Theoretical and Computation Basis for CATNETS - Annual Report Year 3

Bay eu he A bei spapie e zu Wi scha sin o ma ik Leh s uhl ü Wi scha sin o ma ik In o ma ion Sys ems Managemen Bay eu h Repo s on In o ma ion Sys ems Managemen No. 23 2007 Daniel Vei , Geo g Buss (Uni e si y o Mannheim), Bjö n Schnizle , Di k Neumann (Uni e si y o Ka ls uhe), We ne S ei be ge , To s en Eymann (Uni e si y o Bay eu h) Theo e ical and Compu a ional Basis o CATNETS - Annual Repo Yea 3 ISSN 1864-9300 Die A bei spapie e des Leh s uhls ü Wi scha sin o ma ik dienen de Da s ellung o läu ige E gebnisse, die i. d. R. noch ü spä e e Ve ö en lichungen übe a bei e we den. Die Au o en sind deshalb ü k i ische Hinweise dankba . The Bay eu h Repo s on In o ma ion Sys ems Managemen comp ise p elimina y esul s which will usually be e ised o subsequen publica ions. C i ical commen s would be app ecia ed by he au ho s. Alle Rech e o behal en. Insbesonde e die de Übe se zung, des Nachd uckes, des Vo ags, de En nahme on Abbildungen und Tabellen – auch bei nu auszugsweise Ve we ung. All igh s ese ed. No pa o his epo may be ep oduced by any means, o ansla ed. Au ho s: In o ma ion Sys ems and Managemen Wo king Pape Se ies Edi ed by: P o . D . To s en Eymann Managing Assis an and Con ac : Raimund Ma os Uni e si ä Bay eu h Leh s uhl ü Wi scha sin o ma ik (BWL VII) P o . D . To s en Eymann Uni e si ä ss asse 30 95447 Bay eu h Ge many Email: aimund.ma os@uni-bay eu h.de ISSN Daniel Vei , Geo g Buss (Uni e si y o Mannheim), Bjö n Schnizle , Di k Neumann (Uni e si y o Ka ls uhe), We ne S ei be ge , To s en Eymann (Uni e si y o Bay eu h) 1864-9300 IST-FP6-003769 CATNETS Y3 Repo WP 1: Theo e ical and Compu a ional Basis Con ac ual Da e o Deli e y o he CEC: 31. Augus 2007 Ac ual Da e o Deli e y o he CEC: 30. Sep embe 2007 Au ho (s): Daniel Vei , Geo g Buss (Uni e si y o Mannheim) Bj¨o n Schnizle , Di k Neumann (Uni e si y o Ka ls uhe) We ne S ei be ge , To s en Eymann (Uni e si y o Bay eu h) Wo kpackage: WP 1 Theo e ical and Compu a ional Basis Es . pe son mon hs: 19 Secu i y: public Na u e: submi ed e sion Ve sion: 1.0 To al numbe o pages: 50 Abs ac : This epo co e s he esul s o WP1 in Y2, Y3 and hence comp ises mainly he esul s om T1.4 and T1.5. Keywo ds (op ional): Decen alized Ma ke Mechanisms, Cen alized Ma ke Mechanisms, Ca allaxy, Ma ke Enginee ing, Simula o In eg a ion, P o o ype In eg a ion CATNETS Conso ium This documen is pa o a esea ch p ojec pa ially unded by he IST P og amme o he Commission o he Eu- opean Communi ies as p ojec numbe IST-FP6-003769. The pa ne s in his p ojec a e: LS Wi scha sin o ma ik (BWL VII) / Uni e si y o Bay eu h (coo dina o , Ge many), A qui ec u a de Compu ado s / Uni e si a Poli ecnica de Ca alunya (Spain), In o ma ion Managemen and Sys ems / Uni e si y o Ka ls uhe (TH) (Ge many), Dipa imen o di Economia / Uni e si ´a delle ma che Ancona (I aly), School o Compu e Science and he Welsh eScience Cen e / Uni e si y o Ca di (Uni ed Kingdom), Au oma ed Reasoning Sys ems Di ision / ITC-i s T en o (I aly), Chai o Business Adminis a ion and In o ma ion Sys ems - E-Business and E-Go e nmen / Uni e si y o Mannheim (Ge - many). Uni e si y o Bay eu h LS Wi scha sin o ma ik (BWL VII) 95440 Bay eu h Ge many Tel: +49 921 55-2807, Fax: +49 921 55-2816 Con ac pe son: To s en Eymann E-mail: [email p o ec ed] Uni e si a Poli ecnica de Ca alunya A qui ec u a de Compu ado s Jo di Gi ona, 1-3 08034 Ba celona Spain Tel: +34 93 4016882, Fax: +34 93 4017055 Con ac pe son: Felix F ei ag E-mail: [email p o ec ed] Uni e si y o Ka ls uhe Ins i u e o In o ma ion Managemen and Sys ems Engle s . 14 76131 Ka ls uhe Ge many Tel: +49 721 608 8370, Fax: +49 721 608 8399 Con ac pe son: Ch is o Weinha d E-mail: [email p o ec ed] Uni e si ´ a delle ma che Ancona Dipa imen o di Economia Piazzale Ma elli 8 60121 Ancona I aly Tel: 39-071- 220.7088 , Fax: +39-071- 220.7102 Con ac pe son: Mau o Gallega i E-mail: [email p o ec ed] Uni e si y o Ca di School o Compu e Science and he Welsh eScience Cen e Uni e si y o Ca adi , Wales Ca di CF24 3AA, UK Uni ed Kingdom Tel: +44 (0)2920 875542, Fax: +44 (0)2920 874598 Con ac pe son: Ome F. Rana E-mail: [email p o ec ed] .ac.uk ITC-i s T en o Au oma ed Reasoning Sys ems Di ision Via Somma i e, 18 38050 Po o - T en o I aly Tel: +39 0461 314 314, Fax: +39 0461 302 040 Con ac pe son: Flo iano Zini E-mail: [email p o ec ed] Uni e si y o Mannheim Chai o Business Adminis a ion and In o ma ion Sys ems - E-Business and E-Go e nmen - L9, 1-2 68131 Mannheim Ge many Tel: +49 621 181 3321, Fax: +49 621 181 3310 Con ac pe son: Daniel Vei E-mail: [email p o ec ed] Con en s 1 In oduc ion 3 2 Sel -O ganiza ion in Compu ing Sys ems - Pu ing CATNETS in o a G ea e Pe spec i e 5 2.1 In oduc ion o Sel -O ganiza ion . ..................... 5 2.2 In as uc u al Sphe es o Sel -O ganizing Compu ing . . . . ....... 6 2.3 The Open Se ice In as uc u e . ..................... 7 2.4 Abou he Necessi y o C ea e an Open SOC Policies In as uc u e . . . . 8 3 Fo mal Desc ip ion o Mechanisms 10 3.1 Cen alized Mechanisms . . . . . ..................... 10 3.2 The Ca allaxy as an Al e na i e Decen alized App oach . . ....... 11 3.2.1 Se up and Va iables De ini ion . . . . . .............. 13 3.2.2 The Nego ia ion S a egy . ..................... 14 3.2.3 Gossip Lea ning . . . . . ..................... 17 3.2.4 The Lea ning Algo i hm . ..................... 17 4 Bidding Issues 20 4.1 Wha Does an Agen Bid? . . . . ..................... 20 4.1.1 No a ion . . ............................ 20 4.1.2 Valua ion Gene a ion . . . ..................... 21 4.1.3 Calib a ion o he Valua ion Gene a o . .............. 23 4.2 When Does an Agen Bid? . . . . ..................... 24 4.2.1 Complex Se ice Agen . ..................... 24 4.2.2 Basic Se ice Agen . . . ..................... 24 4.2.3 Resou ce Se ice Agen . ..................... 27 4.3 Summa y . . ................................ 28 5 In eg a ion o Mechanisms in o Simula o 29 5.1 In eg a ion o he Auc ion Mechanisms . . . . .............. 29 5.1.1 Implemen a ion o he Ma ke s . . . . . .............. 29 5.1.2 In eg a ion in o Op o Sim ..................... 33 5.2 Decen alized Mechanisms (Ca allaxy) . . . . . .............. 35 1 CONTENTS 2 5.2.1 Implemen a ion o he Ma ke s . . . . . .............. 35 5.2.2 In eg a ion in o Op o Sim ..................... 38 5.3 Simula ion Resul s . ............................ 40 6 Rela ions o o he WPs 43 6.1 WP2..................................... 43 6.2 WP3..................................... 43 6.3 WP4..................................... 44 7 Summa y 45 7.1 Re iew ................................... 45 7.2 Con en o Y3 ................................ 46 7.3 Ou look . . . ................................ 