A bargaining-specific architecture for supporting automated service agreement negotiation systems

Science o Compu e P og amming 77 (2012) 4–28 Con en s lis s a ailable a SciVe se ScienceDi ec Science o Compu e P og amming jou nal homepage: www.else ie .com/loca e/scico A ba gaining-speci ic a chi ec u e o suppo ing au oma ed se ice ag eemen nego ia ion sys ems Manuel Resinas∗, Pablo Fe nández, Ra ael Co chuelo Uni e sidad de Se illa, ETSI In o má ica, A da. Reina Me cedes, s/n, E-41012 Se illa, Spain a icle in o A icle his o y: Recei ed 25 Feb ua y 2009 Recei ed in e ised o m 3 Sep embe 2010 Accep ed 7 Sep embe 2010 A ailable online 1 No embe 2010 Keywo ds: So wa e a chi ec u e Au oma ed nego ia ion Se ice ag eemen s abs ac The p o ision o se ices is o en egula ed by means o ag eemen s ha mus be nego ia ed be o ehand. Au oma ing such nego ia ions is appealing inso a as i o e comes one o he mos o en ci ed sho comings o human nego ia ion: slowness. Ou analysis o he equi emen s o au oma ed nego ia ion sys ems in open en i onmen s sugges s ha some o hem canno be ackled in a p o ocol-independen manne , which mo i a es he need o a p o ocol-speci ic a chi ec u e. Howe e , cu en s a e-o - he-a ba gaining a chi ec u es ail o add ess all o hese equi emen s oge he . Ou key con ibu ion is a ba gaining a chi ec u e ha add esses all o he equi emen s we ha e iden i ied. The de ini ion o he a chi ec u e includes a logical iew ha iden i ies he key a chi ec u al elemen s and hei in e ac ions, a p ocess iew ha iden i ies how he a chi ec u al elemen s can be g ouped oge he in o p ocesses, a de elopmen iew ha includes a so wa e amewo k ha p o ides a e e ence implemen a ion de elope s can use o build hei own nego ia ion sys ems, and a scena ios iew by means o which he a chi ec u e is illus a ed and alida ed. ©2010 Else ie B.V. All igh s ese ed. 1. In oduc ion Ag eemen s play a majo ole o egula e bo h unc ional and non- unc ional p ope ies, as well as gua an ees ega ding he p o isioning o a se ice [1,8,23,26]. Many au ho s ha e ocused on au oma ing he nego ia ion o such ag eemen s as a means o imp o e e iciency and bene i om he many oppo uni ies ha elec onic businesses b ing [25,37]. De eloping an au oma ed nego ia ion sys em equi es selec ing and in eg a ing he mos app op ia e nego ia ion p o ocol, decision- making algo i hm, and da a model. This selec ion depends on he con ex o he nego ia ion, and i has usually been accomplished in ad hoc manne s [16]. Ou wo k ocuses on open en i onmen s in which change is he majo d i ing o ce. This makes ad hoc solu ions o li le in e es because o he de elopmen cos s incu ed. The eason o changes in open en i onmen s is mani old: on he one hand, pa ies may appea o disappea unexpec edly, i is no usually possible o ha e comple e in o ma ion abou hem, hey may implemen a a ie y o nego ia ion p o ocols, and hey may ha e di e ging beha iou s du ing a nego ia ion; on he o he hand, use s’ equi emen s a e subjec o con inuous changes due o new business ules and egula ions, new ypes o business- ela ed e en s, o new ope a ions [29]. In Re . [32], we iden i ied a numbe o objec i e equi emen s ha allowed us o compa e cu en p oposals ega ding se ice ag eemen nego ia ion in open en i onmen s [2,3,6,17,14,19,24,30,33,38,40]. The conclusion was ha none o he ele en p oposals su eyed was adequa e o nego ia e se ice ag eemen s in open en i onmen s. To o e come his issue, he NegoFAST-Co e a chi ec u e was de eloped; i de ines he mos abs ac a chi ec u al elemen s o an au oma ed nego ia ion sys em om a p o ocol-independen and echnology-agnos ic poin o iew. Un o una ely, al hough some o ∗Co esponding au ho . E-mail add ess: [email p o ec ed] (M. Resinas). 0167-6423/$ – see on ma e ©2010 Else ie B.V. All igh s ese ed. doi:10.1016/j.scico.2010.09.008 M. Resinas e al. / Science o Compu e P og amming 77 (2012) 4–28 5 Fig. 1. Sequence diag am o a ypical in e ac ion in a ba gaining p o ocol. he equi emen s can be sa is ied using p o ocol-independen a chi ec u al elemen s, e.g., acqui ing in o ma ion om o he pa ies, modelling he wo ld, o nego ia ing a p o ocol, a ew equi emen s a e p o ocol-dependen and equi e speci ic a chi ec u al elemen s. The con ibu ion ha we p esen in his a icle is an a chi ec u e called NegoFAST-Ba gaining ha e ines and complemen s he NegoFAST-Co e a chi ec u e by de ining p o ocol-speci ic a chi ec u al elemen s o deal wi h ba gaining p o ocols. On he one hand, ha ing a so wa e a chi ec u e [22] o a p oblem domain is appealing inso a i p o ides an abs ac , eusable, easy- o-main ain, and easy- o-adap design ha p o ides he ha ness de elope s need o ace he de elopmen o a sys em wi hou incu ing he cos o de eloping i om sc a ch o in ad hoc manne s. Fu he mo e, a so wa e a chi ec u e p o ides a common ocabula y ha b idges he gap be ween he many exis ing e minologies, which, he e o e, helps ease he communica ion amongs de elope s. No e ha we use he e m ‘a chi ec u e’ in he same sense as in K uch en’s ’s semina a icle [22], acco ding o which an a chi ec u e is composed o 4 +1 iews, namely: a logical iew, which de ails he unc ional iew o he sys em; a p ocess iew, which desc ibes how he a chi ec u al elemen s can be g ouped oge he in o p ocesses; a de elopmen iew, which desc ibes he o ganisa ion o he so wa e modules in he so wa e de elopmen en i onmen ; a scena ios iew in which he a chi ec u e is illus a ed; and a physical iew, which desc ibes how a sys em mus be deployed. On he o he hand, ba gaining p o ocols a e in widesp ead use [5,16,36]. Such p o ocols ha e he ollowing ea u es: hey a e bila e al, which means ha hey a e ca ied ou be ween wo pa ies ha play he oles o ini ia o and esponde ; hey a e sequen ial, which means ha he same pa y canno send wo messages in a ow, excep o nego ia ion messages ha include he ejec Nego ia ion o he wi hd aw pe o ma i es, which can be sen a any ime; inally, messages a e p oposal- based, which means ha hey con ain one pe o ma i e and one o mo e p oposals, bu hey do no include any a gumen s o con ince he o he pa y abou accep ing a p oposal. Fig. 1 shows a sequence diag am ha illus a es a ypical execu ion o a ba gaining p o ocol; no e ha bo h pa ies exchange se e al p oposals using pe o ma i e p opose un il one o hem decides o send a binding p oposal by means o pe o ma i e commi , which is accep ed by he o he pa y by means o pe o ma i e accep . The es o he a icle is o ganised as ollows: in Sec ion 2, we summa ise ou equi emen s, wi h an emphasis on he subse o p o ocol-speci ic ones, and p o e ha none o he s a e-o - he-a p oposals ha we ha e su eyed add ess all o hese equi emen s oge he ; hen, he K uch en a chi ec u al iews [22] o ou p oposal a e p esen ed in Sec ions 3(logical iew), 4(p ocess iew), 5(de elopmen iew), and 6(scena ios iew). (No e ha we do no p o ide a physical iew as such, since ou p oposal is no in ended o be he a chi ec u e o a speci ic sys em, bu a eusable a chi ec u e ha can be used o de ise and implemen a a ie y o ac ual nego ia ion sys ems.) We epo on ou conclusions and u u e esea ch pa hs in Sec ion 7. Finally, Appendix p o ides an o e iew o NegoFAST-Co e. 2. Mo i a ion In his sec ion, we i s epo on a numbe o key p oblems and equi emen s wi h which a se ice ag eemen nego ia o mus deal in open en i onmen s; we hen analyse hese equi emen s and conclude ha some o hem canno be ackled in a p o ocol-independen manne ; inally, we su ey cu en ela ed p oposals and conclude ha none o hem add esses all o he p o ocol-speci ic equi emen s we ha e iden i ied, which mo i a ed us o wo k on a new p oposal. 2.1. Key p oblems and equi emen s Au oma ed nego ia ion sys ems mus cope wi h ou p oblems when acing au oma ed nego ia ions o se ice ag ee- men s in open en i onmen s. These p oblems led o he o mula ion o a numbe o equi emen s ha a e summa ised in Table 1 [32], namely: Nego ia ions a e mul i- e m. The nego ia ions o a ypical se ice ag eemen in ol es e ms like a ailabili y, esponse ime, secu i y, o p ice, o name a ew. This has an in luence on he nego ia ion p o ocol and he ag eemen p e e ences ha he au oma ed nego ia ion sys em mus suppo . On he one hand, i makes i desi able o an au oma ed nego ia ion sys em o suppo mul i- e m nego ia ion p o ocols (REQ 1.1), such as mos (i no e e y) ba gaining 6M. Resinas e al. / Science o Compu e P og amming 77 (2012) 4–28 Table 1 Requi emen s in au oma ed nego ia ion sys ems. Nego ia ions a e mul i- e m REQ Desc ip ion 1.1 Suppo mul i- e m nego ia ion p o ocols. 1.2 Manage exp essi e ag eemen p e e ences. Pa ies a e he e ogenous REQ Desc ip ion 2.1 Suppo mul iple nego ia ion p o ocols. 2.2 Nego ia e he nego ia ion p o ocol in a p e-nego ia ion phase in which a nego ia ion p o ocol is ag eed. 2.3 Suppo mul iple decision-making algo i hms o ace he di e en beha iou s o he o he pa ies du ing he nego ia ion. 2.4 Allow use p e e ences abou he nego ia ion p ocess o be changed o e ime. Pa ial in o ma ion abou pa ies REQ Desc ip ion 3.1 Manage di e en ypes o knowledge abou he o he pa ies. 3.2 Suppo di e se que y capabili ies. 3.3 Build analysis-based models o pa ies. Ma ke s a e dynamic REQ Desc ip ion 4.1 Suppo se e al nego ia ions wi h di e en pa ies a he same ime. 4.2 Selec decision-making algo i hms dynamically acco ding o he e olu ion o he simul aneous nego ia ions. 4.3 Build ma ke models. 