Automating the Procurement of Web Services

Au oma ing he P ocu emen o Web Se ices ? Oc a io Ma ín-Díaz, An onio Ruiz-Co és, Amado Du án, Da id Bena ides, Miguel To o Dp o. de Lenguajes y Sis emas In o má icos E.T.S. de Ingenie ía In o má ica, Uni e sidad de Se illa 41012 Se illa, España - Spain - Phone: +34 95 455 3871 Fax: +34 95 455 7139 - [email p o ec ed], [email p o ec ed], [email p o ec ed], [email p o ec ed], [email p o ec ed] Abs ac . As go e nmen agencies and business become mo e dependen on web se ices, so wa e solu ions o au oma e hei p ocu emen gain impo ance. Cu en app oaches o au oma ing he p ocu emen o web se ices su e om an impo an d awback: nei he unce ain y measu es no non-linea , and com- plex ela ions among pa ame e s can be used by p o ide s o speci y quali y-o - se ice in o e s. In his pape , we look deeply in o he oo s o his d awback and p esen a p oposal which o e comes i . The key poin o achie e his im- p o emen has been using he cons ain p og amming as a o mal basis, since i endows he model wi h a e y powe ul exp essi eness. A XML-based imple- men a ion is p esen ed along wi h some expe imen al esul s and compa isons wi h o he app oaches. Keywo ds so wa e p ocu emen , web se ices, quali y-o -se ice, ade s. 1 In oduc ion As go e nmen agencies and business become mo e dependen on web se ices, so - wa e solu ions o au oma e hei p ocu emen gain impo ance. I is gene ally assumed ha decision c i e ia o choosing so wa e packages s ems om he use equi emen s hey should ul ill. The e a e di e en ypes o equi emen s such as manage ial, po- li ical, and, o cou se, quali y equi emen s. The e a e a numbe o app oaches which au oma e some ac i i ies o he p ocu emen , mos o hem ocus in quali y equi e- men s. Howe e , hese app oaches su e om se e al d awbacks ha hampe hei use when equi emen s ha p o ide s gua an ee include unce ain y measu es, non-linea and complex ela ions among pa ame e s. In ac , i we wan o achie e a compe i i e echnology based on web se ices, hei quali y-o -se ice is an impo an issue o be aken in o accoun , becoming one o challenges o be sol ed in he nea u u e [31]. ? Suppo ed by he Spanish In e minis e ial Commission on Science and he Spanish Minis y o Science and Technology unde g an s TIC2000-1106-C02-01, TIC2003-02737-C02-01 and FIT-150100-2001-78. In his con ex , so wa e p ocu emen [4,5] becomes web se ices p ocu emen (WSP), an ac i i y ocussed on he acquisi ion o web se ices equi ed by a web- se ice-based sys em, hus i is a c i ical ac i i y o cu en web sys em de elope s. Some ypical asks in ol ed in WSP a e: –Speci ica ion o demands and o e s, which should be checked o consis ency in o de o e i y hey do no con ain any inne con adic ion. –Sea ch o o e s, which should be checked o con o mance in o de o e i y hey ul ill he demand, so ha he selec ion is limi ed o such o e s. –Selec ion o he bes choice acco ding o he assessmen c i e ia which is included in he demand. In his pape , we p esen a p oposal o au oma e he p ocu emen o web se ices. Ou p oposal imp o es on o he s in ha i suppo s a symme ic speci ica ion model. Thus, p o ide s can include in hei o e s equi emen s as complex as cus ome s in- clude in hei demands. The key poin o achie e his imp o emen has been using he cons ain p og amming as a o mal basis, since i endows he model wi h a e y powe ul exp essi eness. A XML–based implemen a ion is p esen ed along wi h some