47 Bibliog aphy 47 Chap e 1 In oduc ion The p ima y a ge o he CATNETS p ojec is he quan i a i e compa ison be ween he echnical and economic e iciency o ma ke -based esou ce alloca ion mechanisms in applica ion laye ne wo ks such as G ids. He e, wo undamen ally di e en app oaches a e compa ed. A cen alized – auc ion-based – ma ke mechanism and a decen alized – Ca allaxy-based – ma ke mechanism. A e he eo ganiza ion ( ollowing he Y1 e iew) his endea o has been app oached in he ollowing way: •Wo kpackage 1 (Theo e ical and Compu a ional Basis): The a ge o his wo kpackage is he de ini ion o ma ke mechanisms o he cen alized and he decen alized case. The e o e, so wa e componen s ha e o be iden i ied (T1.1), ma ke equi emen s ha e o be analyzed (T1.2) and an a chi ec u e o se ices and ALNs has o be designed (T1.3). Finally a speci ica ion o bidding and in e ac ion issues has o be ca ied ou (T1.4). A speci ica ion and analysis o he ma ke mechanisms concludes WP1. •Wo kpackage 2 (Simula ion F amewo k): The co e o his wo kpackage is he imple- men a ion o a simula o amewo k in eg a ing bo h, he cen alized and he decen al- ized ma ke mechanism. The goal is o compa e he ou come o he applica ion o bo h mechanisms quan i a i ely. •Wo kpackage 3 (P oo -o -Concep Applica ions): In pa allel o WP2 his wo kpackage ocuses on he implemen a ion o he designed decen alized ma ke mechanisms in o a p o o ypical ALN-middlewa e so wa e. Quan i a i e e alua ions a e ca ied ou by unning expe imen s wi h his pla o m. •Wo kpackage 4 (Pe o mance E alua ion): The aim o his WP is he iden i ica ion and design o me ics in o de o measu e he ou come o WP2 and WP3. He e, a me ics amewo k is designed in o de o enable he measu emen o he quali y o alloca ion esul s in using an economic scale. 3 CHAPTER 1. INTRODUCTION 4 •Wo kpackage 5 (Managemen ): This wo kpackage is designed o ca y ou p ojec managemen and dissemina ion. This epo co e s he esul s o WP1 in Y2, Y3 and hence comp ises mainly he esul s om T1.4 and T1.5. The emainde o his epo is s uc u ed as ollows: Chap e 2 illus a es esea ch ques ions o he CATNETS p ojec in he con ex o sel -o ganizing sys ems and d aws a ision owa ds u u e esea ch opics. In he ollowing, chap e 3 ocuses on he desc ip- ion o he in oduced ma ke mechanisms. In Sec ion 3.1 he p ope ies o he cen alized ma ke mechanisms, which ha e been de ined in Y1 epo [SNV+05b] a e b ie ly de- sc ibed. Sec ion 3.2 p o ides a o mal desc ip ion o he decen alized alloca ion mecha- nisms. The bidding issues, p esen ed in Chap e 4, elabo a e di e en scena ios connec - ing he se ice and esou ce ma ke . Ad an ages and disad an ages o hese scena ios a e compa ed o enable a compa ison o he cen alized and decen alized ma ke mechanism. In Chap e 5 he in eg a ion o he mechanisms in o he Op o Sim simula o is ou lined and links o WP2 a e se . The ein, o e all simula ion esul s a e p o ided. Chap e 6 e- la es Wo kpackage 1 o he o he wo kpackages. Finally, Chap e 7 p o ides a summa y o he wo k ha was done in wo kpackage 1. Chap e 2 Sel -O ganiza ion in Compu ing Sys ems - Pu ing CATNETS in o a G ea e Pe spec i e 2.1 In oduc ion o Sel -O ganiza ion The ision o Sel -O ganiza ion in Compu ing Sys ems and Ne wo ks has gained signi i- can in e es in he las yea s, and e en was labeled wi h a popula buzzwo d: Au onomic Compu ing [KC03] desc ibes a concep o sel -o ganizing in o ma ion echnology, whe e he unc ionali y o he compu ing sys em is an eme gen ea u e o he capabili ies and ac ions o he componen s. Wi hou any cen alized con olle , his sys em is capable o con igu ing, healing, o ganizing and p o ec ing i sel ( he so-called CHOP p ope ies). In con as , classical IT con ol in ol es a cen alized con olle ins ance wi h global knowl- edge abou he cu en s a us o he compu ing sys em, and a de ailed egula ion mecha- nism o ’heal’ de ia ions om a de ined ’no mal’ s a us. Cen alized compu a ion is said o be no ha lexible in e ms o scalabili y and adap abili y. Au onomic Compu ing uses a biological pa adigm as a design and con ol me apho , he au onomic ne ous sys em. The co e CHOP p ope ies o he Au onomic Compu ing concep a e in ended o be an elec onic ealiza ion o he espec i e mechanisms o he human body. Sel -o ganiza ion can be ound in o he pa s o ou na u al en i onmen as well, e.g. biological e olu- ion, social g oup beha io , ma ke dynamics phenomena and o he complex adap i e sys ems. Au onomics e e s o ou own human neu al sys em, Ca allaxy [ESMP03] o sel -o ganizing ma ke s in Economics, S igme gy o coo dina ion wi hou communica- ion in insec colonies. All hese ideas ha e in common ha he solu ion o g owing complexi y bo h in scale and seman ics is decen alized con ol, ha is based on local in o ma ion and execu ed h ough local e ec s which build up o a sys em-wide eme gen beha io . I is no su p ising ha p ojec s labeled Au onomic Compu ing a e hus man- i old, coming om di e se backg ounds and academic habi a s, and aiming a a a ie y 5 CHAPTER 3. FORMAL DESCRIPTION OF MECHANISMS 12 Figu e 3.2: Decen alized Se ice Disco e y decide on hei own, and do no ake he sys em s a e in o accoun . In he Edgewo h p ocess [Va 94], economic subjec s ade bila e ally wi h each o he only i hei u ili y is supposed o inc ease a e he ba e . In ha case, he sum o all u ili ies inc eases a e each success ul ba e ; he inal s a e is Pa e o-op imal and has maximum sys em u ili y. A heo e ical undamen how he concep s o dynamic ma ke p ocesses, he e oge- neous agen s and choice unde incomple e in o ma ion a e linked, can be ound in Neo- Aus ian Economics, in pa icula in F ied ich Augus on Hayeks Ca allaxy concep [HBKC89]. Ca allaxy desc ibes a s a e o spon aneous o de , which comes in o exis- ence by he communi y membe s communica ing (ba e ing) wi h each o he and hus achie ing a communi y goal ha no single use has planned o . The implemen a ion o Ca allaxy, desc ibed in his pape , uses e o s om bo h agen echnology and economics, no ably agen -based compu a ional economics [Tes97]. Au- onomous so wa e agen s nego ia e wi h each o he using an al e na ing o e s p o o- col [Ros94] and adap hei nego ia ion s a egies using eedback lea ning algo i hms (e olu iona y algo i hms, nume ical op imiza ion e.g. Nelde /Meads simplex me hod [PT02], hyb id me hods e.g. B enne s VID model [B e02]). Ongoing communica- ion by using p ice signaling leads o cons an adap a ion o he sys em as a whole and p opaga es changes in he sca ci y o esou ces h oughou he sys em. The esul ing pa e ns a e compa able o hose wi nessed in human ma ke nego ia ion expe imen s [KR95][P u81][Smi62]. CHAPTER 3. FORMAL DESCRIPTION OF MECHANISMS 13 3.2.1 Se up and Va iables De ini ion While he no a ion o buye s, selle s and goods is he same as he one used in he cen al- ized scena io, we need o add de ini ions o he decision-making p ocess ( he s a egy) o he agen s. The nego ia ion s a egy desc ibed he e is based on he AVALANCHE s a egy [ESP98][Eym01]. The s a egy consis s o 5 basic pa ame e s, which de ine he indi idual beha io (geno ype) o each agen . Fo e e y adable good he e a e wo ypes o agen s, buye s and selle s. Le agen k be a buye and agen a selle o a adable good. Le ikbe he numbe o nego ia ions ha agen khas s a ed and