4.4 Suppo decommi men om p e iously es ablished ag eemen s. 4.5 Supe ised c ea ion o ag eemen s o a oid commi ing o ag eemen s ha canno be sa is ied. p o ocol [16]; on he con a y, mos auc ioning p o ocols allow us o nego ia e on a unique e m only, usually he p ice [15], excep o he p oposal by Bichle [7]. On he o he hand, use p e e ences can be exp essed in a a ie y o ways, e.g., u ili y unc ions [11], combina ions o a ibu es [9], o uzzy cons ain s [25], bu mul i- e m nego ia ions equi e he managemen o exp essi e ag eemen p e e ences ega ding mul iple e ms (REQ 1.2) so ha hey cap u e he ela ionships be ween e ms and, hence, enable making ade-o s du ing nego ia ions. Pa ies a e he e ogenous. In an open en i onmen , he pa ies ha may ge in ol ed in a nego ia ion p ocess a e ob iously expec ed o be he e ogeneous, no only ega ding hei echnology, bu also hei seman ics and beha iou s. I is hen desi able o an au oma ed nego ia ion sys em o suppo mul iple nego ia ion p o ocols (REQ 2.1), o be able o nego ia e he nego ia ion p o ocol (REQ 2.2) in cases in which a pa y suppo s se e al o hem, suppo mul iple decision-making algo i hms (REQ 2.3) o adap o di e en beha iou s o he o he pa ies, and o allow use p e e ences abou he nego ia ion p ocess (REQ 2.4), e.g., deadline o eage ness. Pa ial in o ma ion abou pa ies. Ha ing as much in o ma ion as possible abou he o he pa ies is impo an o s eng hen one’s nego ia ion capabili ies [43]. Un o una ely, mo e o en han no , au oma ed nego ia ion sys ems ha e only pa ial in o ma ion abou hei con ex [25]. As a conclusion, i is impo an ha such sys ems can manage di e en ypes o knowledge abou he o he pa ies (REQ 3.1), e.g., whe he hey end o concede [28], hei nego ia ion deadline, hei epu a ion o geog aphical loca ion; i is also impo an o be able o suppo a a ie y o que y capabili ies (REQ 3.2), e.g., a epu a ion p o ide , an ad hoc API, o a WS-Ag eemen empla e [1]; inally, i mus be able o build analysis-based models o pa ies (REQ 3.3) building on he p e ious in o ma ion and on p e ious nego ia ion p ocesses, since his shall de ini ely help make be e decisions in he u u e [43]. Ma ke s a e dynamic. Se ices a e no s o able, which means ha esou ces no used yes e day a e wo hless oday [14]. This is he eason why se ice ma ke s end o be ex emely dynamic. As a consequence, i would be con enien o an au oma ed nego ia ion sys em o suppo se e al nego ia ions wi h di e en pa ies a he same ime (REQ 4.1), o selec decision-making algo i hms dynamically (REQ 4.2), o suppo decommi men om p e iously es ablished ag eemen s (REQ 4.3), o supe ise he c ea ion o ag eemen s (REQ 4.4), and o build ma ke models (REQ 4.5). These equi emen s allow nego ia ion sys ems o choose he pa y wi h which he mos p o i able ag eemen can be made, o change i s beha iou wi h ega d o speci ic pa ies acco ding o how he nego ia ion is going on globally, o e oke an ag eemen in case ano he ha is mo e p o i able is ound [35], and no o commi o an ag eemen ha canno be sa is ied o ha migh no be sa is ac o y. 2.2. P o ocol-speci ic equi emen s Nego ia ion p o ocols go e n how nego ia ions a e pe o med. They ha e a s ong in luence on he decision-making echniques used by he pa ies and he inal ou come o he p ocess. M. Resinas e al. / Science o Compu e P og amming 77 (2012) 4–28 7 Al hough he e a e a a ie y o nego ia ion p o ocols, hey all may be analysed in e ms o i e undamen al aspec s, namely: pa ies, which a e commonly a consume and a p o ide , al hough some nego ia ion p o ocols equi e a media o o acili a e he nego ia ion p ocess [19,20]; pe o ma i es, which e lec he in en ion o a message, e.g., p opose,accep , o commi ; ules, which exp ess es ic ions ega ding how p oposals a e buil o when a pa y can send a p oposal; in o ma ion exchanged, which is he ype o in o ma ion exchanged du ing a nego ia ion; and ag eemen e ms nego iabili y, which is he se o e ms ha a e nego iable. Nego ia ion p o ocols can be g ouped in o a numbe o amilies, namely: ba gaining p o ocols [10], which in ol e exchanging p oposals and coun e -p oposals be ween he pa ies; auc ion p o ocols [15], which in ol e one o mo e pa ies called auc ionee s, who s a he auc ion, and o he pa ies called bidde s ha bid ollowing a p o ocol ha may allow one o se e al ounds; and a gumen a ion-based p o ocols [31], which in ol e exchanging a gumen s o pe suade he o he pa y o accep an o e . This a iabili y in nego ia ion p o ocols causes ha some o he equi emen s desc ibed in he p e ious sec ion mus be deal wi h in a p o ocol-speci ic manne , namely: •Suppo mul i- e m nego ia ion p o ocols (REQ 1.1) and suppo mul iple nego ia ion p o ocols (REQ 2.1). Al hough mos nego ia ion p o ocols in ol e he exchange o p oposals be ween pa ies, his exchange may be ca ied ou wi hou es ic ions on he con en s o he p oposal and coun e -p oposal [10] o wi h es ic ions on he o de in which e ms a e nego ia ed [12] o on he e ms o he p oposals [18]. This a iabili y mus be deal wi h in a p o ocol-speci ic manne . •Suppo se e al decision-making algo i hms (REQ 2.3) and selec hem dynamically (REQ 4.2). The decisions ha mus be made by an au oma ed nego ia ion sys em can be di ided in o wo g oups. Fi s , i has o handle he commi men o new ag eemen s, i.e., deciding i an ag eemen is accep able and con enien o commi o, and he decommi men om p e iously c ea ed ag eemen s [28,35]. Second, i has o gene a e esponses o he o he pa ies in he nego ia ion by implemen ing an app op ia e algo i hm [10,11,21,25]. The i s decision is p o ocol-independen . Howe e , depending on he nego ia ion p o ocol, he way decisions a e made ega ding he gene a ion o esponses can be signi ican ly di e en . Fo ins ance, in a ba gaining p o ocol, he esponse has o be eques ed o each nego ia ion message ecei ed, whe eas in an auc ion p o ocol, bids can be placed a any momen . As a consequence, suppo ing decision-making algo i hms mus be designed in a p o ocol-speci ic manne . •Suppo se e al nego ia ions simul aneously (REQ 4.1) and allow use p e e ences abou he nego ia ion p ocess (REQ 2.4). On he one hand, i is necessa y o ha e a global iew o all nego ia ions o suppo hem p ope ly; on he o he hand, allowing use p e e ences abou he nego ia ion p ocess in ol es changing he nego ia ion beha iou based on he use ’s p e e ences. Consequen ly, he beha iou o each nego ia ion mus be guided based on how well he o he simul aneous nego ia ions a e pe o ming and on he use p e e ences. To his end, i is necessa y o ha e an unde s anding o how he decision-making algo i hms wo k. The e o e, he changes in he beha iou mus be designed in a p o ocol-speci ic manne . 2.3. Analysis o he ela ed wo k The key p oblems desc ibed abo e p o ide a numbe o objec i e equi emen s ha can be used o compa e cu en s a e-o - he-a au oma ed nego ia ion a chi ec u es. Table 2 summa ises ou compa ison. An ‘✓’ in a cell means ha he co esponding p oposal p o ides explici suppo o he co esponding equi emen ; a ‘∼’ indica es ha i is suppo ed pa ially; a ‘✗’ indica es ha he ea u e is no suppo ed; ‘NA’ means ha he e is no in o ma ion a ailable. (No e ha we do no ake in o accoun p oposals like he ones desc ibed in [21,39] because hey a e speci ic-pu pose nego ia ion sys ems, no so wa e a chi ec u es.) A quick look a Table 2 e eals ha none o he s a e-o - he-a p oposals we ha e su eyed is comple e wi h ega d o he p o ocol-dependen equi emen s we ha e iden i ied, which jus i ies he need o esea ch ega ding an a chi ec u e ha add esses hem all. Nex we p o ide addi ional de ails on he compa ison. Fo he sake o homogenei y, we ha e g ouped hem in o p o ocol-o ien ed and in elligence-o ien ed p oposals. P o ocol-o ien ed p oposals ocus on p o iding suppo o deal wi h he nego ia ion p o ocol and low-le el in e - ope abili y issues. Howe e , hey do no p o ide any kind o suppo o implemen decision making algo i hms. P o ocol- o ien ed p oposals can be di ided in o wo ca ego ies: some o hem de ine a nego ia ion hos o ma ke place ha ac s as a media o be ween he nego ia ing pa ies [19,33,38]; con a ily, o he s jus desc ibe he elemen s ha a e equi ed o manage nego ia ion p o ocols p ope ly [3]. Kim and Sege [19] p esen ed a p oposal ha desc ibes a web se ices-enabled ma ke place a chi ec u e ha de ines execu able nego ia ion p o ocols in BPEL. Rinde le and Benyouce [33] ollowed a simila app oach, bu nego ia ion p o ocols a e speci ied as s a echa s and la e mapped on o BPEL. Simila ly, Silk oad [38] elies on a me a-model called oadmap ha is in ended o cap u e he cha ac e is ics o a nego ia ion p ocess, and an applica ion amewo k called skele on ha p o ides se e al modula and con igu able nego ia ion se ice componen s. Con a y o hese p oposals, Ba olini e al. [3] p esen ed a axonomy o ules o cap u e a a ie y o nego ia ion mechanisms and a simple in e ac ion p o ocol based on FIPA speci ica ions ha is used oge he wi h he ules o de ine nego ia ion p o ocols; hey also de ined a se o oles and an OWL-based language o exp ess nego ia ion p oposals and ag eemen s. In elligence-o ien ed p oposals y o make he de elopmen o au oma ed nego ia ion sys ems easie by gi ing a common g ounding o de eloping decision making algo i hms and imp o ing eusabili y. Fu he mo e, some o hese 8M. Resinas e al. / Science o Compu e P og amming 77 (2012) 4–28 Table 2 Compa ison o au oma ed nego ia ion a chi ec u es. P o ocol-o ien ed a chi ec u es P oposal Requi emen 1.1 2.1 2.3 2.4 4.1 4.2 Kim and Sege [19]✓ ✓ ✗ ✗ ✗ ✗ Rinde le and Benyouce [33]✓ ✓ ✗ ✗ ✗ ✗ Ba olini e al. [3]✓ ✓ ✗ ✗ ✗ ✗ Silk oad [38]✓ ✓ ✗ ✗ ✗ ✗ In elligence-o ien ed a chi ec u es P oposal Requi emen 1.1 2.1 2.3 2.4 4.1 4.2 Ash i e al. [2]✓ ✓ NA NA NA NA Ludwig e al. [24]✓ ✓ ✓ ✗ ✗ ✗ PANDA [14]✓ ✓ ✓ ✓ ∼✗ DynamiCS [40]✓✓✓NA ✗ ✗ Benyouce and Ve ons [6] NA ✓ ✓ NA ✗ ✓ Jonke e al. [17]✓NA ✓ ✓ ✗ ✗ Pau obally e al. [30]✓ ✗ ∼✗∼✗ 1.1 Suppo mul i- e m nego ia ion p o ocols. 