-TR-021, So wa e Enginee -
ing Ins i u e. Ca negie-Mellon Uni e si y, No embe 1990
171
E.
A. Kendall. Role modeling o agen sys em analysis. de-
sign. and implemen a ion.
IEEE Concu ency.
8(2):34-41,
Ap ./June 2000.
[81
B.
Lisko and
J.
M Wing. Speci ica ions and hel use in
de ining sub ypes. In
P oc,e ding.s
o
he eigh h annual con-
. e e ce on Objec -o ien ud p og a~nming sysiem.c, lanyuag
,,
a d app1ic.a on.s.
pages 16-28. ACM P ess, 1993.
191
J.
Odell,
H
Pa unak. and M. Fleische . The ole o oles in
designinge ec i e agen o ganisa ions. In
A.
Ga cla and
C. L.
F.
Z
A.
0.
J
Cas o, edi o s,
SojYwa e Enginee ug b La ge-
Scale Mul -Agm Sys e n.~.
numbe 2603 in LNCS. pages 27-
28, Be lin, 2003. Sp inge -Ve lag.
[I 01
H.
V.
D Pa unak and
J.
Odell. Rep esen ing social s uc u es
in UML. In
J.
P
Miille ,
E
And e, S. Sen, and
C.
F asson. edi-
o s.
P oceeding. o he Fi l11 In en a ional Con e enc on Au-
ono nous Agen s.
pages 100-101, Mon eal. Canada. 2001
ACM P ess.
[I I]
J.
Peiia.
On I np o ng The Modelling O Complex Acquain-
ance O gan sa ions O Agen s. A Me hod F ag nen Fo The
Ana1ysi. Phase.
PhD hesis. Uni e si y o Se ille. 2005.
[I21
J.
Pe ia,
R.
Co chuelo. and
J.
L. A jona. Towa ds In e ac-
ion P o ocol Ope a ions o La ge Mul i-agen Sys ems. In
P oceed~ngs e he Second ln e na ional Wo kshop on Fo nal
App oaches o Agen -Based Sys ems
(FAABS
2002).
olume
2699 o
LNAI,
pages 79-91. NASA-Godda d Space Fligh
Cen e , G eenbel , MD, USA, 2002. Sp inge -Ve lag.
[13]
J.
Peiia andM.
G.
Hinchey Mul iagen sys em p od~lc lines:
Challenges and bene i s.
Communi a ions
o
he ACM,
De-
cembe 2006. Submi ed and p e-accep ed.
[14] J. Peiia.
M.
G.
Hinchey. and A. Ruiz-Co is. Building
he
co e a chi ec u e o a mul iagen sys em p oduc line: Wi h
an example om a u u e nasa mission. In
7 h In e na ional
Wo kshop on Ayen O ien ed So wa e Enginee ing
2006.
page o be published, Hakoda e. Japan, May, 2006. LNCS.
[I51
J.
Peiia,
R.
Le y. and
R.
Co chuelo. Towa ds cla i ying he
impo ance o in e ac ions in agen -o ien ed so ~wa e engi-
nee ing.
ln e na ional lb oa ne ican Jou nal
q"
:I'
(25): 19-
-
28, 2005.
[I61 T. Reenskaug.
Wo king wi h Objec s: The OO a n 'o m,a e
Enginee ing Me hod
Manning Publica ions. 1990.
[I71
Y.
Sma agdakis and D. Ba o y. Mixin laye s: an
objec -o ien ed implemen a ion echnique o e inemen s
and collabo a ion-based designs.
ACM T ans
So h .
Ens.
Me hodol.,
1 1(2):215-255,2002.
[18]
F.
Zambonelli.
N.
Jennings, and
M.
Woold idge. De eloping
mul iagen sys ems: he
GAIA
me hodology.
ACM T an~ac
-
ion~ n z So Ma c Enginee ing and Me hodology,
12(3):3 17-
370, July 2003.