i he numbe o nego ia ions ha agen has s a ed. A geno ype de ines he beha io o he agen s in he nego ia ion s a egy. Le he geno ype o agen ∗ o ∗=k, du ing his nego ia ion i∗be Gi∗ ∗∈[0; 1]5 wi h Gi∗ ∗=(Gi∗ ∗,1,...,G i∗ ∗,5)τ=(ai∗ ∗,s i∗ ∗, i∗ ∗,b i∗ ∗,wi∗ ∗)τ whe e ai∗ ∗acquisi i eness si∗ ∗sa is ac ion i∗ ∗p iceS ep bi∗ ∗p iceNex wi∗ ∗weigh Memo y. Acquisi i eness de ines he p obabili y o s icking wi h he las o e made, and no o make an unila e al concession in he ollowing nego ia ion s ep. The alue in e al is be ween 0 and 1, and will be challenged by a s ochas ic p obe in e e y nego ia ion s ep. A alue o 0.7 means a p obabili y o 70% ha he agen will no make a concession – a highly compe i i e s a egy. An agen wi h acquisi i eness alue 1.0 will ne e change his p ice and an agen wi h acquisi i eness alue 0.0 will always make an unila e al con- cession. I he p obe succeeds, a buye agen will ise his o e , a selle agen will lowe his p ice. The exac change o he bid alue is de ined by he concession le el (p iceS ep). The concession le el is ep esen ed by a pe cen age o he di e ence be ween he ini ial s a - ing p ices. A alue o p iceS ep = 0.25 means a compu a ion o he concession le el as 1/4 o he i s s a ed di e ence. I bo h opponen s a e homogenously nego ia ing and always concede, hey mee each o he on he hal way in he hi d nego ia ion ound unde he assump ion o no nego ia ion abo ion. CHAPTER 3. FORMAL DESCRIPTION OF MECHANISMS 14 Ob iously, wi h an acquisi i eness le el se high, and a p iceS ep se low enough, he opponen s migh ne e each an ag eemen . The sa is ac ion pa ame e de e mines i an agen will d op ou om an ongoing nego ia ion. The mo e s eps he nego ia ion akes, o he mo e excessi e he pa ne ’s o e s a e, he soone he nego ia ion will be discon inued. E ec i ely, his pa ame e c ea es ime p essu e. Like o acquisi i eness, i does his by doing a s ochas ic p obe agains a se alue be ween 0 and 1. A sa is ac ion alue o 0.75 means, ha he agen has a chance o 75% o con inue he nego ia ion p ocess. An agen wi h sa is ac ion = 0.0 will abo all nego ia ion a once and an agen wi h sa is ac ion = 1.0 will ne e abo . The las piece o he s a egy is an exp ession o sel ishness. Behind each success ul nego ia ion lies a u u e oppo uni y o gaining mo e o he u ili y sha e, by nego ia ing ha de . p iceNex hus modi ies he s a ing bid. A success ul selle will inc ease his o e p ice, a success ul bidde will s a wi h a lowe bid nex ime. Fo a iable s a egy, he pa icipan s will ha e a close eye on wha o he s deem o be he ma ke p ice. I no , hey isk being agged as ”excessi e” and hei bids will ail he sa is ac ion p obe. They hus weigh cu en p ice in o ma ion and his o ic p ice in o ma ion in a speci ied a io weigh Memo y, balancing sho - ime p ice luc ua ion and longe - e m oppo uni ies. A he beginning o he simula ion he genes Gi∗ ∗,j o ∗=k, and j∈{1,...,5}a e dis ibu ed acco ding o he p obabili ies: U o[mj−δj;mj+δj] The eby, he cons an s mjand δj o j∈{1,...,5}a e de ined so ha [mj−δj;mj+ δj]⊂[0; 1] . Addi ionally, each agen ∗has he ollowing a iables: Mi∗ ∗: he ma ke p ice, which is es ima ed by agen kdu ing his nego ia ion i∗. Pi∗ ∗: he p ice o he he las success ul nego ia ion 1,2,...,i ∗o agen ∗. Oi∗ ∗: he las o e , which he nego ia ion opponen has made in nego ia ion numbe i∗ he agen ∗ be o e he nego ia ion ended. pi∗ ∗: he numbe o s o ed plumages o agen ∗ di ec a e his nego ia ion i∗. 3.2.2 The Nego ia ion S a egy When agen kand agen nego ia e, agen kis he buye and agen he selle . The se- quence (Pj)j∈IN 0⊂[0,∞[cons i u es he o e in ch onological o de . The buye always CHAPTER 3. FORMAL DESCRIPTION OF MECHANISMS 15 makes he i s o e . This means, all o e s P2m∀m∈IN 0 o igina e om he buye and he o e s P2m+1 ∀m∈IN 0 come om he selle , whe e m is he nego ia ion ound. A he beginning o a nego ia ion he buye kde e mines his ini ial p ice Kand his maximum p ice K: K=Mik k·(1 −bik k), K =Mik k The selle de e mines his s a ing p ice Vand his minimum p ice V: V=Mi ·(1 + bi ),V=Mi The buye s a s wi h he i s bid: P0=K •Fi s Case: K≥V Then o e s also P1=K and he nego ia ion will be closed success ully o he p ice P1. •Second Case: K< V Then o e s his ini ial p ice P1=V. Bo h agen s de e mine now hei s eps δj∗ ∗ o p ice concessions: δi∗ ∗=(V−K)· i∗ ∗ o ∗=k, In he subsequen nego ia ion ounds, le A1,A 2,A 3,... and S1,S 2,S 3,... be s ochas ic independen andom a iables wi h he ollowing binomial dis ibu ions: A2m=1wi h p obabili y aik k 0wi h p obabili y 1−aik k ∀m∈IN A2m+1 =1wi h p obabili y ai 0wi h p obabili y 1−ai ∀m∈IN S2m=1wi h p obabili y sik k 0wi h p obabili y 1−sik k ∀m∈IN S2m+1 =1wi h p obabili y si 0wi h p obabili y 1−si ∀m∈IN CHAPTER 3. FORMAL DESCRIPTION OF MECHANISMS 16 •O e numbe 2m; i is he buye ’s k u n: I S2m=0and P2m−1≥P2(m−1)−1wi h m=1, hen he buye kcancels he nego ia- ion. This means, Oik k=P2m−1and Oi =P2(m−1) . O he wise, he buye kmakes he ollowing o e : P2m=min K,(P2(m−1) +δik k),P 2m−11−A2m · P2(m−1)A2m •Bid numbe 2m+1; i is he selle ’s u n: I S2m+1 =0and P2m≤P2(m−1), hen he selle cancels he nego ia ion. Tha means, Oi =P2mand Oik k=P2(m−1)+1 . O he wise he selle makes he ollowing o e : P2m+1 =min V , (P2(m−1)+1 −δi ),P 2mA2m+1 · P2(m−1)1−A2m+1 The nego ia ion ends i ei he one o he agen s cancels he nego ia ion o he nego ia ion ends success ully wi h Pj=Pj+1 o a j∈IN . In his case, i holds Oik k=Pj=Oi . Wi h he end o a success ul nego ia ion o he p ice Pj he nego ia ion compu e hei es ima ed p o i Πik k=Mik k−Pj espec i ely Πi =Pj−Mi . Addi ionally, bo h agen s upda e a e e e y nego ia ion hei es ima ed ma ke p ice using Mik+1 k=wik k·Oik k+(1−wik k)·Mik k espec i ely Mi +1 =wi ·Oi +(1−wi )·Mi . This las s ep is independen o he success o a nego ia ion. CHAPTER 3. FORMAL DESCRIPTION OF MECHANISMS 17 3.2.3 Gossip Lea ning The lea ning concep used in his simula ion is de i ed om so-called gossip lea ning. This means ha he agen s lea n om ecei ed in o ma ion abou o he ansac ions in he ma ke . This in o ma ion may no be accu a e o comple e, bu se es as an indica ion abou he g oss di ec ion o he ma ke . In ou implemen a ion, his gossip in o ma ion is c ea ed and b oadcas by a success ul agen , in analogy o issuing an ad-hoc in o ma ion in s ock ma ke pe iodicals. Le nbe an agen and g1,...,g d he adable goods. The agen nhas inished his ne- go ia ion insuccess ully wi h an es ima ed p o i o Πin n(g) o he good g∈{g1,...,g d}. Alea ning s ep acco ding o he lea ning algo i hm (see subsec ion 3.2.4) is pe o med by agen nlas ime a he end o his nego ia ion jk. This means Gjn+1 n=Gjn+2 n=···=Gin n. I agen nwi h he nego ia ion numbe s jn+1,j n+2,...,i n has success ully comple ed a leas 10 nego ia ions o e e y good, he sends a Plumage (Gin n,Fin n) o all o he agen s o his ype. Then, his upda ed i ness is Fin n, which is compu ed as ollows: (a) Fo e e y good gj∈{g1,...,g d} he nex p o i alue Π(gj)is de e mined: Le Π1(gj),...,Π10(gj) be he es ima ed p o i s o he las 10 success ul nego ia ions o agen n o he good gj. Then, he i ness is Fin n(gj)= 1 10Π1(gj)+···+Π 10(gj). (b) The upda ed i ness Fin n inally is Fin n=1 dΠ(g1)+···+Π(gd). 