expe imen al esul s and compa isons wi h o he app oaches. The es o he pape is s uc u ed as ollows. In Sec ion 2, we in oduce he no- ions o asymme ic and symme ic speci ica ion models, as well as an o e iew o ela ed wo ks. In Sec ion 3, we p opose he use o cons ain p og amming as a means o achie ing a symme ic speci ica ion model. In Sec ion 4, we p esen b ie ly he main implemen a ion aspec s o ou un- ime amewo k, oge he wi h some expe imen al esul s. Finally, in Sec ion 5 we summa ise he p esen ed wo k and he immedia e u- u e wo k. 2 Symme ic e sus Asymme ic Models 2.1 Asymme ic models Le S be a mul idimensional space whose dimensions a e gi en by domains o quali y- o -se ice pa ame e s. T adi ionally, a demand (  ) has been iewed as a subspace in S , whe eas an o e ( ! ) has been iewed as a poin in S . Thus, checking he con- o mance amoun s o checking whe he he poin ( he o e ) belongs o he subspace ( he demand) o no . See Figu es 1.a and 1.b, espec i ely. This checking can be com- pu ed easily by e alua ing ! in  . As an example, i a web se ice owns he o e ! = MT T F =120 g , hen i is con o man o he demand  1 = MT T F  100 g be- cause 120  100 , bu no o he demand  2 = MT T F > 120 g because 120 6 > 100 . This in e p e a ion o con o mance esul s in a model which is asymme ic wi h ega d o he exp essi eness o quali y-o -se ice speci ica ions. This seman ics makes e y di icul o speci y o e s when i is needed some hing else han a poin , as an example o speci y some unce ain yo a space. As mos o p og amminglanguages a e able o check i a poin is inside a space, whe eas checking i a space includes ano he space is a ha d ques ion, mos o pla o ms ha e adop ed an asymme ic speci ica ion model. As well, hese app oaches wi h an asymme ic model usually own a limi ed exp essi eness because condi ions a e es ic ed o simple exp essions in ol ing single pa ame e s, so complex exp essions a e no allowed. (a) S S (b) Fig.1. Con o mance in asymme ic models. 2.2 Symme ic models Al e na i ely, an o e can be also conside ed as a sub-space, jus as demands, so ha i ep esen s he anges o quali y-o -se ice alues ha he co esponding web se ice gua an ees o supply. In his way, an o e ( ! ) is con o man o a demand (  ) when- e e he o e ’s sub-space is inside he demand’s sub-space (see Figu e 2.a), o he wise he o e is no con o man (see Figu e 2.b). As an example, i a web se ice owns he o e ! = MT T F > =120 g , hen i is con o man o he ollowing demand  1 = MT T F > =100 g , bu no o he demand  2 = MTTF > 120 g because he o e ’s ins ance alue { MT T F = 120 } is ou o he demand’s space. This in e p e a iono con o mance esul s in asymme icmodel becausequali y-o - se ice in demands and o e s can be speci ied in he same way. This seman ics makes he o e gua an ee he comple e ange, no only a conc e e alue, i.e., we can no make any assump ion on a conc e e alue, because i is equally possible any alue in he sub- space, and he e is no con ol o ge a conc e e alue. As well, symme ic app oaches usuallyachie eag ea e dealo exp essi eness ospeci yquali y-o -se ice,since he e is usually no es ic ion on he numbe o in ol ed pa ame e s o ype o ope a o s, so ha non-linea o mo e complex exp essions a e allowed. (b) (a) SS Fig.2. Con o mance in symme ic models. 