2.4 Allow use p e e ences abou nego ia ion p ocesses. 2.1 Mul iple p o ocol suppo . 4.1 Suppo se e al nego ia ions simul aneously. 2.3 Mul iple decision-making algo i hms. 4.2 Selec decision-making algo i hms dynamically. p oposals [2,14,40] also deal wi h p o ocol managemen . Howe e , ins ead o dealing wi h he in e ac ions amongs au oma ed nego ia ion sys ems like p o ocol-o ien ed p oposals, hey ocus on decoupling he p o ocol and he in elligence algo i hms o make he la e compa ible wi h a numbe o di e en p o ocols. Ash i e al. [2] desc ibed an agen -o ien ed a chi ec u e a a e y high le el o abs ac ion, which means ha hey le many aspec s ela ed o he decision-making algo i hms unspeci ied. Ludwig e al. [24] p esen ed an a chi ec u e o se ice ag eemen nego ia ion in se ice g ids. I builds on WS-Ag eemen and p o ides a p o ocol se ice p o ide and a decision making se ice p o ide o deal wi h he nego ia ion p ocess. Un o una ely, i does no suppo se e al simul aneous nego ia ions o changing he decision-making algo i hms dynamically. PANDA [14] is an a chi ec u e ha mixes u ili y unc ions and ules o ca y ou he decision- making p ocess. The decision-making componen elies on ules, u ili y unc ions and an objec pool wi h se e al es ima ion lib a ies, he nego ia ion his o y and he cu en o e . Howe e , i does no allow us o selec decision-making algo i hms dynamically and, al hough i claims o suppo se e al simul aneous nego ia ions, he coo dina ion mechanism amongs hem is no desc ibed. DynamiCS is an ac o -based a chi ec u e ha was de ised by Tu e al. [40]. I makes a clea dis inc ion be ween he nego ia ion p o ocol and he decision-making model, and i uses a plug-in mechanism o suppo new p o ocols and s a egies. Howe e , i does no suppo se e al simul aneous nego ia ions o p o ide suppo o change he decision- making algo i hms a un ime. Benyouce and Ve ons [6] p esen ed an a chi ec u e ha is based on he sepa a ion o p o ocols and s a egies wi hin a se ice-o ien ed a chi ec u e ha makes he deploymen and in eg a ion wi h cu en in as uc u es easie . Howe e , he se ices ha can be composed in o he sys em a e aguely de ined. In addi ion, i does no p o ide any mechanisms o allow se e al simul aneous nego ia ions. Jonke e al. [17] desc ibed a componen -based gene ic agen a chi ec u e o mul i-a ibu e nego ia ion. I copes wi h mul i- e m nego ia ions success ully. Howe e , i is no clea i i suppo s se e al nego ia ion p o ocols and i does no p o ide a mechanism o coo dina e se e al simul aneous nego ia ions o o change he decision-making algo i hm a un ime. Pau obally e al. [30] desc ibed an a chi ec u e o web se ice nego ia ion in g ids. I deals wi h he exp essi eness o se ice ag eemen s success ully, and i allows ag eemen s o be c ea ed in a supe ised manne by means o he so-called WS-DAIOn se ice; un o una ely, i only suppo s a nego ia ion p o ocol, and, al hough he a chi ec u e seems o suppo se e al simul aneous nego ia ions, he au ho s do no del e in o hese de ails. 3. The NegoFAST-Ba gaining logical iew Acco ding o K uch en [22], he logical iew o an a chi ec u e p o ides an insigh in o he unc ional iew o he sys em, which includes bo h he s a ic and dynamic models o i s a chi ec u al elemen s. We used he Gaia me hodology and i s o ganisa ional me apho o wo k on his iew [42]. The Gaia me hodology decomposes he logical iew o an a chi ec u e in o o ganisa ions, oles, in e ac ions, and en i onmen al esou ces. O ganisa ions g oup se e al oles and ha e a conc e e and well de ined goal; oles a e p ecisely de ined asks ha mus be ca ied ou by one o mo e so wa e a e ac s; in e ac ions ep esen he exchange o messages amongs wo o mo e oles; he en i onmen de e mines he esou ces ha oles can use, con ol, o consume o achie e an o ganisa ional goal. As men ioned in he in oduc ion, NegoFAST-Co e [32] de ines he mos abs ac a chi ec u al elemen s o an au oma ed nego ia ion sys em om a p o ocol-independen and echnology-agnos ic poin o iew, whe eas NegoFAST-Ba gaining M. Resinas e al. / Science o Compu e P og amming 77 (2012) 4–28 9 Fig. 2. Logical iew o NegoFAST-Ba gaining. e ines and complemen s i by means o ba gaining-speci ic elemen s. (C . Appendix o a sho in oduc ion o NegoFAST- Co e.) Consequen ly, he logical iew o NegoFAST-Ba gaining is o ganised in o hose p o ocol-speci ic oles o a chi ec u al elemen s, namely: P o ocolHandle , which p o ides suppo o se e al mul i- e m nego ia ion p o ocols, Nego ia ionCoo - dina o , which coo dina es he P o ocolHandle and he ResponseGene a o o handle se e al nego ia ions simul aneously, ResponseGene a o , which decides he mos app op ia e message o answe an incoming nego ia ion message, and Ne- go ia ionCon ex Da a, which s o es in o ma ion ela ed o he nego ia ions ha a e being ca ied ou by he au oma ed nego ia ion sys em. NegoFAST-Ba gaining e ines he p e ious a chi ec u al elemen s in o he ollowing ba gaining-speci ic oles, c . Fig. 2: •Al hough di e en , all nego ia ion p o ocols suppo ed by one p o ocol-speci ic a chi ec u e ha e many commonali ies, e.g., all ba gaining p o ocols in ol e he sequen ial exchange o p oposals be ween wo pa ies. As a consequence, he suppo o mul i- e m nego ia ion p o ocols (REQ 1.1) and se e al ba gaining nego ia ion p o ocols (REQ 2.1) is deal wi h by de ining a gene ic ba gaining p o ocol and specialising he P o ocolHandle in a Ba gainingP o ocolHandle , which con e s speci ic ba gaining p o ocols in o he gene ic one. •Since all nego ia ion p o ocols suppo ed by a p o ocol-speci ic a chi ec u e use a common se o pe o ma i es and he same ype o message con en s, e.g., a p oposal composed o a se o e ms, he suppo o mul iple decision-making algo i hms (REQ 2.3) and he dynamic selec ion o such algo i hms (REQ 4.2) is ackled by di iding he ResponseGene a o in o he MessageCompose , which selec s he pe o ma i e ha is going o be used and composes he message, and he Builde Manage and P oposalBuilde , which deal wi h he c ea ion o a p oposal. •The suppo o se e al simul aneous ba gaining nego ia ions (REQ 4.1) and o allowing use p e e ences abou he ne- go ia ion p ocess (REQ 2.4) is implemen ed by in oducing he concep o nego ia ion policy o guide he gene a ion o esponses based on how well he o he nego ia ions a e pe o ming and he use p e e ences abou he nego ia ion p o- 10 M. Resinas e al. / Science o Compu e P og amming 77 (2012) 4–28 cess. To his end, he Nego ia ionCoo dina o is di ided in o a ole ha is esponsible o hose policies (PoliciesManage ), and wo oles (Bila e alNego ia o and Ba gainingCoo dina o ) ha deal wi h he coo dina ion ask. These elemen s mus in e ac wi h se e al NegoFAST-Co e p o ocol-independen oles ha a e depic ed on he le side o Fig. 2, namely: he P o ocolNego ia o , which decides he conc e e nego ia ion p o ocol ha shall be used du ing a nego ia ion and con igu es he P o ocolHandle acco ding o i ; he Sys emCoo dina o and he Pa yCoo dina o , which coo dina e when new p o ocol-speci ic nego ia ions mus be s a ed, and he Commi Handle , which mus app o e he submission o all binding nego ia ion messages, i.e., hose messages ha in ol e a i m commi men wi h he o he pa y. Fu he mo e, he gene a ion o esponses mus be p o ided wi h in o ma ion s o ed in se e al di e en p o ocol- independen en i onmen al esou ces, namely: he use ’s p e e ences, he wo ld model, he nego ia ion his o y, and o he aspec s ela ed o he cu en s a e o he au oma ed nego ia ion sys em. The way in which he p o ocol-speci ic equi emen s a e ackled shape he logical iew o NegoFAST-Ba gaining, which ollows a hyb id a chi ec u al s yle [4]. On he one hand, i uses a cen alised con ol s uc u e in which he Ba gainingCo- o dina o and he Bila e alNego ia o play he main oles: he o me ocuses on he in e ac ion wi h p o ocol-independen elemen s o an au oma ed nego ia ion sys em, whe eas he la e ocuses on he in e ac ion wi h he elemen s o he p o ocol-speci ic a chi ec u e. On he o he hand, i uses he eposi o y s yle, in which he Nego ia ionCon ex Da a s o es and p o ides in o ma ion gene a ed by he o he elemen s o he a chi ec u e o each nego ia ion. The Nego ia ionCon ex - Da a also uses he publishe –subsc ibe pa e n o ac as an ac i e eposi o y and no i y he o he elemen s o he a chi ec- u e abou he occu ence o a numbe o e en s. This is pa icula ly use ul o he PoliciesManage because i s beha iou depends on how he concu en nego ia ions a e e ol ing. Finally, he plug-in a chi ec u al pa e n [13] is used o allow se e al Ba gainingP o ocolHandle s and P oposalBuilde s a he same ime o enable he suppo o se e al ba gaining ne- go ia ion p o ocols and se e al decision-making algo i hms o c ea e p oposals. The Builde Manage allows he selec ion o he P oposalBuilde ha be e sui s he cu en nego ia ion scena io, whe eas he selec ed Ba gainingP o ocolHandle depends on he speci ic ba gaining nego ia ion p o ocol ha is being execu ed wi h he o he pa y. 3.1. P o ocol handle In NegoFAST-Ba gaining, he P o ocolHandle emains as one ole called Ba gainingP o ocolHandle ha deals wi h he in e ac ion wi h he o he pa ies