Multiagent System Product Lines: Challenges and Benefits

Mul iagen Sys em P oduc Lines: Challenges and Bene i s Joaquin Pe˜ na Uni e si y o Se ille Spain [email p o ec ed] Michael G. Hinchey NASA Godda d Space Fligh Cen e USA Michael.G.Hinche[email p o ec ed] An onio Ruiz-Co ´ es Uni e si y o Se ille Spain [email p o ec ed] 1 In oduc ion On he one hand, he ield o So wa e P oduc Lines (SPL), as desc ibed elsewhe e in his special issue, co e s all he so wa e de elopmen li ecycle necessa y o de elop a amily o p oduc s whe e he de i a ion o conc e e p od- uc s is made sys ema ically and apidly [10]. On he o he hand, Agen -O ien ed So wa e Enginee ing (AOSE) is a new so wa e enginee ing pa adigm ha a ose o apply bes p ac ice in he de elopmen o complex Mul i-Agen Sys- ems (MAS) by ocusing on he use o agen s, and o ganiza- ions (communi ies) o agen s as he main abs ac ions [7]. Following a a he alse s a , agen echnology has be- gun o come in o i s own. Wi h he ad en o biologically- inspi ed, pe asi e, and au onomic compu ing, he ad an- ages o , and necessi y o , agen -based echnologies and MASs has become ob ious. Un o una ely, cu en AOSE me hodologies a e dedica ed o de eloping single MASs. Clea ly, many MASs will make use o signi ican ly he same echniques, adap a ions, and app oaches. The ield is hus ipe o exploi ing he bene i s o SPL: educed cos s, imp o ed ime- o-ma ke , e c., and enhancing agen ech- nology in such a way ha i is mo e indus ially applicable. We belie e ha he e is much ha can be achie ed by combining he wo app oaches: applying he SPL philoso- phy o building a MAS will a o d all o he ad an ages o SPLs and make MAS de elopmen mo e p ac ical. Thus, he con ibu ion o his wo k is wo old: s ess he easibil- i y and bene i s o wha we call Mul i-Agen Sys ems P od- uc Lines (MAS-PL); and show eade s he main esea ch challenges in he de elopmen o MAS-PLs. 2 Feasibili y o MAS-PL and bene i s The so wa e p ocess p oposed in AOSE p esen s many simila i ies wi h he p ocess ollowed in SPL o he i s ac i i ies o he domain enginee ing, which is in cha ge o p o iding he eusable co e asse s ha a e exploi ed du - ing he de i a ion o p oduc s, done a he applica ion en- ginee ing [10]. Following he nomencla u e used in [10], he ac i i ies, usually pe o med i e a i ely and in pa allel, o domain enginee ing ha p esen co ela ion wi h AOSE a e: Domain Requi emen s Enginee ing. Bo h app oaches use models based on simila concep s: ea u es in he case o SPLs, and sys em-goals in he case o AOSE [3, 4]. Bo h ep esen equi emen s obse able by he end use . Bo h app oaches use hie a chical dia- g ams whe e ea u es/goals a e decomposed in o ine g ain ones. Howe e , SPL emphasizes he analysis o he scope o he SPL, i.e. he p oduc s inside i , and he analysis o common and a iable ea u es ac oss he SPL, which is no ca ied ou by AOSE. In [5, 9], a i s s ep owa d adap ing sys em-goals o MAS-PL and documen ing a iabili y is shown. Domain Design. Bo h app oaches de elop a chi ec u e- independen models ha a emp o analyze how ea- u es and i s a iabili y can be ma e ialized. In AOSE, ole models a e used wi h his pu pose [12], and some app oaches in SPL also p opose he same ap- p oach [6, 11]. Howe e , agen - ocused models show addi ional in o ma ion ha is no needed in SPL- ole models, such as he goals o he agen s, o whe he hey a e used o abs ac IA echniques, while no showing how hese ole models can be eused o di e en p od- uc s. Domain Realiza ion. Bo h app oaches ocus on designing a de ailed a chi ec u e. In he case o SPL, a common a chi ec u e o all p oduc s and a se o eusable as- se s. In he case o AOSE, a single a chi ec u e ha ul ills all o he sys em-goals o he MAS. Some ap- p oaches in bo h ields base he cons uc ion o he a chi ec u e on ole model composi ion [6, 11]. In [9], au ho s p esen ed he i s s eps owa d building he co e a chi ec u e o a MAS-PL based on au o- ma ic analysis o sys em-goals models adap ed o ea- u e models using [2]. This, along wi h he shown i s esea ch pape s de el- oped unde his ield, shows ha he bene i s o enabling MAS-PL a e eachable. The main bene i is s aigh o - wa d: AOSE can bene i s o all SPL ad an ages helping i o each he indus ial wo ld. Howe e , as we show in he nex sec ion, he e exis also a numbe o esea ch chal- lenges ha mus cons i u e he esea ch agenda needed o allow MAS-PL o become a eali y. 3 Fu u e Challenges SPL o dis ibu ed sys ems. Dis ibu ed sys ems ha e no been a ho opic in he SPL ield. Howe e , MASs a e dis ibu ed sys ems ha will need new adap ed echniques o be co e ed. Al hough ce ainly i will a ec o he whole de elopmen cycle, one o he i s s eps we o esee is he need o in es ing in he use o in e ac ion-based models, such