2.3 Rela ed wo k Figu e 3 shows a compa a i e s udy among he mos p ominen (as a as we know) quali y-awa e app oaches o WSP. B ie ly: –The UDDI Ex ension (UDDIe) [28] is based on he UDDI (Uni e sal Desc ip- ion Disco e y and In eg a ion) se ices. UDDIe owns an asymme ic model when speci ying demands and o e s. IBM's WSME MME HP's MME Se ice UDDIe Ou P oposal Cus ome 's P o ide 's Name-Value Pai P ope ies S a ic/Dynamic Binding Sc ip s o Rule-based Reqs. Composi ion Single-Pa ame e Cons ain s on Pa ame e s o Se ice (expandable) Name-Value Pai P ope ies Single-Pa . Condi ions on P ope ies (Quali ie s) Composi ion Mul iple-Pa ame e Cons ain s on Pa ame e s o Se ice Da a S uc u ing Da a Dic iona y: p e-de . basic ypes sequences eco ds DAML+OIL On ology: da a ypes and ypes subsump ion Blue Pages Ca alogues: p e-de . basic ypes ca alogue ex ension basic and de i ed p. Dynamic View: The P ocess Model Ad e isemen /Submission Que y/Submission Ma chmaking Selec ing P o ide s' O e s Ad e ising Que ying B owsing Publishing Sea ch and Disco e y C ea ing Ca alogues O e s Submission Demands Submission Ma chmaking Ca alogues, Pa ame e s & Measu es S akeholde s P o ide s Cos ume s Ad e ise s Reques o s P o ide s Consume s P o ide s Cos ume s Quali y-o -Se ice Documen s Ad e isemen s Que ies Ag eemen s Se ice O e s & Reques s Publishing Inqui y Demands O e s Ag eemen s S a ic View: The Lexicon The Re e ence Model Fig.3. A compa ison o quali y-awa e app oaches o WSP. –TheHP’sMa chmakingEngine(MME) [10] is basedon heDAML(DARPAAgen Ma kup Language) seman ic web language [2]. I is he closes p oposal o ou s, because i owns a symme ic model o speci y quali y-o -se ice, and i uses cons- ain s o do i , so i owns a g ea exp essi eness. As well, i uses a Desc ip ion Logic DL’s sol e as a mean o ca ying ou he WSP- ela ed asks. Ne e heless, he e is no cu en ly any DL’s sol e e sion able o p ocess some o he mos complex exp essions which can be speci ied in MME. –The IBM’sWeb Se ices Ma chmaking Engine (WSME) [12], which is ela ed o Web Se ice Le el Ag eemen (WSLA) [15,17], is based on he CORBA/ODP ade se ice. I owns an asymme ic model and he e is no op ima ion o he se- lec ion because sea ch esul s a e only he lis s o con o man o e s. None heless, he e is a di e ence: ela ionships be ween demands and o e s a e bila e al. In he same way quali y-o -se ice in o e s is based on pa ame e / alue pai s whe eas demands impose condi ions on hem, i is also allowed ha demands de ine hei own quali y-o -se ice pa ame e s whe eas he o e s impose condi ions on hem. As an example, le an o e be gi en by he ollowing quali y-o -se ice speci i- ca ion ! = me:M T T F =120 & you :na ional i y 2 BE;:::;UK gg and a demand  = me:na ional i y = IS g & y ou :M T T F > 100 g , hen he o e ! is no con o man o he demand  , because he condi ion i imposes on he de- mand ( he Eu ope Union membe ship) is no ul illed, despi e o he o e ul ills condi ions imposed by he demand. –O he languages o speci ying quali y-o -se ice and ade se ices he Quali y- o -se iceModelingLanguage(QML)[8], heNoFunlanguage[6],and he CORBA ade se ice [22]. These p oposals a e no di ec ly ela ed o WSP. 3 Suppo ing WSP wi h Cons ain P og amming We ha e chosen ma hema ical cons ain s as he way o speci ying quali y-o -se ice in demands and o e s. In his way, checking con o mance can be ca ied ou jus as a cons ain sa is ac ion p oblem (CSP) o a cons ain sa is ac ion op imisa ion p o- blem (CSOP) [7,11,18,29]. In gene al, CSP-based modelling is qui e simple and in ui- i e (in mos cases) in he con