ollowing a conc e e ba gaining p o ocol. I is con igu ed o manage a ba gaining p o ocol by means o in e ac ion Con igu eHandle , bu adap s i o he gene ic nego ia ion p o ocol ha is suppo ed by he Nego ia ionCoo dina o . This in ol es ans o ming he syn ax o he nego ia ion p o ocol in o nego ia ion messages ha a e unde s ood by he o he oles and sending hem by means o in e ac ion Con e P o ocol, ans o ming nego ia ion messages in o he conc e e syn ax o a nego ia ion p o ocol and sending hem ou o he o he pa ies, en o cing he ules o he nego ia ion p o ocol, and coping wi h e o s, e.g., missing messages, messages ha a i e oo la e, o unexpec ed messages. In addi ion, i may also ecei e eques s om o he pa ies o s a nego ia ions, in which case i o wa ds hem o he Sys emCoo dina o by means o in e ac ion IncomingNego ia ion. Below, we epo on he key ea u es o ou design: Roles. The nego ia ion is ca ied ou be ween wo pa ies: he ini ia o , which is he pa y ha ini ia es he nego ia ion, and he esponde , which is he o he pa y. No e ha bo h consume and p o ide can ac as ini ia o s o esponde s, and ha we do no p eclude he abili y o a pa y o pe o m se e al simul aneous nego ia ions. Pe o ma i es. The pe o ma i es in ou gene ic p o ocol a e c p (call o p oposals), p opose,commi , ejec Nego ia ion, ejec P oposal,wi hd aw and accep . No e ha , depending on he pe o ma i e, nego ia ion messages can be clas- si ied as binding nego ia ion messages, which in ol e a i m commi men wi h he o he pa y, e.g., pe o ma i e commi o accep , and non-binding nego ia ion messages, which do no in ol e such a i m commi men , e.g., pe o ma i e p opose. Rules. The ules imposed by he gene ic p o ocol a e as ollows: i mus s a wi h a nego ia ion message ha includes a c p, a p opose o a commi pe o ma i e; he c p pe o ma i e can only be used in an ini ial nego ia ion message; only commi ing p oposals, i.e. commi pe o ma i es, can be accep ed; i is an al e na ing p o ocol, i.e., a e one pa y sends a message, he o he pa y mus espond and so on. The only excep ion a e nego ia ion messages ha include ejec Nego ia ion o wi hd aw pe o ma i es ha can be sen a any ime du ing he nego ia ion. These ules a e o malised as he p o ocol s a e-machine in Fig. 3. In o ma ion exchanged. The ype o in o ma ion exchanged is es ic ed o p oposals and, hence, addi ional a gumen s can- no be sen oge he wi h a p oposal as in a gumen a ion-based p o ocols. Ag eemen e ms nego iabili y. Again, he gene ic p o ocol does no impose any es ic ion on he nego iabili y o ag eemen e ms. The e o e, as is he case o mos ba gaining p o ocols, e e y e m can be nego ia ed. 3.2. Nego ia ion coo dina o The Nego ia ionCoo dina o coo dina es he nego ia ions by linking he P o ocolHandle wi h he ResponseGene a o and he Commi Handle , c . Fig. 2. No e ha , al hough he Commi Handle is a p o ocol-independen ole, he M. Resinas e al. / Science o Compu e P og amming 77 (2012) 4–28 11 Fig. 3. S a e machine o he gene ic ba gaining nego ia ion p o ocol. Nego ia ionCoo dina o mus in e ac wi h i in o de o ca y ou a nego ia ion. The eason is ha he decision o send a binding nego ia ion message is made by he Commi Handle . As a consequence, he Nego ia ionCoo dina o may send as many non-binding nego ia ion messages as necessa y; bu , i needs he app o al o he Commi Handle be o e sending a binding nego ia ion message. To handle se e al simul aneous nego ia ions, he Nego ia ionCoo dina o uses nego ia ion con ex s. Each ongoing nego- ia ion de ines one nego ia ion con ex , which co esponds o he execu ion o one nego ia ion p o ocol. Howe e , o gi e ull suppo o se e al simul aneous nego ia ions, no only mus he Nego ia ionCoo dina o coo dina e each nego ia ion con ex independen ly, bu also mus ha e a global iew o all nego ia ion con ex s. The e o e, he Nego ia ionCoo dina o mus guide he beha iou o one nego ia ion con ex based on how well he o he concu en nego ia ions a e pe o ming. This is achie ed by means o he so-called nego ia ion policies, which a e guidelines abou how o gene a e esponses. Fo ins ance, i one nego ia ion is pe o ming pa icula ly well, i.e., he p oposals om he o he pa y a e e y appealing, he nego ia ion policies o he o he nego ia ion con ex s can be se o make he ResponseGene a o concede less. Fu he mo e, nego ia ion policies a e also used o guiding he nego ia ion acco ding o he use p e e ences abou he nego ia ion p ocess. No e ha he use o policies helps decouple he PoliciesManage om P oposalBuilde s since he communica ion amongs hem akes place jus by means o policies whose seman ics a e well-de ined. This enables he PoliciesManage o be buil independen ly om he P oposalBuilde s, which is impo an because he implemen a ion o ole P oposalBuilde s is a a ia ion poin o he au oma ed nego ia ion sys em, i.e., se e al P oposalBuilde s mus usually co-exis . In NegoFAST-Ba gaining, he Nego ia ionCoo dina o is di ided in o se e al oles o suppo se e al simul aneous ba gaining nego ia ions, namely: Bila e alNego ia o ,Ba gainingCoo dina o , and PoliciesManage . The coo dina ion ask is di ided in o he Bila e alNego ia o , which coo dina es one nego ia ion con ex , and he Ba gainingCoo dina o , which coo dina es se e al Bila e alNego ia o s and in e ac s wi h he Commi Handle . To add suppo o nego ia ion policies, he PoliciesManage , which chooses nego ia ion policies o each nego ia ion o guide hei beha iou , is in oduced. Nex , we p o ide addi ional de ails abou he p e ious oles. Role Bila e alNego ia o .I s goal is o ca y ou a single bila e al nego ia ion by o ches a ing he Ba gainingP o ocolHandle and he MessageCompose . Fu he mo e, i communica es wi h he Ba gainingCoo dina o o ask o app o al be o e sending a binding nego ia ion message, and i ecei es nego ia ion policies om he PoliciesManage . (No e ha he Ba gainingCoo dina o delega es he decision on whe he o app o e a binding nego ia ion message o he Commi Handle .) This ole implemen s, o each nego ia ion con ex , he s a e machine o a nego ia ion con ex o a ba gaining p o ocol. Fig. 4 p esen s he s a e machine and Fig. 5 p esen s a sequence diag am ha illus a es how i wo ks in a ypical scena io. The i s s ep is o con ac he Ba gainingP o ocolHandle o ini ialise i . The Ba gainingP o ocolHandle ac s as an in e media y be ween he Bila e alNego ia o and he o he pa y. Thus, all nego ia ion messages a e sen and ecei ed by means o in e ac ion Con e P o ocol, which makes he Bila e alNego ia o independen om conc e e nego ia ion p o ocols. The Ba gainingP o ocolHandle in o ms he Bila e alNego ia o o whe he i is ac ing as he ini ia o o he esponde o he 12 M. Resinas e al. / Science o Compu e P og amming 77 (2012) 4–28 Fig. 4. S a e machine o a nego ia ion con ex o a ba gaining p o ocol. Fig. 5. Sequence diag am o a nego ia ion con ex o a ba gaining p o ocol. in e ac ion: i i is ini ia o , i en e s s a e gene a ing and s a s he nego ia ion; o he wise, i en e s s a e wai ing and wai s o a message. Nex , we desc ibe each o he s a es in de ail: Wai ing. This is he s a e in which he Bila e alNego ia o is wai ing o a nego ia ion message om he o he pa y. When a new nego ia ion message is ecei ed by he Ba gainingP o ocolHandle , i sends he message o he Bila e alNego ia o ia in e ac ion Con e P o ocol, and he s a e machine en e s s a e gene a ing o eply he message ha was ecei ed. Howe e , i he nego ia ion message ecei ed is a ejec Nego ia ion message o he e is an e o in he p o ocol o in he communica ion link wi h he o he pa y, i en e s s a e inished and no i ies ha he nego ia ion has ailed o he Ba gainingCoo dina o . Gene a ing. In his s a e, he Bila e alNego ia o in e ac s wi h he MessageCompose by means o in e ac ion Reques Re- sponse o ob ain he nego ia ion message o be sen o he o he pa y. I he nego ia ion message gene a ed is a binding one, he Bila e alNego ia o i s o wa ds i o he Ba gainingCoo dina o ia in e ac ion Coo dina eNego- ia ion and en e s s a e asked app o al; o he wise, i sends he nego ia ion message o he o he pa y by means o he Ba gainingP o ocolHandle and en e s s a e wai ing. Asked app o al. In his s a e, he Bila e alNego ia o is wai ing o he app o al o he binding nego ia ion message. I he app o al is g an ed, he nego ia ion con ex en e s s a e app o ed; o he wise, i mo es back o s a e gene a ing. App o ed. I he binding nego ia ion message is app o ed, he nego ia ion con ex en e s his s a e and he binding nego ia- ion message is sen . In his case, i his is he las message o he nego ia ion p o ocol, e.g., i con ains pe o ma i e accep , i en e s s a e inished. O he wise, i wai s in his s a e un il he esponse o he o he pa y is ecei ed. I he o he pa y ejec s he binding nego ia ion message, i mo es back o s a e gene a ing; o he wise, i en e s s a e inished. M. Resinas e al. / Science o Compu e P og amming 77 (2012) 4–28 19 Fig. 14. Typical sequence diag am o in e ac ion Coo dina eNego ia ion wi h wo bila e al nego ia o s. Fig. 15. In e ac ion Submi Policies. message commi Rejec ed and he in e ac ion en e s back s a e nego ia ing. In addi ion, in s a es nego ia ing and wai ing esul , he Bila e alNego ia o may send message inishedUnsuccess ully, which causes he in e ac ion o inish. Simila ly, he Ba gainingCoo dina o may decide o cancel he nego ia ion by in oking me hod cancel on he Bila e al- Nego ia o . I his me hod is in oked in s a e nego ia ing o asking app o al, he in e ac ion inishes. Howe e , i he in e - ac ion is