3.2.4 The Lea ning Algo i hm I is assumed ha he agen s show a coope a i e beha io . This means, he agen s epo u h ully hei lea ning in o ma ion. CHAPTER 3. FORMAL DESCRIPTION OF MECHANISMS 18 A e ha ing ecei ed some gossip in o ma ion message, he agen may modi y his own s a egy. The compa ison o he own esul s wi h hose o he s a egy ecei ed may show ha he o he s a egy supe io o he own. In his case, he agen will y o c oss bo h s a egies o gain compe i i e ad an age. In p ac ice, ou o a lis o ecei ed geno ype/pe o mance- uples, he agen will choose he bes pe o ming ex e nal geno- ype, and hen mix, c oss and mu a e wi h his own geno ype. Le be nan a bi a y agen a he end o his nego ia ion inand le be pin n he numbe o plumages, he agen nhas s o ed di ec ly a e his nego ia ion in. The las lea ning s ep was pe o med by agen na e his nego ia ion jk. Le be ein n he numbe o nego ia ions, an agen no he nego ia ion numbe s jn+1,j n+2,...,i n has success ully inished. Le be p=1. I pin n<p o ein n<10 applies o agen na e his nego ia ion in,nolea ning s ep will be pe o med. This means, his geno ype will no change: Gin+1 n=Gin n. Hence, i pin n≥pand ein n≥10 applies, he agen npe o ms a lea ning s ep. The geno ype o agen nchanges as ollows: Fi s , he s o ed plumage o agen nwi h he highes i ness is selec ed. Le be G =(G ,1,...,G ,5)τ=(a ,s , ,b ,w )τ he ela ed geno ype. Second, a c osso e is pe o med. In doing so, a new geno ype ˜ Gin+1 nis c ea ed, which con ains a andom mix u e o genes o he geno ypes Gin nand G . This p ocess ollows a mu a ion s ep hi d: Using he geno ype ˜ Gin+1 nand changing i s genes sligh ly will esul in he geno ype Gin+1 n. 3.2.4.1 C osso e Le be C1,...,C 5s ochas ic independen andom a iables wi h he ollowing binomial dis ibu ion: Cj=1wi h p obabili y 0,5 0wi h p obabili y 0,5∀j∈{1,...,5} CHAPTER 3. FORMAL DESCRIPTION OF MECHANISMS 19 Then i is impe a i e ˜ Gin+1 n,j =(1−Cj)·Gin n,j +Cj·G ,j ∀j∈{1,...,5}. 3.2.4.2 Mu a ion Le be M1,...,M 5,X 1,...,X 5s ochas ic independen andom a iables wi h he ollow- ing dis ibu ions: Mj=1wi h p obabili y 0,05 0wi h p obabili y 0,05 ∀j∈{1,...,5} Xj∼N(0 ,1) ∀j∈{1,...,5} Tha means, Xjis ∀j∈{1,...,5}s anda d no mal dis ibu ed. Then, i holds Gin+1 n,j =max0; min˜ Gin+1 n,j + Mj·(1 10Xj)mod(1);1 ∀j∈{1,...,5}. Chap e 4 Bidding Issues The pu pose o his chap e is o cla i y wha and when an agen bids in he CATNETS scena io. Wha deno es he alua ion and he ese a ion p ices o agen s, i.e. he maximal p ice which an agen is willing o pay o a se ice ( esp. he minimum p ice an agen has o selling a se ice). When deno es he iming o bids, i.e. which e en induces an agen o bid o a se ice. Bo h cases a e di e en in he cen alized and decen alized scena io. As such, i is impo an o ind concep s ha a e applicable o bo h scena ios and, hus, make he esul s compa able. The chap e is s uc u ed as ollows: Sec ion 4.1 ou lines he alua ion gene a ion o an agen , i.e. he p ocedu e ha de e mines he alue o an agen ’s bid. Sec ion 4.2 desc ibes he iming o he agen ’s bids, i.e. when does an agen bid o a se ice. Finally, Sec ion 4.3 summa izes he chap e . 4.1 Wha Does an Agen Bid? The ollowing sec ion desc ibes wha an agen bids, i.e. he alua ion and ese a ion p ices. Fo his, a gene ic unc ion is de eloped ha is applicable o bo h, he cen alized and he decen alized case. The concep is applied o buye s and selle s in bo h ma ke s, i.e. in he se ice ma ke and in he esou ce ma ke . 4.1.1 No a ion Be o e he alua ion gene a o is in oduced, he gene al no a ion as deno ed in able 4.1 is p esen ed. The ansac ion objec gdeno es he se ice o ha a alua ion is o be gene a ed. 20 CHAPTER 4. BIDDING ISSUES 21 T ansac ion objec g Valua ion in pe iod i V i Ma ke p ice Mi Weigh ma ke p ice β Weigh ed A e age wa i Weigh ed Memo y wi Table 4.1: No a ion o he alua ion gene a ion Fo ins ance, his could be a PDF c ea o in he se ice ma ke . Fo each se ice, an agen has a alua ion Vi( esp. ese a ion p ice) in each pe iod i. This alua ion deno es he maximum p ice, an agen is willing o bid o his pa icula se ice. The alua ion gene a ion is in luenced by ex e nal ac o s such as he ma ke p ice. In case such a ma ke p ice exis s o a ansac ion objec in pe iod i, i is deno ed by Mi. Fo he cen alized case, ma ke p ices may no exis s in each pe iod. In hese cases, an app oxima ed ma ke p ice is used. The e ec a ma ke p ice has on he alua ion gene a ion is deno ed by β. The lowe his alue is, he lesse he impo ance o old ma ke p ices. Finally, he weigh ed a e age wa iand he weigh ed memo y wideno e noise pa am- e e s. 4.1.2 Valua ion Gene a ion Based upon he no a ion as in oduced in he p e ious sec ion, he alua ion o a se ice gin ime pe iod i+1is calcula ed as ollows: Vi+1(g)=βMi(g)+(1−β)wa i(g)+YX +Z(4.1) The alua ion o he nex pe iod depends on he ma ke p ice o he cu en pe iod, he weigh ed a e age o o me alua ions, and some s a is ical noise. The weigh o he ma ke p ice and he weigh ed a e age depends on he s a ic alue β∈{0,1}which is p ede ined and ixed. F om an implemen a ion poin -o - iew, he a iable βshould be de inable ia an ex e nal con igu a ion ile. I is o no e, ha he s a is ical noise unc ions a e only applied in he cen alized case. These unc ions a e esponsible o inse ing exogenous ac o s (e.g. di e en dynamics, densi y scena ios) in he cen alized simula ion. In he decen alized case, his noise is gene a ed by a gene ic algo i hm. Howe e , his gene ic algo i hm will no be applied CHAPTER 4. BIDDING ISSUES 28 dle {A, B, C}would be highe han he sum o alua ions o he esou ces {A},{B} and {C}. 4.3 Summa y This sec ion ou lines bidding issues in he CATNETS scena io. A alua ion gene a o is in oduced ha can be applied o buye s and selle s in bo h ma ke s in o de o de e mine alues o hei bids. The challenge o such a gene a o is o de ine a concep ha is applicable o he cen alized and he decen alized case and ha leads o compa able ou comes. Chap e 5 In eg a ion o Mechanisms in o Simula o Subjec o his chap e is he echnical in eg a ion o he cen alized and decen alized mechanisms in o he simula o (Sec ion 5.1 and Sec ion 5.2). He e, a s ing in e ac ion wi h WP2 will be ca ied ou in o de o a oid o e laps in documen a ions. Fu he mo e in Sec ion 5.3 an insigh in o he esul s o he compa ison o he cen alized o he decen- alized mechanism is gi en. Fo an in dep h analysis he eade is e e ed o [BCC+07]. 5.1 In eg a ion o he Auc ion Mechanisms The objec i e o his sec ion is o desc ibe he implemen a ion o he auc ion mechanisms (Sec ion 5.1.1) and hei in eg a ion in o Op o Sim (Sec ion 5.1.2). 5.1.1 Implemen a ion o he Ma ke s In he ollowing, he implemen a ion o he se ice ma ke and he esou ce ma ke is desc ibed. Bo h ma ke mechanisms a e implemen ed as independen so wa e se ices which allows us o in eg a e hem in o o he sys ems easily. Beside he in eg a ion in o Op o Sim, his lexibili y allows us o in eg a e he ma ke s in o he p o o ype in he u u e such as p oposed in [CJSF06]. 