as ole models. The esea ch unde aken o e his opic may ex end he ap- plicabili y o SPL no only o MASs, bu also o o he kind o dis ibu ed sys ems such as web se ices[1]. AOSE de iciencies. As shown be o e, AOSE does no co e some o he ac i i ies o SPL. These a e mainly concen a ed on commonali y analysis, and i s impli- ca ions a whole SPL app oach. Ano he ho opic can be ound in he p oduc managemen ac i i y ha is pe o med in pa allel wi h domain and applica ion en- ginee ing. I is in cha ge o managing he economic aspec s o a SPL. Gi en ha he p oduc s and he ma - ke s o MASs a e qui e di e en om he ones yp- ically used in SPL, a deal o e o mus be pu on s udying hese aspec s. Finally, as AOSE is de o ed o de elop single p oduc s, applica ion enginee ing is no p esen in AOSE. Resea che s should also in es e o s on s udying his ac i i y. Managemen o e ol ing sys ems. Agen -based e ol ing sys ems esul s on la ge so wa e sys ems ha adap and lea ns o m changes in he en i onmen . The de- elopmen o hese sys ems esul s in a complex ask whe e sys ems usually become unmanageable om an enginee ing poin o iew. SPL can help his ask by iewing an e ol ing sys em as a SPL whe e a di e - en s a e in he sys em is iewed as a sepa a e p oduc [8]. This decomposes he sys em in o well iden i ied chunks and a well de ined con ex whe e each p oduc will appea , wha helps o deal wi h he inhe en com- plexi y o MASs. Sel -* p ope ies o agen s We belie e ha agen echnol- ogy may b ing also ad an ages o SPL. Gi en esea ch e o s in es ed in p o iding agen s wi h he capabil- i ies o communica e wi h each o he a he seman ic le el, o o p o ide capabili ies o sel -o ganiza ion, sel -op imiza ion, sel -healing, e c., he main enance and e olu ion o he co e a chi ec u e may be simpli- ied. Addi ionally, he in eg a ion cos s o new ea u es o a ce ain p oduc , o e en he en i e SPL, may be dec eased. 4 Conclusions MAS-PL, d awing bene i s om bo h SPL and AOSE, will help in he indus ial exploi a ion o agen echnology, sa ing bo h e o and cos . We ha e iden i ied se e al chal- lenges, such as adap ing cu en AOSE enginee ing ech- niques o he SPL philosophy, which in many cases equi es he de elopmen o new ac i i ies/models om sc a ch. Howe e , a symbiosis be ween bo h AOSE and SPL a ises when AOSE also p o ides bene i s o SPL, mainly h ough encou aging and imp o ing esea ch on SPL o complex dis ibu ed sys ems. As can be seen, MAS-PLs ep esen a g ea , and wo hwhile, challenge ha will ce ainly a ac he in e es o many p ac i ione s and esea che s. Re e ences [1] D. Bena ides, A. R. Co ´ es, M. A. Se ano, and C. M. de Oca. A i s app oach o build p oduc lines o mul i-o ganiza ional web based sys ems (mows). In T. B¨ ohme, V. La ios-Rosillo, H. Unge , and H. Unge , edi o s, IICS, olume 3473 o Lec- u e No es in Compu e Science, pages 91–98. Sp inge , 2004. [2] D. Bena ides, A. Ruiz-Co ´ es, and P. T inidad. Au oma ed easoning on ea u e models. LNCS, Ad anced In o ma ion Sys ems Enginee ing: 17 h In e na ional Con e ence, CAiSE 2005, 3520:491–503, 2005. [3] P. B esciani, A. Pe ini, P. Gio gini, F. Giunchiglia, and J. My- lopoulos. T opos: an agen -o ien ed so wa e de elopmen me hodology. Jou nal o Au onomous agen s and Mul iagen Sys ems, 8(3), 2004. [4] K. Cza necki and U. Eisenecke . Gene a i e P og amming: Me hods, Tools, and Applica ions. Addison–Wesley, 2000. [5] J. Dehlinge and R. Lu z. A p oduc -line app oach o p o- mo e asse euse in mul i-agen sys ems. In SELMAS, olume 3914 o Lec u e No es in Compu e Science, pages 161–178. Sp inge , 2005. [6] 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 oceedings - So wa e, 151(4):187–198, 2004. [7] N. Jennings. An agen -based app oach o building complex so wa e sys ems. Communica ions o he ACM, 44(4):35–41, 2001. [8] J. Pe˜ na, M. G. Hinchey, and A. Ruiz-Co ´ es. Managing he e olu ion o an en e p ise a chi ec u e using a mas-p oduc - line app oach. In In e na ional Wo kshop on Wo shop on Sys- em/So wa e A chi ec u es 2006, Ne ada,USA, June, 2006. CSREA P ess. 2 [9] J. Pe˜ na, M. G. Hinchey, and A. Ruiz-Co ´ es. Building he co e a chi ec u e o a mul iagen sys em p oduc line: Wi h an ex- ample om a u u e nasa mission. In 7 h In e na ional Wo k- shop on Agen O ien ed So wa e Enginee ing 2006, page o be published, Hakoda e, Japan, May, 2006. LNCS. [10] K. Pohl, G. B¨ ockle, and F. an de Linden. So wa e P oduc Line Enginee ing : Founda ions, P inciples and Techniques. Sp inge , Sep embe 2005. [11] 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 w. Eng. Me hodol., 11(2):215–255, 2002. [12] F. Zambonelli, N. R. Jennings, and M. Woold idge. De elop- ing mul iagen sys ems: The gaia me hodology. ACM T ans. So w. Eng. Me hodol., 12(3):317–370, 2003. 3