in s a e wai ing esul , i means ha a binding nego ia ion message was sen o he o he pa y. As a consequence, i me hod cancel is in oked in ha s a e, he in e ac ion mus no inish immedia ely, bu en e s a e cancelling; now, he Bila e alNego ia o can in oke ei he me hod inishedSuccess ully o me hod inishedUnsuccess ully. Fig. 14 depic s a sequence diag am o his in e ac ion wi h one Ba gainingCoo dina o (coo dina o ) and wo Bila e al- Nego ia o s (neg1 and neg2). In his sequence diag am, a e he ini ialisa ion, neg1 eques s an app o al, which is ejec ed by Coo dina o a e que ying he Commi Handle (no shown in his igu e). Nex , bo h neg1 and neg2 eques app o als. Coo dina o ejec s he eques om neg1 and accep s he eques om neg2. Finally, neg2 no i ies ha he nego ia ion inished success ully and Coo dina o cancels he nego ia ion o neg1. In e ac ion Submi Policies.This in e ac ion implemen s he submission o nego ia ion policies om he PoliciesManage , which implemen s in e ace IPoliciesManage , o he Bila e alNego ia o , which implemen s in e ace IPolicyRecei e . The in e ac ion p o ocol is simple, c . Fig. 15. When a new ba gaining nego ia ion s a s, he Bila e alNego ia o in okes me hod ini Nego ia ion on he PoliciesManage o no i y ha he e is a new ba gaining nego ia ion and o ecei e he ini ial se o nego ia ion policies. Then, when he PoliciesManage inds i app op ia e, i submi s a new se o nego ia ion policies by in oking me hod se Nego ia ionPolicies. Finally, when he nego ia ion con ex inishes, he Bila e alNego ia o in okes me hod endNego ia ion o no i y ha no new nego ia ion policies a e needed. In e ac ion Reques Response.The goal o his in e ac ion is o ob ain a nego ia ion message ha shall be sen as a esponse o he o he nego ia ing pa y. The in e ac ion is asynch onous and is depic ed in Fig. 16. I has wo pa icipan s: he Bila - e alNego ia o , which eques s he nego ia ion message and implemen s in e ace IResponseReques e , and he Message- Compose , which e u ns he gene a ed nego ia ion message and implemen s in e ace IResponseGene a o . The in e ac ion akes place as ollows: when he Bila e alNego ia o needs a nego ia ion message as esponse, i in okes me hod gene a eResponse on he MessageCompose oge he wi h in o ma ion abou he ba gaining pe o ma i es ha can be used in he nego ia ion message, he cu en s a us o he nego ia ion and a e e ence o he eques e ha shall ecei e 20 M. Resinas e al. / Science o Compu e P og amming 77 (2012) 4–28 Fig. 16. In e ac ion Reques Response. Fig. 17. In e ac ion Reques P oposal. he gene a ed nego ia ion message. Then, he MessageCompose c ea es he nego ia ion message and in okes me hod nego ia ionMessage on he Bila e alNego ia o wi h he gene a ed nego ia ion message. In addi ion, a any momen , he Bila e alNego ia o may cancel he gene a ion by in oking me hod cancelGene a ion. In ha case, he MessageCompose mus espond synch onously wi h a alid nego ia ion message o a null alue i no alid nego ia ion message could be gene a ed. In e ac ion Reques P oposal.The goal o his in e ac ion is o ob ain a p oposal ha shall be sen as a pa o a nego ia ion message o he o he nego ia ing pa y. This in e ac ion is e y simila o he p e ious one, he di e ence being ha he goal o he o me is o ob ain a whole nego ia ion message, i.e., a pe o ma i e and a p oposal, whe eas he goal o he la e is jus o ob ain he p oposal. The in e ac ion has wo pa icipan s, c . Fig. 17: he Builde Manage , which p o ides he p oposal and implemen s in e ace IBuilde Manage , and he MessageCompose , which eques s he p oposal and implemen s in e ace IP oposalReques e . The in e ac ion p o ocol is exac ly he same as he p e ious one. In e ac ion C ea eP oposal.The goal o his in e ac ion is o ob ain a p oposal ha shall be sen as pa o a nego ia ion message o he o he nego ia ing pa y. The di e ence is ha his in e ac ion allows a lowe -le el con igu a ion p io o he eques o he new p oposal. In addi ion, he pa icipan s a e also di e en . In his case, he P oposalBuilde is he c ea o o he p oposal and implemen s in e ace IP oposalBuilde , whe eas he Builde Manage is now he eques e o he p oposal and, hence, i implemen s in e ace IP oposalReques e , c . Fig. 18. The in e ac ion p o ocol is simila o he p e ious one, he di e ence being ha be o e in oking me hod gene a eP o- posal, he Builde Manage may con igu e he P oposalBuilde by in oking me hod con igu e. The main in en o his s ep is o con e nego ia ion policies in o builde -speci ic con igu a ion pa ame e s, and, hence, making independen he M. Resinas e al. / Science o Compu e P og amming 77 (2012) 4–28 21 Fig. 18. In e ac ion C ea eP oposal. P oposalBuilde o he nego ia ion policies de ined o an au oma ed nego ia ion sys em. Howe e , his s ep is op ional; i he Builde Manage does no con igu e he P oposalBuilde , i shall ake i s de aul con igu a ion. 5.4. Coo dina ion beha iou This laye p o ides he speci ica ion o he s a e machine o he Ba gainingCoo dina o and he Bila e alNego ia o , which coo dina es he o he oles o he a chi ec u e. This speci ica ion de ines he beha iou o he whole a chi ec u e. In his sec ion, we epo on he beha iou o he Bila e alNego ia o and he Ba gainingCoo dina o oles, which coo di- na e he o he s. Role Bila e alNego ia o .The goal o he Bila e alNego ia o is o ca y ou a single bila e al nego ia ion by o ches a ing he Ba gainingP o ocolHandle and he esponse gene a ion oles. Fu he mo e, i mus communica e wi h he Ba gainingCo- o dina o o coo dina e wi h he o he simul aneous ba gaining nego ia ions and o ask o app o al be o e commi ing o a p oposal, and wi h he PoliciesManage o ecei e nego ia ion policies ha shall guide he gene a ion o esponses. Fig. 19 depic s he s a e machine o he Bila e alNego ia o . I consis s o he ollowing s a es: ou s a es (wai ing ini ial, wai ing,wai ing accep and cancelling) in which he Bila e alNego ia o is wai ing o a message om he o he pa y, one s a e (gene a ing esponse) in which he Bila e alNego ia o is wai ing o he MessageCompose o gene a e a esponse, and wo s a es (app o ing commi and app o ing accep ) in which he Bila e alNego ia o is wai ing o he Commi Handle o decide on whe he o send a binding nego ia ion message o no . The s a e machine s a s when he Ba gainingCoo dina o ini ia es he Bila e alNego ia o by in oking me hod ini wi h he URI o he nego ia ion con ex and a e e ence o he Ba gainingP o ocolHandle ha manages he communica ion wi h he o he pa y. Then, he Bila e alNego ia o in okes me hod ini o he Ba gainingP o ocolHandle o ini ialise he in e ac- ion and wai s o i s esponse in s a e wai ing ini ial. The esponse can be s a Nego ia ion o a nego ia ion pe o ma i e. In any case, when he esponse is ecei ed he Bila e alNego ia o in okes me hod gene a eResponse on he MessageCom- pose and wai s o he esponse in s a e gene a ing esponse. In addi ion, in s a e wai ing ini ial, he Bila e alNego ia o may also ecei e message cancel, in which case, i en e s s a e cancelling. The Bila e alNego ia o lea es s a e gene a ing esponse when ei he he Ba gainingP o ocolHandle in okes me hod e - o , he Ba gainingCoo dina o in okes me hod cancel o he MessageCompose in okes me hod nego ia ionMessage. In he i s wo cases, he Bila e alNego ia o in okes me hod cancelGene a ion in he MessageCompose and me hod inishedUn- success ully in he Ba gainingCoo dina o and en e s s a e inished. In he hi d case, he ansi ion o he Bila e alNego ia o depends on he pe o ma i e o he gene a ed nego ia ion message: i he pe o ma i e is accep , i in okes me hod accep- App o alReques in he Ba gainingCoo dina o and en e s s a e app o ing accep ; i he pe o ma i e is ejec Nego ia ion, i in okes me hod ejec Nego ia ion in he Ba gainingP o ocolHandle and me hod inishedUnsuccess ully in he Ba gaining- Coo dina o and mo es o s a e inished; i he pe o ma i e is commi , i in okes me hod commi App o alReques in he Ba gainingCoo dina o and en e s s a e app o ing commi ; and i he pe o ma i e is ei he p opose,wi hd aw o ejec P o- posal, i in okes he co esponding me hod in he Ba gainingP o ocolHandle and en e s s a e wai ing. In s a e app o ing accep , he Bila e alNego ia o wai s o an app o e o ejec message. I he message is app o e, he Bila e alNego ia o in okes me hod accep in he Ba gainingP o ocolHandle and me hod inishedSuccess ully in he Ba gainingCoo dina o and en e s s a e inished. I he message is ejec , he Bila e alNego ia o in okes again me hod gene a eResponse in he MessageCompose and mo es back o s a e gene a ing esponse o wai o ano he esponse. 22 M. Resinas e al. / Science o Compu e P og amming 77 (2012) 4–28 Fig. 19. S a e machine o he Bila e alNego ia o . In addi ion, he Bila e alNego ia o may also ecei e message cancel o message e o . I his is he case, i in okes me hod ejec Nego ia ion in he Ba gainingP o ocolHandle and me hod inishedUnsuccess ully in he Ba gainingCoo dina o and en e s s a e inished. Simila ly, in s a e app o ing commi , he Bila e alNego ia o wai s o he Ba gainingCoo dina o o espond wi h an app o e o ejec message. I he esponse is app o e, he Bila e alNego ia o in okes me hod commi in he Ba gainingP o- ocolHandle and en e s s a e wai ing accep . I he esponse is ejec , he Bila e alNego ia o beha es as in s a e app o ing accep : i in okes me hod gene a eResponse and en e s s a e gene a ing esponse. In addi ion, he Bila e alNego ia o may also ecei e message cancel o message e o , in which case i beha es as in s a e app o ing accep . In s a es wai ing and wai ing accep , he Bila e alNego ia o wai s o a new nego ia ion message. The di e ence being ha in s a e wai ing accep , he Bila e alNego ia o has sen a binding nego ia ion message and i is wai ing o i s accep ance whe eas in s a e wai ing, i has no . In bo h s a es, i he nego ia ion message is ejec Nego ia ion o e o , he Bila e alNego- ia o in okes me hod inishedUnsuccess ully in he Ba gainingCoo dina o and en e s s a e inished. I he nego ia ion mes- M. Resinas e al. / Science o Compu e P og amming 77 (2012) 4–28 23 Table 3 Requi emen s co e ed by he scena ios. Scena ios Requi emen 1.1 2.1 2.3 2.4 4.1 4.2 SCEN1: A compu ing job submission sys em ha nego ia es wi h se e al IaaS p o ide s. ✓✓✓✓✓✓ SCEN2: An IaaS p o ide ha nego ia es wi h se e al consume s. ✓ ✓ ✓ ✓ SCEN3: A mee ing schedule nego ia ion sys em. ✓ ✓ 1.1 Suppo mul i- e m nego ia ion p o ocols. 