ex o p oblems which we a e dealing wi h. Cons ain p og amming is an excellen suppo o symme ic speci ica ions models, because i makes possible o check whe he a space is included in ano he one, being hese spaces ea ed as cons ain s. Ou p oposal owns a symme ic speci ica ion model wi h a g ea deal o exp essi eness because o using cons ain s. 3.1 Cons ain p og amming in a nu shell Cons ain P og amming (CP) has ecen ly a ac ed high a en ion among expe s om many a eas because o i s po en ial o sol ing ha d eal-li e p oblems. No only i is based on a s ong heo e ical ounda ion, bu i is an a ac ing widesp ead comme cial in e es , as well. Cons ain s o malise hose dependencies in physical wo lds and hei ma hema ical abs ac ions na u ally and anspa en ly. A cons ain is simply a logical ela ion among se e al a iables, each aking a alue in a gi en domain. The cons ain hus es ic s he possible alues ha a iables can ake, and i ep esen s a pa ial in- o ma ion abou he a iables o in e es . An impo an ea u e o cons ain s is hei decla a i e manne , i.e., hey speci y wha ela ionships mus hold wi hou speci ying a compu a ional p ocedu e o en o ce hem. CP is he s udy o compu a ional sys ems based on cons ain s. The idea o CP is o sol e p oblems by s a ing cons ain s ( e- qui emen s) abou he p oblem a ea and, consequen ly, inding solu ion sa is ying all he cons ain s. The ea lies ideas leading o CP may be ound in he A i icial In elligence da ing back o six ies and se en ies. The scene labelling p oblem [30] is p obably he i s cons ain sa is ac ion p oblem ha was o malised. The main s ep owa ds CP was achie ed when Gallai e [9] and Ja a & Lassez [14] no ed ha logic p og ammingwas jus a pa icula kind o cons ain p og amming.The basic idea behind Logic P og am- ming (LP), and decla a i e p og amming in gene al, is ha he use s a es wha has o be sol ed ins ead o how o sol e i , which is e y close o he idea o cons ain s. The e o e he combina ion o cons ain s and logic p og amming is e y na u al, and Cons ain Logic P og amming (CLP) makes a nice decla a i e en i onmen o sol - ing p oblems by means o cons ain s. Howe e , i does no mean ha CP is es ic ed o CLP. Cons ain s we e in eg a ed o ypical impe a i e languages like C++ and Ja a, as well. The nowadays eal-li e applica ions o CP in he a ea o planning, scheduling and op imisa ion ise he ques ion i he adi ional ield o Ope a ions Resea ch (OR) is a compe i o o an associa e o CP. The e is a signi ican o e lapo CP and OR in he ield o NP-Ha d combina o ial p oblems. While he OR has a long esea ch adi ion and ( e y success ul) me hod o sol ing p oblems using linea p og amming, he CP em- phasis is on highe le el modelling and solu ions me hods ha a e easie o unde s and by he inal cus ome . Mos ecen ad ances p omise ha bo h me hodologies can ex- ploi each o he , in pa icula , he CP can se e as a oo pla o m o in eg a ing a ious cons ain sol ing algo i hms including hose de eloped and checked o be success ul in OR. As he abo e pa ag aphs show, he CP has an inne in e disciplina y na u e. I combines and exploi s ideas om a numbe o ields including A i icial In elligence, Combina o ial Algo i hms, Compu a ional Logic, Disc e e Ma hema ics, Neu al Ne - wo ks, Ope a ions Resea ch, P og amming Languages, and Symbolic Compu a ion. Cu en ly, we see wo b anches o CP, namely cons ain sa is ac ion and cons ain sol ing. Bo h sha e he same e minology bu he o