5.1.1.1 Se ice Ma ke As ou lined in he las deli e able [SNV+05a], a double auc ion is applied o he se ice ma ke . In a double auc ion ma ke [F i91], a la ge numbe o pa icipan s ade a common objec and can submi bids (buy o de s) and asks (sell o de s). T ading in 29 CHAPTER 5. INTEGRATION OF MECHANISMS INTO SIMULATOR 30 double auc ions is o ganized by means o o de books, each o a se o homogeneous goods. In he CATNETS scena io he e will be ndi e en o de books, each o one o he ndi e en se ices. Figu e 5.1 depic s he high le el a chi ec u e o he se ice ma ke o CATNETS [SNV+05a]. Complex se ice agen s can submi buy o de s o he o de books; basic se ice agen s can submi sell o de s. Each se o homogeneous se ices (e.g. PDF c ea o se ices) is aded in a single o de book. Figu e 5.1: The se ice ma ke including se e al double auc ion o de books. A simpli ied class diag am o he se ice ma ke implemen a ion is shown in ig- u e 5.2: Fo each ype o basic se ice aded in he se ice ma ke , an ins ance o he O de book class is gene a ed. The o de book p o ides unc ionali y o add o de s, e- mo e hem, and o s a he ou come de e mina ion. Whene e an agen wan s o submi an o de o he ma ke , i gene a es an ins ance o he O de class and submi s i o he o - de book. The o de book is also esponsible o igge ing he clea ing p ocess. In case, a con inuous clea ing is used, he o de book ins an ia es he Alloca o CDA class; o he wise i uses he call ma ke as implemen ed in he Alloca o CallMa ke class. Bo h alloca o classes use he ma chmake Ma ch o ind co esponding coun- e pa o de s. A e he alloca ion and he p ices a e compu ed, an Alloca ion objec is gene a ed o each ansac ion. This objec poin s o he pa ies ha a e in ol ed in he ansac ion, i.e. i poin s o an o de om a buye and an o de om a selle . The alloca ion objec s a e s o ed in a ec o and can be pa sed by he simula o . As he diag am shows, he implemen a ion suppo s con inuous clea ing and a call ma ke . In a con inuous clea ing auc ion, buye s and selle s simul aneously and asyn- ch onously announce bids and o e s. In case a new o de en e s he ma ke , he auc- ionee ies o clea he ma ke immedia ely. A call ma ke is an auc ion wi h pe iodic CHAPTER 5. INTEGRATION OF MECHANISMS INTO SIMULATOR 31 Figu e 5.2: Class diag am o he mos undamen al classes o he se ice ma ke uni o m clea ing, e.g. he auc ionee clea s he ma ke e e y i es minu es. All o de s in a pe iod a e collec ed in an o de book and will be clea ed pe iodically [SNV+05a]. The clea ing s a egy ha is applied can be selec ed by means o a con igu a ion ile. Till he end o Y3 we did no succeed in implemen ing he e en d i en ime model o he con inuous clea ing auc ion in o he simula o . So i was no possible o speed up simula ion uns om eal ime. The e o e we chose o use he con inuous clea ing auc ions o he simula ion uns in Y3. Tha enabled us o calib a e and e alua e he cen alized app oach on a lage da a se . 5.1.1.2 Resou ce Ma ke Fo alloca ing se ices in he esou ce ma ke , we apply a mul i-a ibu e combina o ial exchange (MACE) [SNV+05a, SNVW06]. Figu e 5.3 depic s he sequence o he auc ion in he CATNETS scena io. Agen s (buye s and selle s) submi hei bids o he auc ionee ins ance. A e ha , he bids a e ans o med in o an in e nal ep esen a ion o m and, subsequen ly, he winne s a e compu ed (alloca ion). Finally, p ices a e compu ed in conside a ion o he alloca ion. As a esul o he ma ke mechanism, he agen s ge in o med whe he o no hey a e pa o he alloca ion. Figu e 5.4 depic s some o he mos basic componen s o he implemen a ion as a UML class diag am: The Ma ke class is he cen al componen o he implemen a ion. On he one hand, i is esponsible o ini ializing all ele an classes. On he o he hand, he ma ke class con ols he p ocess o submi o de s and o compu e an ou come. Agen s can submi o de s o an o de book, whe e an o de consis s o a p ice and a CHAPTER 5. INTEGRATION OF MECHANISMS INTO SIMULATOR 32 Figu e 5.3: Sequence o an auc ion o he esou ce ma ke Bundle, whe e a bundle is a collec ion o Good ins ances. A e he bids a e submi ed by he agen s, an Ou come is compu ed. Fo his, he ma ke uses a ModelFac o y and a P icingFac o y. The ModelFac o y is esponsible o p o iding a winne de e mina ion model in o de o compu e an alloca ion. In he CATNETS scena io, his model is implemen ed in he Ca ne sResou ceMa ke class. The P icingFac o y p o ides a se o p ice mechanisms ha can be used. In CATNETS, he p icing schema as implemen ed in he KP ice class is applied. A e an alloca ion and co esponding p ices a e de e mined, he esul is s o ed in he Ou come objec . This objec can be que ied in o de o e ie e he equi ed in o ma ion such as alloca ion decisions and p ices o each ansac ion. Fo implemen ing he esou ce ma ke , we make use o he s anda d linea p og am- ming lib a ies CPLEX and LPSol e. CPLEX is a comme cial p oduc and is cu en ly he s a e o he a op imiza ion engine1. LPSol e is a ee linea p og amming sol e ha implemen s he b anch-and-bound me hod o sol ing in ege p oblems2. CPLEX is cu en ly one o he as es sol ing lib a ies and will be used o he e alua ion o he mechanisms. The use o LPSol e (mo e speci ically, i s license) allows us o ins all he esou ce ma ke implemen a ion on e e y machine. As such, he de elopmen and es ing o he mechanisms can be os e ed. The beha io o he implemen a ion can be con olled by means o con igu a ion iles. Among o he s, se e al al e na i e p icing schemas a e implemen ed, e.g. he app oxi- ma ed Vick ey p icing algo i hm [PKE01]. The conc e e p icing mechanism can be se- lec ed by means o a con igu a ion pa ame e . Fu he mo e, he clea ing in e al3o he 1See h p://www.cplex.com/ o de ails. 2See h p://www.geoci ies.com/lpsol e/ o de ails. 3The esou ce ma ke is cu en ly es ic ed o a pe iodical clea ing. CHAPTER 5. INTEGRATION OF MECHANISMS INTO SIMULATOR 33 Figu e 5.4: Class diag am o he mos undamen al classes o he esou ce ma ke ma ke can be con olled by he con igu a ion ile. 5.1.2 In eg a ion in o Op o Sim This sec ion b ie ly ou lines he in eg a ion o he ma ke s in o Op o Sim. Fo a de ailed desc ip ion o how hese concep s a e implemen ed, he eade is e e ed o deli e able D2.2 [CSSZ06]. In Op o Sim, he auc ionee s o he se ice ma ke and esou ce ma ke a e bo h ealized as agen s. They ge ins an ia ed by he simula o du ing i s ini ializa ion and can be con ac ed by e e y o he agen . Agen s communica e wi h he auc ionee s in o de o submi hei bids and o e ie e s a us in o ma ion such as he las ma ke p ice o a se ice o he alloca ion decision. The communica ion be ween ading agen s and auc ionee s is ealized by means o messages. Figu e 5.5 shows he gene al in e ac ion be ween ading agen s and an auc ionee . Each ime, a complex se ice agen (CSA) wan s o acqui e o se ice, i submi s a message o i s Pee - o-Pee Manage (P2P). This manage is capable o ad e ising and ou ing messages o o he agen s. In case a