2.4 Allow use p e e ences abou nego ia ion p ocesses. 2.1 Mul iple p o ocol suppo . 4.1 Suppo se e al nego ia ions simul aneously. 2.3 Mul iple decision-making algo i hms. 4.2 Selec decision-making algo i hms dynamically. sage is ei he c p,commi ,p opose, ejec P oposal o wi hd aw, he Bila e alNego ia o in okes me hod gene a eResponse in he MessageCompose and en e s s a e gene a ing esponse. In his case, i he Bila e alNego ia o is in s a e wai ing accep , i also in okes me hod commi Rejec ed in he Ba gainingCoo dina o . In s a e wai ing accep , he nego ia ion message may also be accep , in which case i in okes me hod inishedSuccess ully in he Ba gainingCoo dina o and en e s s a e inished. Be- sides a nego ia ion message, he Bila e alNego ia o may also ecei e message cancel, in which case, i en e s s a e cancelling. In s a e cancelling, he nego ia ion has been cancelled bu by he Ba gainingCoo dina o , bu he o he pa y has no been no i ied because he Bila e alNego ia o was wai ing o a nego ia ion message om he o he pa y, i.e., message ejec Ne- go ia ion has no been sen . When his nego ia ion message a i es, i i is accep , he Bila e alNego ia o in okes me hod inishedSuccess ully in he Ba gainingCoo dina o and en e s s a e inished. O he wise, he Bila e alNego ia o in okes bo h me hods ejec Nego ia ion and inishedUnsuccess ully in he Ba gainingP o ocolHandle and he Ba gainingCoo dina o , e- spec i ely, and en e s s a e inished. In addi ion, a any momen , he PoliciesManage may send nego ia ion policies o he Bila e alNego ia o (se Nego ia ionPolicy). Role Ba gainingCoo dina o .The Ba gainingCoo dina o is a dispa che ha coo dina es he messages exchanged amongs he Bila e alNego ia o , he Commi Handle , he Pa yCoo dina o and he Ba gainingCon ex Da a. I s asks include ini ialising and inalising nego ia ions, as well as coo dina ing he app o al eques s be ween he Commi Handle and he Bila e alNego ia o s. Rega ding he i s ask, whene e a Pa yCoo dina o in okes me hod nego ia e, i c ea es a new nego ia ion con ex and in okes me hod ini on a Bila e alNego ia o . Simila ly, when a Bila e alNego ia o in okes ei he me hod ail o me hod succeed, i o wa ds hem o he Pa yCoo dina o . Finally, i he Pa yCoo dina o in okes he cancel me hod, i o wa ds i o he co esponding Bila e alNego ia o . The second ask in ol es coo dina ing he app o al o eques s: he Ba gainingCoo dina o ecei es commi app o al e- ques s and accep app o al eques s om he Bila e alNego ia o s. Fo each o hem, i o wa ds hem o he Commi Handle by in oking he app o alReques me hod. Simila ly, i o wa ds messages accep and ejec om he Commi Handle o he co esponding Bila e alNego ia o s. Finally, i o wa ds he commi Rejec ed o success message om he Bila e alNego ia o o he Commi Handle . 6. The NegoFAST-Ba gaining scena ios iew Acco ding o K uch en [22], he scena ios iew p o ides addi ional de ails on how an a chi ec u e is ins an ia ed in ypical cases, which se es wo pu poses: on he one hand, i illus a es he a chi ec u e and shows how i s accompanying so wa e amewo k can be used o build au oma ed ba gaining nego ia ion sys ems; on he o he hand, i alida es he p oposal because, in a sense, he scena ios a e an abs ac ion o he mos impo an equi emen s. We ha e de eloped h ee scena ios, namely: SCEN1 and SCEN2 a e au oma ed nego ia ion sys ems in an In as uc u e- as-a-Se ice (IaaS) con ex om he pe spec i e o he consume and p o ide , espec i ely; SCEN3 ocuses on he implemen a ion o an au oma ed nego ia ion sys em o schedule mee ings by means o he mul i-agen nego ia ions desc ibed by Waine e al. [41]. Table 3 summa ises he equi emen s co e ed by each scena io; no e, oo, ha hese scena ios helped us check desi able non- unc ional p ope ies since SCEN1 and SCEN2 we e chosen o es he eusabili y o he a chi ec u e and SCEN3 was chosen o es i s adap abili y o new si ua ions. Implemen ing hese scena ios equi ed us o make h ee g oups o ela ed decisions, namely: Nego ia ion p o ocol: The i s s ep is o decide which he mos app op ia e p o ocols a e. To make his decision, we mus pay a en ion o he exp essi eness o he p oposals o which i allows, he es ic ions ha i poses on he con en s o he p oposals and he pe o ma i es i allows. In addi ion, he nego ia ion p o ocol also has an in luence on he decision-making algo i hms. Nego ia ion policies: The second s ep is o decide on he nego ia ion policies ha a e used by he au oma ed nego ia ion sys ems. These policies a e used o allow use p e e ences abou he nego ia ion p ocess and o enable ad anced coo dina ion mechanisms o concu en ba gaining nego ia ions. Decision making: The las s ep is o decide on he algo i hms o implemen he decisions ega ding which esponses mus be c ea ed du ing a nego ia ion. The e a e wo di e en decisions: selec ing he pe o ma i e o be used and c ea ing 24 M. Resinas e al. / Science o Compu e P og amming 77 (2012) 4–28 Fig. 20. In as uc u e-as-a-Se ice scena io. he p oposal o be sen i necessa y. Bo h decisions a e in luenced by he ba gaining p o ocol because he use o a pa icula algo i hm may depend on he ea u es o a conc e e ba gaining p o ocol. I is also in luenced by he nego ia ion policies because he decision-making algo i hm mus be con igu able enough o suppo he deg ee o con ol se by he nego ia ion policies. Nex , we p esen u he de ails on hese scena ios and p o ide addi ional de ails ega ding he p e ious decisions. 6.1. SCEN1: compu ing job submission This scena io ocuses on he submission o a compu ing job o an IaaS p o ide . In his con ex , se ice ag eemen s mus be c ea ed amongs job submi e s and IaaS p o ide s o se he e ms unde which jobs shall be execu ed. These e ms may include de ails such as he na u e o he p ocess o be execu ed, he esou ces equi ed o he execu ion o scheduling equi emen s like job s a o job comple ion deadlines [1]. Fig. 20 depic s a communica ion diag am ha o e iews he whole scena io: i s , he job submi e sends i s ag eemen p e e ences o i s au oma ed nego ia ion sys em. These p e e ences may include bo h equi emen s abou he job execu ion and guidelines ega ding he nego ia ion p ocess. Then, when he au oma ed nego ia ion sys em ecei es a numbe o e e ences o IaaS p o ide s, i s a s a ba gaining nego ia ion wi h hem. When an ag eemen is made, he au oma ed nego ia ion sys em no i ies he job submi e and sends he ag eemen o i . Finally, he job submi e sends he job o he IaaS p o ide , which execu es i ollowing he e ms es ablished in he ag eemen . Nex we p o ide addi ional de ails on he h ee g oups o decisions made o implemen his scena io: Nego ia ion p o ocol: In his scena io, we ha e selec ed he mul i- e m nego ia ion p o ocol desc ibed by Fa a in e al. [10] (REQ 1.1), which is a ba gaining nego ia ion p o ocol in which bo h pa ies exchange binding p oposals un il an ag eemen is made o one pa y abandons he nego ia ion. The implemen a ion o his nego ia ion p o ocol is encapsula ed in componen Fa a inP o ocolHandle , which plays ole Ba gainingP o ocolHandle o deal wi h pa icula ba gaining p o ocols. Nego ia ion policies: In his scena io, ou policies a e de ined o con ol he nego ia ion p ocess (REQ 2.4): nego ia ion deadline, numbe o ag eemen s o each, eage ness o each an ag eemen , and he minimum u ili y h eshold. These policies a e managed by componen PoliciesManage , which se s he alues o hese policies acco ding o he use p e e ences and he s a e o cu en nego ia ions (REQ 4.1). Decision making: Fig. 21 depic s he so wa e a i ac s ha implemen he decision making oles in SCEN1. The i s decision, i.e., selec ing he pe o ma i e, is encapsula ed in componen MessageCompose , speci ically in class Pe o ma i eSelec o . I is designed o selec pe o ma i e accep i he u ili y o he ecei ed p oposal exceeds a use -de ined h eshold. O he wise, i selec s pe o ma i e commi . The second decision is ca ied ou by he Builde Manage , which elies on wo addi ional componen s o c ea e p oposals, namely: componen Ini ialBuilde , which c ea es a p oposal by selec ing he alues ha maximise he u ili y, and componen NDFBuilde , which implemen s he decision-making algo i hms p oposed by Fa a in e al. [10] based on nego ia ion decision unc ions and wo ac ics: ime ac ic and beha iou ac ic. The Builde Manage selec s Ini ialBuilde a he beginning o he nego ia ion and, as he nego ia ion goes on, i s a s selec ing an NDFBuilde (REQ 4.2). Fu he mo e, i also con igu es he P oposalBuilde s wi h he policies p o ided by he PoliciesManage (REQ 2.4). M. Resinas e al. / Science o Compu e P og amming 77 (2012) 4–28 25 