igins and sol ing echnologies a e di e en . The o me deals wi h p oblems de ined o e ini e domains and, cu en ly, p obably mo e han 95% o all indus ial cons ain applica ions use ini e domains. The e o e,we deal wi h cons ain sa is ac ion p oblems mos ly in his pape . The la e sha es he basis o CP, i.e., desc ibing he p oblem as a se o cons ain s and sol - ing hese cons ain s. Bu now, he cons ain s a e de ined (mos ly) o e in ini e o mo e complex domains. Ins ead o combina o ial me hods o cons ain sa is ac ion, he cons ain sol ing algo i hms a e based on ma hema ical echniques such as au o- ma ic di e en ia ion, Taylo se ies o New on me hod. Cons ain Sa is ac ion P oblems [29] ha e been a subjec o esea ch in A i icial In elligence o many yea s. A Cons ain Sa is ac ion P oblem (CSP) is de ined as a se o a iables each anging on a ini e domain, and a se o cons ain s es ic ing all he alues ha a iables can simul aneously ake. A solu ion o a CSP is an assignmen o a alue om i s domain o e e y a iable, in such a way ha all cons ain s a e sa is ied a once. We may wan o ind: i) jus one solu ion, wi h no p e e ence as o which one, ii) all solu ions, iii) an op imal, o a leas a good solu ion, gi en some objec i e unc ion de ined in e ms o some o all o a iables. Solu ions o a CSP can be ound by sea ching (sys ema ically) h ough all possible alue assignmen s o a iables. In many eal-li e applica ions, we do no wan o ind any solu ion bu a good solu- ion. The quali y o solu ion is usually measu ed by an applica ion dependen unc ion called objec i e unc ion. The goal is o ind such solu ion ha sa is ies all he cons- ain s and minimise o maximise he objec i e unc ion, espec i ely. Such p oblems a e e e ed o as Cons ain Sa is ac ion Op imisa ion P oblems (CSOP), which con- sis s o a s anda d CSP and an op imisa ion unc ion ha maps e e y solu ion (comple e labelling o a iables) o a nume ical alue [29]. 3.2 Consis ency and con o mance Whene e a new demand o o e is submi ed, i s consis ency needs o be checked, i.e., whe he o no i con ains any inne con adic ion. This is in e p e ed as a CSP, so ha i he co espondingCSP is sa is iable, hen he demand o o e can be conside ed as consis en . The co esponding CSP o a demand o o e is composed o all he cons ain s i con ains. On he o he hand, he bes choice selec ion ega ding wi h a de- mand implies he p e ious checking o con o mance, because he sea ch is educed o con o man o e s. As we a e using cons ain p og amming, checking o con o mance lies in de e mining whe he each and e e y solu ion o he o e ’s CSP is also a solu ion o he demand’s CSP. In his way, he co esponding CSP o checking he con o mance is cons uc ed acco ding o he de ini ion gi en in [18]: con o mance ( !;  ) , sa ( c ! ^: c  )= alse whe e ! is he o e and c ! i s co esponding CSP,  is he demand and c  i s co es- ponding CSP, and sa is a unc ion ha we iden i y wi h he CSP sol e which is being used. I can be applied on a CSP c so ha i e u ns one o he ollowing esul s: ue i c is sa is iable, alse i no , and ? i he sol e canno de e minewhe he c is sa is iable o no . 