manage ecei es a message om i s CSA, i CHAPTER 5. INTEGRATION OF MECHANISMS INTO SIMULATOR 34 o wa ds he message o he auc ionee (SMAA). Likewise, a basic se ice agen (BSA) can also submi an o e message o i s Pee - o-Pee Manage which also o wa ds i o he auc ionee . On he basis o he messages, he auc ionee compu es an ou come, i.e. alloca ion decisions and p ices. The esul o his ou come (success ul o unsuccess ul bid) is subsequen ly sen back o he agen s. The igu e shows he p ocess o he se ice ma ke exempla ily; o he esou ce ma ke he p ocess is iden ical. Figu e 5.5: Sequence o an auc ion o he esou ce ma ke . The cen al in e aces be ween Op o Sim and he auc ionee s a e messages. F om a concep ual poin o iew, di e en message ypes a e equi ed o di e en ac ions o he agen s: Reques Message: A eques message is sen , whene e an agen wan s o buy a se ice. Fo ins ance, a complex se ice agen submi s such a message o he auc ionee o bid o a basic se ice. The message con ains in o ma ion abou he ansac ion objec (which se ice), he agen ’s alua ion p ice (maximum p ice), as well as an ID o he agen . O e Message: An o e message is sen , whene e an agen wan s o sell a se ice. Fo ins ance, a esou ce se ice agen submi s such a message when i wan s o sell i s esou ces. In analogy o he eques message, he o e message con ains in o ma ion abou he ansac ion objec (which se ice), he agen ’s ese a ion p ice (minimum p ice), as as well as an ID o he agen . Alloca ion Message: A e he auc ionee has compu ed an ou come, alloca ion mes- sages a e sen back o each pa icipa ing agen . In case he agen was success ul CHAPTER 5. INTEGRATION OF MECHANISMS INTO SIMULATOR 35 in he auc ion (i.e. i is pa o he alloca ion), he message con ains in o ma ion abou he p ice o he se ice and i s coun e pa . Fo ins ance, i a basic se ice agen e ie es a success ul alloca ion message om he esou ce ma ke auc ion- ee , he message con ains in o ma ion abou he esou ce se ice agen who will p o ide he esou ces. In case an agen was unsuccess ul, he con ains a nega i e p ice (p=−1). Dele e Message: Some imes agen s need o cancel o de s which hey ha e submi ed o he auc ionee . In his case, hey submi a dele e message o auc ionee . This message con ains all ele an in o ma ion such as he agen ID and an o de ID. Each message ype is implemen ed o he se ice ma ke and he esou ce ma ke . The implemen a ion o he messages is desc ibed in he deli e able D2.2 in mo e de ail [CSSZ06]. 5.2 Decen alized Mechanisms (Ca allaxy) This sec ion b ie ly ou lines he implemen a ion o he decen alized ma ke s and i s in e- g a ion in o Op o Sim. Sec ion 5.2.1 ocuses on he implemen a ion o he decen alized (ca allac ic) se ice and esou ce ma ke . The implemen a ion o he s a egy module and i s adop ion o he ma ke s a e desc ibed in de ail. The in eg a ion in o Op o Sim, message pa e ns and he in oduced message ypes desc ibes sec ion 5.2.2. 5.2.1 Implemen a ion o he Ma ke s The implemen a ion o he se ice and esou ce ma ke s use he ca allac ic easone im- plemen a ion p esen ed in he deli e able D1.1. Bo h ma ke s use he same s a egy im- plemen a ion o easoning abou p oposals. The e o e, he objec i e o his sec ion is o desc ibe he implemen a ion o he se ice and esou ce ma ke using he ca allac ic easone . Di e ences be ween he se ice and he esou ce ma ke occu a he ini ializa ion o he s a egy, hei in e ac ion pa e ns and he in eg a ion in o he simula o and p o o ype en i onmen . In he ollowing, he ocus lies on he in e aces o he s a egy and i s ini ializa ion. The in e ac ion pa e ns a e dependen on he en i onmen o be in eg a ed. Thus, hey a e desc ibed in hei co esponding sec ions. The in e ace o he easone shows igu e 5.6. The implemen a ion o e s cus- omized easone s o e e y agen ype in he CATNETS scena io, which ex end he Agen Sou ce easone empla e. The Agen Sou ce class is an implemen a ion o he bila e al nego ia ion p o ocol which calls he A alanche s a egy o decision making. Addi ionally, i p o ides access o he e olu iona y lea ning algo i hm. CHAPTER 5. INTEGRATION OF MECHANISMS INTO SIMULATOR 36 F om a concep ual poin o iew, he agen calls he easone wi h a Msg objec which con ains all in o ma ion acco ding o he bidding language p esen ed in he D1.1 deli - e able and some addi ional iden i ica ion in o ma ion. The s a egy in e p e s his objec (in e p e Message) and e u ns a Msg objec . This p ocess sepa a es he message p opaga ion om he easoning abou he con en using he de ined nego ia ion p o ocol. The unde lying in as uc u e anspo s he con en o i s des ina ion. The same concep is applied in he simula o and p o o ype en i onmen . «Ja a Class» Agen Sou ce CLASS_NAME : S ing s a egy : IP icing budge : double Agen Sou ce ( ) ge S a egy ( ) in e p e Message ( ) se P ice ( ) se Geno ype ( ) dec easeBudge ( ) inc easeBudge ( ) in e p e C p ( ) in e p e P oposal ( ) in e p e Accep ( ) in e p e Rejec ( ) pos Rejec anceMe hod ( ) pos Accep anceMe hod ( ) checkRes ic ions ( ) inc easeP iceDis ibu ion ( ) dec easeP iceDis ibu ion ( ) lea n ( ) in e p e Plumage ( ) «Ja a Class» ComplexSe iceReasone ComplexSe iceReasone ( ) pos Rejec anceMe hod ( ) pos Accep anceMe hod ( ) execu e ( ) «Ja a Class» Ca allac icReasone c ea eC p ( ) c ea eResou ceC p ( ) checkRes ic ions ( ) pos Accep anceMe hod ( ) pos Rejec anceMe hod ( ) «Ja a Class» BasicSe iceReasone BasicSe iceReasone ( ) BasicSe iceReasone ( ) pos Rejec anceMe hod ( ) pos Accep anceMe hod ( ) execu eSe iceMa ke ( ) execu eResou ceMa ke ( ) «Ja a Class» Resou ceReasone Resou ceReasone ( ) pos Rejec anceMe hod ( ) pos Accep anceMe hod ( ) execu e ( ) «Ja a Class» Msg se ialVe sionUID : long messageType : in documen Type : in i emID : S ing p ice : double budge : double basicSe ices : BasicSe iceDa a e dic : in msg_plumage : Plumage hopCoun e : in basicSe ice : BasicSe iceDa a ma ke : S ing smMessage : Objec eqMessage : Objec «use» «use» «use» «use» «use» Figu e 5.6: The in e ace o he ca allac ic easone . The ele an a ibu es o he Msg class a e in de ail: CHAPTER 5. INTEGRATION OF MECHANISMS INTO SIMULATOR 37 MessageType:The ype o he nego ia ion message speci ies he communica i e ac e e ed o he nego ia ion p o ocol. In he bila e al nego ia ion p o ocol o he ca allac ic easone he ollowing alues a e used: c p (call- o -p oposal), accep , ejec , p oposal. Con e sa ionID:The iden i ie o he nego ia ion is an unique numbe gene a ed o each new eques du ing he c ea ion o new c p message. The alues a e andom gene a ed UUIDs o he Ja a buil -in UUID gene a o . Documen Type:This pa ame e signals he ca allac ic easone he ype o p ice o eason abou . A BID e e s o he p oposal ype o be gene a ed by a selle and an ASK ela es o a buye o e . I emID:This is he iden i ie o he aded good. In he cu en implemen a ion, his could be any kind o ex . In Op o Sim, se ices usually a e iden i ied using hei ype (cs o bs o complex se ice o basic se ice) ollowed by a numbe . The bundles on he esou ce ma ke a e iden i ied using a sequence o hei single