Fig. 21. Implemen a ion o decision making oles in SCEN1. We mus also p o ide an implemen a ion o ole Commi Handle o app o e he submission o all binding nego ia ion messages. In his scena io, he implemen a ion is based on decision poin s. Decision poin s ake place when a gi en e en occu s, e.g., when he numbe o binding nego ia ion messages wai ing o app o al exceeds a ce ain h eshold o he nego ia ion deadline is close. When a decision poin akes place, some binding nego ia ion messages a e app o ed, whe eas o he s a e ejec ed. 6.2. SCEN2: in as uc u e-as-a-Se ice P o ide This scena io uses he same con ex as he p e ious scena io, bu i ocuses on he IaaS p o ide ’s pe spec i e. No e ha since he con ex is he same as in he p e ious scena io, i is no su p ising ha bo h au oma ed nego ia ion sys ems sha e many componen s. The e o e, his scena io helps illus a e he eusabili y o he elemen s o NegoFAST-Ba gaining by analysing he changes ha ha e o be made o he au oma ed nego ia ion sys em desc ibed in he p e ious scena io o adap o he new one. A desc ip ion o he h ee g oups o decisions made o implemen his scena io ollows: Nego ia ion p o ocol: Since he au oma ed nego ia ion sys em is going o nego ia e in he same con ex as be o e, we can keep he same nego ia ion p o ocol. The e o e, he implemen a ion o he Ba gainingP o ocolHandle can be eused (REQ 1.1). Nego ia ion policies: Rega ding he nego ia ion policies, h eshold, deadline and eage ness can be eused since hey a e also use ul om he poin o iew o he IaaS p o ide (REQ 2.4). The only conside a ion is he policy ega ding he numbe o ag eemen s, since a p o ide has no es ic ions on he numbe o ag eemen s i can each p o ided ha i has enough esou ces o comply wi h hem. The e o e, he PoliciesManage can be eused, excep o he ac ha i should upda e he policy ega ding he numbe o ag eemen s wi h an es ima ion based on he a ailabili y o he p o ide ’s esou ces. Decision making: Finally, his scena io uses a new decision-making model inspi ed by Nash’s ba gaining solu ion [27] ha does no impose any es ic ions on he numbe o ag eemen s ha can be eached a p io i. This causes ha new implemen a ions o se e al decision-making oles mus be p o ided, namely: he implemen a ion o he Pe o ma i eSelec o mus change because he c i e ion o selec pe o ma i e accep is no a use -de ined h eshold, bu depends on he p oposal ecei ed and i s po en ial esponse; a new Builde Manage mus be de eloped o use he Nash-inspi ed builde ins ead o he NDFBuilde (REQ 2.3), and he implemen a ion o he Builde Manage mus change o use he Nash-inspi ed builde ins ead o he nego ia ion decision unc ions builde . Fu he mo e, a new Commi Handle mus be de eloped so ha he commi men o an ag eemen does no depend on a s a ic numbe o ag eemen s ha can be eached, bu on he a ailabili y o he p o ide ’s esou ces. 6.3. SCEN3: scheduling mee ings This scena io ocuses on he implemen a ion o a mechanism o schedule mee ings by means o he mul i-agen nego ia ions desc ibed by Waine e al. [41]. The ea u es o he au oma ed nego ia ion sys em used in his scena io a e 26 M. Resinas e al. / Science o Compu e P og amming 77 (2012) 4–28 signi ican ly di e en om he p e ious ones since he e a e mul iple pa icipan s in he nego ia ion and nego ia ions a e ca ied ou one a a ime. The e o e, his scena io helps alida e he adap abili y o NegoFAST-Ba gaining o new si ua ions. Nex we epo on he h ee g oups o decisions ela ed o his scena io: Nego ia ion p o ocol: The sugges ions p o ocol by Waine e al. [41] is a signi ican ly di e en nego ia ion p o ocol (REQ 2.1). I allows he in e ac ion o mul iple pa icipan s in a nego ia ion, one o each pe son in i ed o a mee ing. This in e ac ion is done by means o a hos ha ecei es nego ia ion messages om all pa icipan s and b oadcas s hem o he es . The implemen a ion o his p o ocol is encapsula ed in componen Sugges ionsP o ocolHandle , which ans o ms i in o he gene ic nego ia ion p o ocol de ailed in Sec ion 3so ha he conc e e de ails o he p o ocol a e anspa en o he es o he au oma ed nego ia ion sys em. Nego ia ion policies: In his scena io, he e is only one mee ing nego ia ion a a ime. Consequen ly, he e is no need o an ad anced coo dina ion o au oma ed nego ia ions and he nego ia ion policies can be educed o only one: in o ma ion, which se s he amoun o in o ma ion abou he own schedule ha he sys em is willing o p o ide. Decision making: Rega ding he decision-making algo i hm o c ea e p oposals, Waine e al. [41] de ailed h ee algo i hms called ego is ic, laconic, and decei ing. Each o hem o e s a di e en ade-o be ween he amoun o in o ma ion p o ided o he o he pa icipan s abou i s own schedule and he bes esul in he nego ia ion. These algo i hms a e implemen ed in h ee di e en componen s ha play ole P oposalBuilde (REQ 2.3) and he Builde Manage selec s one amongs hem based on he policies p o ided by he use in his o he p e e ences. 7. Conclusions and u u e wo k In his a icle, we ha e ackled he p oblem o building au oma ed se ice ag eemen nego ia ion sys ems ha ely on a ba gaining p o ocol and wo k in open en i onmen s. In Re . [32], we iden i ied a numbe o key equi emen s o such sys ems, some o which a e p o ocol-dependen ; ou ocus in his pape has been on his subse o equi emen s. We ha e su eyed he cu en li e a u e on au oma ic nego ia ion, and ou conclusion was ha none o he cu en s a e-o - he-a p oposals suppo all o he p o ocol-dependen equi emen s oge he , which mo i a ed us o wo k on a new p oposal called NegoFAST-Ba gaining. Ou analysis o he ela ed wo k also e ealed a ew weaknesses o he exis ing li e a u e; o ins ance, he e exis s a ple ho a o p oposals o c ea e P oposalBuilde s and P o ocolHandle s [10,11,21,25,39]; con a ily, he e a e e y ew p oposals o c ea e Builde Manage s [34]. Ou a chi ec u e has been p esen ed acco ding o he guidelines ha we e es ablished by K uch en [22]. We ha e p esen ed a logical iew, a p ocess iew, a de elopmen iew, and a scena ios iew. The logical iew iden i ies he unc ional a chi ec u al elemen s o ou p oposal, and helps de ine a ocabula y ha b idges he gap be ween he many di e en e minologies used in his ield. The p ocess iew iden i ies how he a chi ec u al elemen s can be g ouped oge he in o p ocesses; we ha e discussed wo di e en o ganisa ions o his iew. The de elopmen iew includes a so wa e amewo k ha p o ides a e e ence implemen a ion o ou p oposal and can, hus, be seen as a s a ing poin o de elope s who need o c ea e hei own nego ia ion sys ems. The scena ios iew illus a es he a chi ec u e in ypical cases and also helped us alida e i . Fu u e esea ch pa hs include he ollowing: allowing o in e -dependen nego ia ions and allowing o p o ocols ha a e no based solely on p oposals. (No e ha ex ending NegoFAST-Co e o suppo auc ioning, o ins ance, is also qui e in e es ing, bu i is a esea ch line in i s own. The p e ious p oposals a e esea ch pa hs ha would help enhance he esul s p esen ed in his a icle.) No e ha so-called composi e web se ices a e gaining impo ance as languages such as BPEL a e becoming mo e and mo e mains eam. The p oblem wi h such composi e se ices is ha nego ia ing hei componen s may equi e nego ia ions o be in e -dependen . Al hough he cu en e sion o NegoFAST-Ba gaining p o ides a li le suppo o such nego ia ions by means o ou policy- ela ed oles, he p oblem has no been s udied in i s ull ex en . Fu he mo e, suppo ing p o ocols ha a e no based on p oposals only would equi e new esponse gene a ion elemen s o c ea e his addi ional in o ma ion and changes o he da a model. Acknowledgemen s The wo k on which his pape epo s was pa ially unded by he Eu opean Commission (FEDER), he Spanish Minis y o Science and Inno a ion, and he Andalusian Go e nmen . The wo k by MR and PF was suppo ed by g an s TIN2009- 07366 (SETI) and P07-TIC-2533 (Isabel); he wo k by RC was suppo ed by g an s TIN2010-21744-C02, TIN2007-64119, P07-TIC-02602, P08-TIC-4100, and TIN2008-04718-E (In eg aWeb). Appendix. An o e iew o NegoFAST-Co e This appendix p o ides a sho in oduc ion o NegoFAST-Co e, which cons i u es he con ex o ou wo k. A de ailed p esen a ion o NegoFAST-Co e was published in Re . [32]. Fig. 22 p o ides an o e all pic u e o he ou o ganisa ions in o which a nego ia ion sys em is decomposed, namely: p o ocol managemen , coo dina ion, wo ld modelling, and decision making, which a e ep esen ed as la ge whi e boxes M. Resinas e al. / Science o Compu e P og amming 77 (2012) 4–28 27 Fig. 22. The NegoFAST-Co e e e ence a chi ec u e. in he middle o he igu e, and se en en i onmen al esou ces, which a e depic ed as small boxes a he bo om o he igu e. The ex e nal en i ies a e depic ed as da k g ey boxes a he op and he le o he igu e. Roles a e depic ed as boxes inside he o ganisa ions. They ep esen a e ac s ha model he wo ld (Wo ldModelle ), ga he and p o ide in o ma ion om and o o he pa ies (Inqui e and In o man ), in e ac wi h he use (Sys emCoo dina o ), decide on he conc e e nego ia ion p o ocol o be used (P o ocolNego ia o ), handle he nego ia ion p o ocol (P o ocolHandle ), gene a e nego ia ion messages du ing a nego ia ion (ResponseGene a o ), decide whe he o commi o o accep an ag eemen p oposal (Commi Handle and Commi Ad iso ), and coo dina e hem all (Sys emCoo dina o ,Pa yCoo dina o and Nego ia ionCoo dina o ). NegoFAST-Co e also includes en i onmen al esou ces ha s o e he da a ha a e equi ed by he p e ious oles du ing a nego ia ion p ocess, namely: Ag eemen sResou ce,P e e encesResou ce,Sys emCon ex Da a, Pa yCon ex Da a,Nego ia ionCon ex Da a,Wo ldModel and Nego ia ionHis o y. The majo i y o he p e ious a chi ec u al elemen s a e p o ocol independen , wi h he excep ion o he ollowing, which a e deal wi h by NegoFAST-Ba gaining: P o ocolHandle s o p o ide suppo o se e al mul i- e m nego ia ion p o ocols; Nego ia ionCoo dina o o suppo se e al nego ia ions simul aneously; ResponseGene a o s o decide which is he mos app op ia e message o answe an incoming nego ia ion message, and Nego ia ionCon ex Da a o s o e in o ma ion ela ed o he nego ia ions ha a e being ca ied ou by he au oma ed nego ia ion sys em. In addi ion o hese a chi ec u al elemen s, he e is a decision ha , despi e being p o ocol-independen , has an in luence on he p o ocol-speci ic a chi ec u e: he decision on sending a binding nego ia ion message o no . Depending on he pe o ma i e, nego ia ion messages can be classi ied as binding nego ia ion messages, which in ol e a i m commi men wi h he o he pa y, and non-binding nego ia ion messages, which do no in ol e such a i m commi men . This decision is made by he Commi Handle . As a consequence, he Nego ia ionCoo dina o may send as many non-binding nego ia ion messages as necessa y; bu , i needs he app o al o he Commi Handle be o e sending a binding nego ia ion message. The e o e, al hough he Commi Handle is a p o ocol-independen ole, he Nego ia ionCoo dina o mus in e ac wi h i in o de o ca y ou a nego ia ion. Fu he mo e, he gene a ion o esponses mus be p o ided wi h he use ’s p e e ences, he 28 M. Resinas e al. / Science o Compu e P og amming 77 (2012) 4–28 wo ld model, he nego ia ion his o y, and o he aspec s ela ed o he cu en s a e o he au oma ed nego ia ion sys em. Such in o ma ion is s o ed in he a o emen ioned en i onmen al esou ces. Re e ences [1] A. And ieux, K. Czajkowski, A. Dan, K. Keahey, H. Ludwig, T. Naka a, J. P uyne, J. Ro ano, S. Tuecke, M. Xu, WS-Ag eemen speci ica ion, 2007. URL: h p://www.og .o g/documen s/GFD.107.pd . [2] R. Ash i, I. Rahwan, M. Luck, A chi ec u es o nego ia ing agen s, in: V. Ma ík, J.P. Mülle , M. Pechoucek (Eds.), Mul i-Agen Sys ems and Applica ions III: 3 d In e na ional Cen al and Eas e n Eu opean Con e ence on Mul i-Agen Sys ems, in: Lec u e No es in Compu e Science, ol. 2691, Sp inge , 2003, pp. 136–146. [3] C. Ba olini, C. P eis , N.R. Jennings, A so wa e amewo k o au oma ed nego ia ion, in: R. Cho en, A. Ga cía, C. Lucena, A. Ramono sky (Eds.), So wa e Enginee ing Fo Mul i-Agen Sys ems III: Resea ch Issues and P ac ical Applica ions, in: Lec u e No es in Compu e Science, ol. 3390, Sp inge , 2005, pp. 213–235. [4] L. Bass, P. Clemens, R. Ka zman, So wa e A chi ec u e in P ac ice, Addison-Wesley, 2003. [5] D. Ba é, O. Wäld ich, W. Ziegle , F. B azie , K. Cla k, M. Oey, A. Papaspy uo, P. Wiede , A p oposal o WS-Ag eemen nego ia ion, in: P oceedings o 11 h ACM/IEEE In e na ional Con e ence on G id Compu ing, G id 2010, IEEE, 2010, pp. 233–241. [6] M. Benyouce , M.-H. Ve ons, Con igu able e-nego ia ion sys ems o la ge scale and anspa en decision making, G oup Decision and Nego ia ion 17 (3) (2008) 211–224. [7] M. Bichle , An expe imen al analysis o mul i-a ibu e auc ions, Decision Suppo Sys ems 29 (3) (2000) 249–268. [8] A. El a a y, P. Layzell, Nego ia ing in se ice-o ien ed en i onmen s, Communica ions o he ACM 47 (8) (2004) 103–108. [9] A. El a a y, P.J. Layzell, A nego ia ion desc ip ion language, So wa e, P ac ice and Expe ience 35 (4) (2005) 323–343. [10] P. Fa a in, C. Sie a, N.R. Jennings, Nego ia ion decision unc ions o au onomous agen s, In e na ional Jou nal o Robo ics and Au onomous Sys ems 24 (3–4) (1998) 159–182. [11] P. Fa a in, C. Sie a, N.R. Jennings, Using simila i y c i e ia o make ade-o s in au oma ed nego ia ions, A i icial In elligence 142 (2) (2002) 205–237. [12] S.S. Fa ima, M. Woold idge, N.R. Jennings, An agenda-based amewo k o mul i-issue nego ia ion, A i icial In elligence 152 (1) (2004) 1–45. [13] M. Fowle , Pa e ns o En e p ise Applica ion A chi ec u e, Addison-Wesley Longman Publishing Co., Inc., 2002. [14] H. Gimpel, H. Ludwig, A. Dan, B. Kea ney, PANDA: Speci ying policies o au oma ed nego ia ions o se ice con ac s, in: M.E. O lowska, S. Wee awa ana, M.P. Papazoglou, J. Yang (Eds.), Fi s In e na ional Con e ence on Se ice-O ien ed Compu ing, ICSOC 2003, in: Lec u e No es in Compu e Science, ol. 2910, Sp inge -Ve lag, 2003, pp. 287–302. [15] M. He, N.R. Jennings, H.-F. Leung, On agen -media ed elec onic comme ce, IEEE T ansac ions on Knowledge and Da a Enginee ing 15 (4) (2003) 985–1003. [16] N.R. Jennings, P. Fa a in, A.R. Lomuscio, S. Pa sons, M. Woold idge, C. Sie a, Au oma ed nego ia ion: p ospec s, me hods and challenges, G oup Decision and Nego ia ion 10 (2) (2001) 199–215. [17] C. Jonke , V. Robu, J. T eu , An agen a chi ec u e o mul i-a ibu e nego ia ion using incomple e p e e ence in o ma ion, Au onomous Agen s and Mul i-Agen Sys ems 15 (2) (2007) 221–252. [18] A.H. Ka p, Rules o engagemen o au oma ed nego ia ion, in: B. Bena allah, C. Goda (Eds.), Fi s IEEE In e na ional Wo kshop on Elec onic Con ac ing, IEEE Compu e Socie y, 2004, pp. 32–39. [19] J.B. Kim, A. Sege , A web se ices-enabled ma ke place a chi ec u e o nego ia ion p ocess managemen , Decision Suppo Sys ems 40 (1) (2005) 71–87. [20] M. Klein, P. Fa a in, H. Sayama, Y. Ba -Yam, P o ocols o nego ia ing complex con ac s, IEEE In elligen Sys ems 18 (6) (2003) 32–38. [21] R. Kowalczyk, Fuzzy e-nego ia ion agen s, So Compu ing 6 (5) (2002) 337–347. [22] P. K uch en, The iew model o a chi ec u e, IEEE So wa e 12 (6) (1995) 42–50. [23] G. Lodi, F. Panzie i, D. Rossi, E. Tu ini, SLA-D i en clus e ing o qos-awa e applica ion se e s, IEEE T ansac ions on So wa e Enginee ing 33 (3) (2007) 186–197. [24] A. Ludwig, P. B aun, R. Kowalczyk, B. F anczyk, A amewo k o au oma ed nego ia ion o se ice le el ag eemen s in se ices g ids, in: Business P ocess Managemen Wo kshops, in: Lec u e No es in Compu e Science, ol. 3812, Sp inge , 2006, pp. 89–101. [25] X. Luo, N.R. Jennings, N. Shadbol , H.-F. Leung, J.H. Lee, A uzzy cons ain based model o bila e al, mul i-issue nego ia ions in semi-compe i i e en i onmen s, A i icial In elligence 148 (1–2) (2003) 53–102. [26] C. Molina-Jimenez, J. P uyne, Aad an Moo sel, The ole o ag eemen s in i managemen so wa e, in: R. de Lemos, C. Gacek, A. Romano sky (Eds.), A chi ec ing Dependable Sys ems III, in: Lec u e No es in Compu e Science., ol. 3549, Sp inge , 2005, pp. 36–58. [27] J.F. Nash, The ba gaining p oblem, Econome ica 18 (2) (1950) 155–162. [28] T.D. Nguyen, N.R. Jennings, Managing commi men s in mul iple concu en nego ia ions, Elec onic Comme ce Resea ch and Applica ions 4 (4) (2005) 362–376. [29] M.P. Papazoglou, The challenges o se ice e olu ion, in: Z. Bellahsene, M. Léona d (Eds.), 20 h In e na ional Con e ence on Ad anced In o ma ion Sys ems Enginee ing, CAiSE 2008, in: Lec u e No es in Compu e Science, ol. 5074, Sp inge , 2008, pp. 1–15. [30] S. Pau obally, V. Tamma, M. Woold dige, A amewo k o web se ice nego ia ion, ACM T ansac ions on Au onomous and Adap i e Sys ems 2 (4) (2007) 14. [31] I. Rahwan, S.D. Ramchu n, N. R Jennings, P. McBu ney, S. Pa sons, L. Sonenbe g, A gumen a ion-based nego ia ion, The Knowledge Enginee ing Re iew 18 (4) (2003) 343–375. [32] M. Resinas, P. Fe nandez, R. Co chuelo, Au oma ic se ice ag eemen nego ia o s in open comme ce en i onmen s, In e na ional Jou nal o Elec onic Comme ce 14 (3) (2010) 93–128. [33] S. Rinde le, M. Benyouce , Towa ds he au oma ion o e-nego ia ion p ocesses based on web se ices, in: A.H. Ngu, M. Ki su egawa, E.J. Neuhold, J.-Y. Chung, Q.Z. Sheng (Eds.), Six h In e na ional Con e ence on Web In o ma ion Sys ems Enginee ing, in: Lec u e No es in Compu e Science, ol. 3806, Sp inge , 2005, pp. 443–453. [34] R. Ros, C. Sie a, A nego ia ion me a s a egy combining ade-o and concession mo es, Au onomous Agen s and Mul i-Agen Sys ems 12 (2) (2006) 163–181. [35] T.W. Sandholm, V.R. Lesse , Le eled commi men con ac s and s a egic b each, Games and Economic Beha io 35 (1–2) (2001) 212–270. [36] K.M. Sim, G id esou ce nego ia ion: su ey and new di ec ions, IEEE T ansac ions on Sys ems, Man, and Cybe ne ics, Pa C: Applica ions and Re iews 40 (3) (2010) 245–257. [37] K.M. Sim, S.Y. Wang, Flexible nego ia ion agen wi h elaxed decision ules, IEEE T ansac ions on Sys ems, Man and Cybe ne ics, Pa B 34 (3) (2004) 1602–1608. [38] M. S öbel, Design o oles and p o ocols o elec onic nego ia ions, Elec onic Comme ce Resea ch 1 (3) (2001) 335–353. [39] S.Y. Su, C. Huang, J. Hamme , Y. Huang, H. Li, L. Wang, Y. Liu, C. Pluempi iwi iyawej, M. Lee, H. Lam, An in e ne -based nego ia ion se e o e- comme ce, The In e na ional Jou nal on Ve y La ge Da a Bases 10 (1) (2001) 72–90. [40] T. Tu, C. Seebode, F. G i el, W. Lame sdo , DynamiCS: an ac o -based amewo k o nego ia ing mobile agen s, Elec onic Comme ce Resea ch 1 (1–2) (2001) 101–117. [41] J. Waine , P.R. Fe ei a, E.R. Cons an ino, Scheduling mee ings h ough mul i-agen nego ia ions, Decision Suppo Sys ems 44 (1) (2007) 285–297. [42] F. Zambonelli, N.R. Jennings, M. Woold idge, De eloping mul iagen sys ems: he Gaia me hodology, ACM T ansac ions on So wa e Enginee ing and Me hodology 12 (3) (2003) 317–370. [43] D. Zeng, K. Syca a, Bayesian lea ning in nego ia ion, In e na ional Jou nal Human–Compu e S udies 48 (1) (1998) 125–141.