3.3 Op imali y Mo e o en han no , i is possible o ha e se e al o e s which a e con o man o he same demand o a web se ice, hen we should selec ha o e which is he bes choice. This selec ion is ca ied ou acco ding o he assessmen c i e ia he cus ome includes in his o he demand. These c i e ia may be gi en by u ili y unc ions [3,16, 21] which, in gene al, ha e he signa u e U :  ! [0 ; 1] whe e  is he measu ing domain o a quali y-o -se icepa ame e . U ili y unc ions assign an u ili y assessmen ( anging om 0 o 1) o e e y quali y-o -se ice alue i can ake, so he g ea e he assessmen , he be e he conside a ion o he cus ome . The e o e, u ili y unc ions allow he es ablishmen o an objec i e c i e ia, gi en by cus ome s, in o de o selec hose o e s which be e ul ill he demands. Figu e 4 shows se e al u ili y unc ions co esponding o examples in his sec ion. U ili y o Mean Time To Failu e 0,5 1 0,25 1209060 MTTF 0,75 45 75 105 140 010 20 30 0,5 1 0,25 0,75 05152535 MTTR U ili y o Mean Time To Repai Modem Modem ISDN ISDN Modem ADSL ISDN ADSL Modem ISDN ADSL ADSL U ili y o Media Suppo MEDIA 0,5 1 0,25 0,75 0 { } Fig.4. U ili y unc ions o MTTF , MT T R , and MEDIA . Al houghwe can make useo anykind o unc ion o speci y u ili y unc ions,linea piecewise unc ionsa e o en he p e e ed.As an example, he u ili y unc ion o a nu- me ic quali y-o -se icepa ame e can be de inedby means o polylines de e mined by a sequence o coo dina epoin s such as ( x 1 ;u 1 ) ; ( x 2 ;u 2 ) ;:::; ( x n ;u n ) , whe e e e y x ep esen s a alue in he measu ing domain o he quali y-o -se ice, and u i s assess- men in he ange [0,1]. The co esponding u ili y unc ion is hen gi en by: U ( x )= 8 > > < > > : u 1 + u 2 ; u 1 x 2 ; x 1 ( x ; x 1 ) i x 1  x<x 2  u n ; 1 + u n ; u n ; 1 x n ; x n ; 1 ( x ; x n ; 1 ) i x n ; 1  x  x n ? i x<x 1 o x>x n We a e no usually in e es ed in compu ing he u ili y assessmen o an unique quali y-o -se ice pa ame e , bu on maximising he global assessmen o o e s in o de o selec he bes one, being hese o e s con o man o he demand. Ne e he- less, we can no compu e he maximum o e s’u ili y assessmen s when compa ing hem. As an example, le he ollowing o e s ! 1 = 60  MT T F  120 g and ! 2 = 90  MT T F  110 g . In ui i ely, he i s is be e , because i MT T F =120 hen U ( ! 1 ) = 1 . Howe e , he o e is gua an eeing he comple e ange, no only a conc e e alue, so we can no make such assump ion because i is equally possible ha MTTF =60 , and he e is no con ol o ge a conc e e alue. The e o e, we compa e he minimum u ili y assessmen s o o e s. In his way, he la e o e is he be e , be- cause i MTTF =90 hen U ( ! 2 ) = 0 : 5 , whe eas he wo s assessmen o he i s o e is 0.25, a mos . Fo mally, he bes o e ( ! S ) can be de ined as: ! S = ! 2    8 ! i 2   ; ! g U  ( ! ) U  ( ! i ) whe e ! and ! i s and o o e s in he se   o con o man o e s o he demand  , and he U  ( ! ) u ili y unc ion o an o e ! acco ding o assessmen c i e ia in demand  is de ined as: U  ( ! )= min P  2 c ! w   U  (  ) s c ! whe e  ep esen s a quali y-o -se ice pa ame e which is in ol ed in he o e ’s CSP c ! , and U  (  ) i s u ili y unc ion, and w   i s assigned weigh , acco ding o assessmen c i e ia in demand  . On he o he hand, weigh s a e needed o exp ess ha a quali y- o -se ice pa ame e is p e e ed o ano he . 