i ems. Fo example, ”cpu;mem;hdd” e e s o a bundle o he h ee single bundle i ems cpu, memo y (mem) and ha d disk space (hdd). P ice:This is he cu en p ice o he aded good. On he se ice ma ke , i is he p ice o a basic se ice ins ance; on he esou ce ma ke i is he p ice o a esou ce bundle. In he cu en implemen a ion, he e is a p ede ined p ice o e e y e- sou ce bundle combina ion. The agen has a p e e ence o ading ce ain esou ce bundles. Ma ke :The ma ke pa ame e is used o assign he message o a ma ke . In he CAT- NETS scena io, he alues a e: SERVICEMARKET, RESOURCEMARKET Ve dic :This is he e dic on bid se by he s a egy. I ela es o he esul o he ca allac ic easone . Valid alues a e: accep , ejec , p oposal Plumage:The plumage pa ame e ep esen s a con aine o lea ning in o ma ion. This con aine includes he i ness in o ma ion and he geno ype alues o o he agen s which aded he same good. The ma ke agen s use he desc ibed message objec o decide he nex ope a ion. This ope a ion is dependen on he message ype o he ma ke agen and i s ole in he ma ke . I s gene al bidding beha io o ming he wo ma ke is desc ibed in sec ion 4.2. The implemen a ion o he agen s ollows his desc ip ion. The nex sec ion p esen s he ini ializa ion o s a egy o he wo ma ke s in he CATNETS scena io. ini ializa ion o he s a egy o he se ice and esou ce ma ke : The ini ializa ion o he easone spli s in o wo a eas: The i s a ea is he ini ializa ion o he i ems and CHAPTER 6. RELATIONS TO OTHER WPS 44 6.3 WP4 The goal o WP4 is o e alua e he pe o mance o he Ca allac ic app oach by means o a simula o (see deli e able yea 2 o WP2) and a p o o ype (see deli e able yea 2 o WP3). The ela ions o WP4 a e as ollows: •The iden i ica ion and o maliza ion o ele an me ics o compa e cen alized and decen alized ma ke mechanisms. •Collabo a ion wi h he de elopmen o he pe o mance measu ing amewo k, due o he ac ha some o he measu ed me ics a e aken a he economic agen le el. Repo ing o measu ed da a o he pe o mance measu ing amewo k. Chap e 7 Summa y In his chap e he achie emen s om wo kpackage 1 du ing Y2 and Y3 o he p ojec a e summa ized. Sec ion 7.1 ocuses on he wo k ha has been pe o med in Y2 and Y3. A b ie e iew is gi en on how his ela es o he i s p ojec yea . In Sec ion 7.2 he asks pe o med in Y3 a e speci ied. Finally sec ion 7.3 pu s he CATNETS esul s in o a g ea e pe spec i e. 7.1 Re iew As desc ibed in he in oduc ion and in he p ojec summa y, he main con ibu ion o he CATNETS p ojec is he compa ison o cen alized and decen alized economic al- loca ion mechanisms o esou ces in G ids and applica ion laye ne wo ks (ALNs). In o de o achie e his, he p ojec has been di ided in o i e wo kpackages. The con en o he indi idual wo kpackage as well as he di ision and in eg a ion o hose has been elabo a ed on in Chap e 1 and Chap e 6. In his epo , he line o wo k ca ied ou in he second p ojec yea (mon h 25 o mon h 48) is desc ibed. The eby, Chap e 3 ocuses on he o mal desc ip ion o he cen alized and decen alized mechanisms. The cen alized mechanisms, which also ha e been con en o yea 1 epo , a e b ie ly discussed and an in dep h p esen a ion o he decen alized mechanisms as well as he lea ning algo i hms and nego ia ion s a egies is gi en. The no ion o wo di e en ma ke s – a esou ce and a se ice ma ke – is in oduced. Bo h ma ke s a e in e connec ed ia a in e media ies. A e his, Chap e 4 cla i ies when and wha a pa icipan in esou ce and se ice ma ke s bids. The de ini ions p o ided he e a e subs an ial o he design o bo h (i) he implemen a ion o he mechanisms in he simula o as well as (ii) he in eg a ion o he decen alized mechanisms in o he middlewa e. In Chap e 5 he in eg a ion o he cen alized and decen alized mechanisms in o he 45 CHAPTER 7. SUMMARY 46 Op o Sim simula o amewo k is desc ibed. Besides he p epa a o y wo k ha is con en o he chap e s be o e, his has been he mos demanding and ex ensi e wo k ha has been ca ied ou in wo kpackage 1 in yea 2. The challenge he e is o p o ide a lexible, adap able and dynamic simula ion amewo k ha enables bo h, simula ions based on cen alized and decen alized ma ke se ups in one scena io in o de o keep he esul s compa able. Addi ionally o he issues conce ning he implemen a ion i sel , simula ion esul s om bo h, he cen alized and he decen alized case a e compa ed o each o he . 7.2 Con en o Y3 A e he in eg a ion o he cen alized and he decen alized economic mechanisms in o he simula o and he in eg a ion o he cen alized mechanism in o he middlewa e was inished and e ined he ollowing asks we e he main issues o wo kpackage 1: •2.3 Simula ion o applica ion laye ne wo ks and e inemen : He e, he e o s in wo k- package 1 was ocused on he calib a ion, alida ion and e i ica ion o he simula ion model. Addi ional issues we e he assis ance o wo kpackage 2 leade s in ca ying ou simula ions, ob aining and e alua ing la ge-scale da a om simula ion uns as well as o in e p e he esul s de i ed om he applied me ics. •3.3 Pe o mance measu ing componen s o expe imen s: Conce ning his issue, mos wo k is ca ied ou in wo kpackage 3. Howe e , he economic expe ise in designing and moni o ing he pe o mance e alua ion bo h, om a echnical and an economic pe spec i e has been ou mission in yea 3. •3.4 Dis ibu ed applica ion o execu e on economic-enhanced G id/P2P pla o m and middlewa e in eg a ion: In his ask, he in eg a ion o middlewa e concep s o no el app oaches in G id and P2P a chi ec u es was e alua ed. Besides gene a ing sub- s an ial esul s, he a ge o wo kpackage 1 was he o assis wo kpackage 3 lead- e s in p o isioning o a s ong oo p in o he wo k pe o med in CATNETS in he G id/SOA/P2P/Pe asi e Compu ing communi ies. •4.4 Pe o mance analysis, compa ison, e alua ion: As o all o he pa icipan s, i was one o he co e issues o analyze, documen and compa e he e alua ion o he p oposed mechanisms. Wo kpackage 1 con ibu ed o his e o by assis ing wo kpackage 4 p o- agonis s in o de o show he e iciency and e ec i eness o he p oposed mechanisms in app op ia e scena ios and ind channels o dis ibu e hese esul s in o all ele an communi ies. CHAPTER 7. SUMMARY 47 7.3 Ou look The co e con ibu ion o he CATNETS p ojec is he quan i a i e compa ison be ween common cen alized economic alloca ion mechanisms and decen alized nego ia ion o - ma s based on on Hayek’s Ca allaxy. The e o e, se e al me ics ha e been de ined in o de o iden i y he quali y o he economic alloca ions. The esul s show, ha he appli- cabili y o alloca ion mechanisms highly depends on he pa ame e iza ion o he indi id- ual se up. Co e issues along which he iden i ica ion o he app op ia e mechanism has o be aligned a a e: • he size o he alloca ion p oblem, • he communica ion in ensi y, • he dis ibu ion o he p ices o e ed by he pa icipan s and • he dynamics o he ma ke . All hese pa ame e s again depend upon he indus y b anch in which he indi idual applica ion, o which he mechanism should be deployed, is loca ed. The app oaches in es iga ed wi hin he CATNETS p ojec may be subsumed in he ield o G id Economics. In his ield, cu en ly s ong e o s a e bundled in o de o iden i y me hodologies ha a e applicable o dynamically alloca ing compu a ional e- sou ces o applica ions. The ision o his ield is o enable a seamless in eg a ion o dis- ibu ed ha dwa e o he compu a ion o he e ogeneous on -end applica ions. The idea is o allow o dynamic alloca ion o esou ces in o de o de e mine he p ices o compu- a ional esou ces along he ime. In o de o lay he undamen s o such an a chi ec u e, subs an ial e o s ha e o be ca ied ou in he G id middlewa e domain. Applying he e- sul s om CATNETS an icipa es a ully unc ional G id middlewa e, which is capable o ex-an e de e mining he ime speci ic jobs a e unning. This implies a componen , which judges he un ime o jobs s emming om he e ogeneous applica ions. Ha ing such a componen in place will allow he applica ion o ma ke based alloca ion schemes such as hey ha e been p oposed in his wo k. The key ideas o ou app oaches ha e been p esen ed in di e en communi ies. A la ge numbe o expe s om he e-In as uc u e communi y, he G id communi y as well as he SOA and he dis ibu ed sys ems communi ies see g ea po en ial in hese app oaches. Se e al in dep h coope a ions ha e been s a ed he e. Bibliog aphy [AG04] Nadia Ben Azzouna and Fab ice Guillemin. Cha c e is ic o ip a - ic in comme cial wide a ea ne wo ks. In P oceedings o he In e na- ional Con e ence on Compu ing, Communca ions and Con ol Technologies (CCCT’2004), Aus in, Texas (TX), Augus 2004. [AH00] E. Ada and B.A Hube man. F ee iding on gnu ella. Fi s Monday, 5(10), 2000. [BCC+07] Geo g Buss, Michele Ca alano, Pablo Chacin, Isaac Chao, To s en Ey- mann, Felix F ei ag, Sebas ian Hude , Li iu Joi a, Leand o Na a o, Nils Pa asie, Ome F. Rana, Bj¨o n Schnizle , We ne S ei be ge , and Daniel Vei . D4.3: Pe o mance e alua ion. Ca ne s deli e able, Uni e si y o Mannheim,Uni e si ´a delle Ma che Ancona, Uni e si a Poli ecnica de Ca alunya, Uni e si y o Bay eu h, Uni e si y o Ka ls uhe, Uni e si y o Ca di , 2007. [B e02] T. B enne . A beha iou al lea ning app oach o he dynamics o p ices. Com- pu a ional Economics, pages 67–94, 2002. [Ca 04] Nicholas G. Ca . Does IT Ma e ? In o ma ion Technology and he Co o- sion o Compe i i e Ad an age. Ha a d Business School P ess, May 2004. [CGS+05] Michele Ca alano, Gian anco Giulioni, We ne S ei be ge , Michael Reinicke, and To s en Eymann. 4.1: E alua ion and me ics amewo k. Ca - ne s deli e able, Uni e si ´a delle Ma che Ancona, Uni e si y o Bay eu h, 2005. [CJSF06] P. Chacin, L. Joi a, B. Schnizle , and F. F ei ag. Flexible a chi ec u e o sup- po ing auc ions in g ids. In P oceedings o he 2nd In e na ional Wo kshop On Sma G id Technologies 2006 (SGT2006) Wo kshop, 2006. [CSSZ06] Gae ano Calab ese, Bj¨o n Schnizle , We ne S ei be ge , and Flo iano Zini. D2.2: Annual epo o wp2. Ca ne s deli e able, ITC-i s T en o, Uni e si y o Ka ls uhe, Uni e si y o Bay eu h, 2006. 48 BIBLIOGRAPHY 49 [ESH07] To s en Eymann, We ne S ei be ge , and Sebas ian Hude . 5.6: Pe iodic p og ess epo . Ca ne s deli e able, Uni e si y o Bay eu h, 2007. [ESMP03] T. Eymann, S. Sackmann, G. M¨ulle , and I. Pippow. Hayek’s ca allaxy: A o wa d-looking concep o in o ma ion sys ems. In P oceedings o he Ame ican Con e ence on In o ma ion Sys ems (AMCIS), Tampa, Flo ida, 2003. [ESP98] To s en Eymann, De le Schode , and Bo is Pado an. A alanche - an agen based alue chain coo dina ion expe imen . In Wo kshop on A i icial Soci- e ies and Compu a ional Ma ke s (ASCMA’98), pages 48–53, Minneapolis, 1998. [Eym01] To s en Eymann. Decen alized economic coo dina ion in mul i-agen sys- ems. In Hans-Ul ich Buhl, F. Hu he , and A. Rei wiesne , edi o s, In o - ma ion Age Economy. P oceedings WI-2001., pages 575–588, Heidelbe g, 2001. Physica Ve lag. [F i91] D. F iedman. The double auc ion ma ke ins i u ion: A su ey. In D. F ied- man and J. Rus , edi o s, The Double Auc ion Ma ke - Ins i u ions, Theo ies, and E idence, pages 3–26. Camb idge MA, Pe seus Publishing, 1991. [HBKC89] F.A. . Hayek, W.W. Ba ley, P.G. Klein, and B. Caldwell. The collec ed wo ks o .a. hayek. Uni e si y o Chicago P ess, 1989. [KC03] Je ey O. Kepha and Da id M. Chess. The ision o au onomic compu ing. Compu e , 36(1):41–50, Janua y 2003. [KR95] J.H. Kagel and A.E. Ro h. The handbook o expe imen al economics. P ince- on Uni e si y P ess, 1995. [Ma 99] Humbe o R. Ma u ana. The o ganiza ion o he li ing: a heo y o he li ing o ganiza ion. In . J. Hum.-Compu . S ud., 51(2):149–168, 1999. [PKE01] Da id C. Pa kes, Jayan Kalagnanam, and Ma a Eso. Achie ing budge - balance wi h ick ey-based paymen schemes in exchanges. In P oceedings o he Se en een h In e na ional Join Con e ence on A i icial In elligence, 2001. [P u81] D.G. P ui . Nego ia ion beha io . O ganiza ional and occupa ional psy- chology. New Yo k: Academic P ess, 1981. [PT02] W. H. P ess and S. A. Teukolsky. Nume ical Recipes in C++ - The A o Scien i ic Compu ing. Camb idge, MA, Camb idge Uni e si y P ess, 2002. [RFH+01] Syl ia Ra nasamy, Paul F ancis, Ma k Handley, Richa d Ka p, and Sco Schenke . A scalable con en -add essable ne wo k. Technical epo , Be ke- ley P ess, 2001. BIBLIOGRAPHY 50 [Ros94] G. Rosenschein, J. S.; Zlo kin. Rules o encoun e - designing con en ions o au oma ed nego ia ion among compu e s. MIT P ess, Camb idge, 1994. [Smi62] V.L. Smi h. An expe imen al s udy o compe i i e ma ke beha io . Jou nal o Poli ical Economy, 70:111–137, 1962. [SNV+05a] Bj¨o n Schnizle , Di k Neumann, Daniel Vei , Mau o Napole ano, Michele Ca alano, Mau o Gallega i, Michael Reinicke, We ne S ei be ge , and To s en Eymann. : En i onmen al analysis o applica ion laye ne wo ks. Ca ne s deli e able, Uni e si y o Ka ls uhe, Uni e si ´a delle Ma che An- cona, Uni e si y o Bay eu h, 2005. [SNV+05b] Bj¨o n Schnizle , Di k Neumann, Daniel Vei , Michael Reinicke, We ne S ei be ge , To s en Eymann, Felix F ei ag, Isaac Chao, and Pablo Chacin. Deli e able 1.1; wp 1: Theo e ical and compu a ional basis. Technical e- po , CATNETS, 2005. [SNVW06] Bj¨o n Schnizle , Di k Neumann, Daniel Vei , and Ch is o Weinha d . T ad- ing G id Se ices – A Mul i-a ibu e Combina o ial App oach. Eu opean Jou nal o Ope a ional Resea ch, o hcoming, 2006. [Tes97] L. Tes a sion. How economis s can ge ali e. In The Economy as a E ol ing Complex Sys em II, pages 533–564. A hu , W.B. and Du lau , S. and Lane, D.A. (H sg.), 1997. [Va 94] Hal R. Va ian. Mik okonomie. Oldenbou g, 1994. [Wei99] Ge ha d Weiss, edi o . Mul iagen sys ems: a mode n app oach o dis- ibu ed a i icial in elligence. MIT P ess, Camb idge, MA, USA, 1999. [Wie98] N. Wiene . The his o y and p ehis o y o cybe ne ics. Kybe ne es, 27:29–37, 1998. ISSN In his documen he de elopmen s in de ining he compu a ional and heo e ical amewo k o economical esou ce alloca ion a e desc ibed. Acco dingly he o mal speci ica ion o he ma ke mechanisms, bidding s a egies o he in ol ed agen s and he in eg a ion o he ma ke mechanisms in o he simula o we e e ined. 1864-9300