3.4 An example o con ain -based quali y-o -se ice speci ica ion Figu e 5 shows se e al ca alogues, demands, and o e s w i en in QRL [23,26], he language which we ha e de ined o speci ying quali y equi emen s. Figu e 4 shows he g aphical ep esen a ion o u ili y unc ions appea ing in Figu e 5. These demands and o e s will be used in he examples along hese pa ag aphs. In his case, he in ol ed quali y-o -se ice pa ame e s a e he Mean Time To Failu e (MTTF), heMean Time To Repai (MTTR),and heMedia Sup- po (MEDIA).No e he included demand and o e s a e all consis en , because hei co esponding CSP a e sa is iable, as well as o e s a e con o man o he demand, because he co esponding CSP o checking he con o mance a e no sa is iable, ac- co ding o de ini ions in Sec ion 3.2. Since bo h o e s a e con o man o he demand, we will ha e o compu e he u i- li y unc ions o compa e hem. Acco ding o de ini ions in Sec ion 3.3, bo h o e s own U ( MTTF = 110) = 0 : 83 and U ( MT T R =10) = 0 : 8 , elazquez owns U ( MEDIA )=1 , and zipi owns U ( MEDIA )=0 : 5 . The e o e, u ili y assessmen o elazquez is 0 : 9  0 : 83 + 0 : 05  0 : 04 + 0 : 05  1=0 : 84 , and u ili y assessmen o zipi is 0 : 9  0 : 83 + 0 : 05  0 : 04+0 : 05  0 : 5= 0 : 815 , so he bes o e is elazquez. // A ca alogue o Reliabili y- ela ed QoS pa ame e s ca alogue Reliabili y { MTTF { desc ip ion: "Mean Time o Failu e"; domain: eal [0,+in ) minu e; }; MTTR { desc ip ion: "Mean Time To Repai "; domain: eal [0,+in ) minu e; }; } // A ca alogue o Mul imedia- ela ed QoS pa ame e s ca alogue Mul imedia { MEDIA { desc ip ion: "Media Suppo "; domain: se { modem, ISDN, ADSL }; } } // Web se ice o e supplied by Velazquez using Reliabili y, Mul imedia; o e o IVideoSe e { O1: MTTF >= 110 and MTTF <= 120; O2: MTTR > 5 and MTTR <= 10; O3: MEDIA = {ADSL,ISDN,modem}; } c) Se e al o e s. a) Ca alogues o quali y-o -se ice pa ame e s. // Web se ice demand o IVideoSe e using Reliabili y, Mul imedia; demands o IVideoSe e { D1: MTTF / (MTTF + MTTR) >= 0.9; D2: MEDIA includes {modem,ISDN}; } assessmen { MTTF {90, { (0,0), (90,0.5), (120,1) } }; MTTR {05, { (0,1), (20,0.6), (30,0) } }; MODEM {05, case MEDIA = { } : 0.01; case MEDIA = {modem} : 0.1; case MEDIA = {ISDN} : 0.3; case MEDIA = {ISDN,modem} : 0.5; case MEDIA = {ADSL} : 0.9; case MEDIA = {modem, ADSL} : 1; case MEDIA = {ISDN, ADSL} : 1; case MEDIA = {modem, ISDN, ADSL} : 1; } } b) A demand. // Web se ice o e supplied by Zipi using Reliabili y, Mul imedia; o e o IVideoSe e { O1: MTTF >= 110 and MTTF <= 120; O2: MTTR > 5 and MTTR <= 10; O3: MEDIA = {ISDN,modem}; } Fig.5. Demands and o e s w i en in QRL. 4 Implemen a ion and Expe imen al Resul s 4.1 O e iew o he p o o ype’s a chi ec u e We a e de elopinga p o o ypeo a un- ime amewo k o WSP [19,20,24,27], whose p elimina y e sion is a ailable a h p://www.lsi.us.es/˜oc a io. In his pape , we gi e a b ie e iew, oge he wi h some expe imen al esul s we ha e ecen ly ob ained. A componen s iew o he un- ime amewo k is shown in Figu e 6. Selec ing a mul i-le el a chi ec u e along wi h he deploymen o he componen s as web applica ions o web se ices ha e been c i ical design decisions. Componen s a e spli up among he uppe use -in e ace le el, he in e media e se ice and u ili y le els, and he bo om eposi o y le el. These componen s can be eusable and in e - changeable.Se ice le el includes hose componen swhich implemen he IImpo - Se ice in e ace ( unc ions ela ed o submission o demands and sea ching o bes con o man o e ), and he IExpo Se ice in e ace ( unc ions ela ed o submis- sion o o e s). These componen s ha e need o in oking checkings o consis ency, con o mance, and op imum sea ch. These unc ions a e implemen ed by he Quali y T ade Web Se - ice [19] a he u ili y le el. Each unc ion has a simila ope a ion: 1. I akes he in ol ed demands and o e s w i en in XML as pa ame e s.