Exploring the Spatial Density of Strategy Models in a Realistic Distributed Interactive Application

Explo ing he Spa ial Densi y o S a egy Models in a Realis ic Dis ibu ed In e ac i e Applica ion Damien Ma shall, Declan Delaney Depa men o Compu e Science, Na ional Uni e si y o I eland, Maynoo h, Co. Kilda e, Republic o I eland. E-mail:{damienm, decland}@cs.may.ie Seamus McLoone, Tomas Wa d Depa men o Elec onic Enginee ing, Na ional Uni e si y o I eland, Maynoo h, Co. Kilda e, Republic o I eland. E-mail: {seamus.mcloone, omas.wa d}@eeng.may.ie Abs ac As Dis ibu ed In e ac i e Applica ions (DIAs) become inc easingly mo e p ominen in he ideo game indus y hey mus scale o accommoda e p og essi ely mo e use s and main ain a globally consis en wo ld iew. Howe e , ne wo k cons ain s, such as bandwid h, limi he amoun o communica ion allowed be ween use s. Se e al me hods o educing ne wo k communica ion packe s, while main aining consis ency, exis . These include dead eckoning and he hyb id s a egy-based modelling app oach. This la e me hod combines a sho - e m model such as dead eckoning wi h a long- e m s a egy model o use beha iou . By employing he s a egy ha mos closely ep esen s use beha iou , a educ ion in he numbe o ne wo k packe s ha mus be ansmi ed o main ain consis ency has been shown. In his pape a no el me hod o cons uc ing mul iple long- e m s a egies using dead eckoning and polygons is desc ibed. Fu he mo e he algo i hms a e implemen ed in an indus y-p o en game engine known as To que. A se ies o expe imen s a e execu ed o in es iga e he e ec s o a ying he spa ial densi y o s a egy models on he numbe o packe s ha need o be ansmi ed o main ain he global consis ency o he DIA. The esul s show ha inc easing he spa ial densi y o s a egy models allows a highe consis ency o be achie ed wi h ewe packe s using he hyb id s a egy-based model han wi h pu e dead eckoning. In some cases, he hyb id s a egy-based model comple ely eplaces dead eckoning as a means o communica ing upda es. 1. In oduc ion Dis ibu ed compu e games a e becoming one o he mo e impo an ace s o an al eady bu geoning games indus y. Such games pe ain o a class o compu e applica ion known as Dis ibu ed In e ac i e Applica ions (DIAs), and can in ol e ens o housands o simul aneous playe s in e ac ing in a simula ed en i onmen . Gi en he scale o DIAs, he olume o da a ansmi ed can be conside able, o example, E e ques [1]. As he complexi y o he DIAs inc eases and he numbe o concu en use s mul iplies, he applica ion and connec ing ne wo k mus be able o scale wi hou losing he eal- ime expe ience o pa icipa ing in a sha ed en i onmen [2]. This abili y o scale is inhe en ly linked o he numbe o ne wo k packe s ha need o be ansmi ed be ween pa icipan s o main ain a consis en global wo ld iew o he DIA. Se e al echniques exis o educe he quan i y o ne wo k a ic, including ele ance il e ing, comp ession and dead eckoning [3, 4]. Dead eckoning uses a sho - e m model o p edic en i y dynamics. Howe e , i does no ake a p io i in o ma ion ega ding en i y beha iou in ela ion o an objec i e o goal in o accoun . In con as he ecen ly p oposed hyb id s a egy-based modelling app oach combines a sho - e m dead eckoning model wi h a long- e m s a egy model [5]. Using his app oach a educ ion in he numbe o packe s ha a e equi ed o main ain global consis ency has been demons a ed compa ed o using dead eckoning alone. These esul s we e based on a es ic ed wo-dimensional es en i onmen ha se ed o p o e he concep bu did no allow he expe imen s o scale o la ge numbe s o s a egy models. The long- e m s a egy models hemsel es we e cons uc ed using isual heu is ics. S a egy models can be o wo ypes: an explo a o y o ansien s a egy ha use s employ when hey a e un amilia wi h he en i onmen and a s eady- s a e s a egy, which co esponds o a p e e ed way o sa is y he use goal in he en i onmen . He e, a no el implemen a ion o he s a egy model echnique in an indus y-p o en game engine called To que is p esen ed [6]. This me hod builds on p e ious wo k by acili a ing he cons uc ion o an a bi a y numbe o s a egy models. Fu he mo e, simula ion esul s a e p esen ed o expe imen s ea u ing a a ying spa ial densi y o s a egy models using his new es en i onmen . The esul s show ha as he numbe o models inc eases, he e is a dec ease in he numbe o packe s equi ed in o de o main ain consis ency in he DIA. An op imum numbe o hese models exis o s eady-s a e ajec o ies, abo e which no u he packe educ ion is ob ained. The ac ual alue i sel will a y om applica ion o applica ion. 2. Implemen ing he Hyb id S a egy-Based Model One o he key di icul ies wi h he hyb id s a egy- based model app oach is he de e mina ion o he s a egy models. To da e, ac ual use da a has been eco ded in a limi ed game-like en i onmen and a ep esen a i e s eady s a e s a egy was chosen based on a isual analysis o ac ual use s eady-s a e ajec o ies[5]. He e, a no el app oach based on dead eckoning is p oposed. When dead eckoning is employed, he e o h eshold is always exceeded a ce ain key a eas. By eco ding he poin s whe e he dead eckoning h eshold is exceeded by his use , a se o poin s is ob ained ha p o ides an a e age piecewise linea model o a use ’s ajec o y – see Figu e 1. The piecewise linea model is hen combined wi h he e o h eshold alues which de ines bands on bo h sides o he model o desc ibe a polygon shape known as a “s a egy polygon”. An en i y loca ed wi hin he s a egy polygon is deemed o be ollowing ha pa icula s a egy. The i s s ep in he c ea ion o he s a egy polygon is he de ini ion o he polygon edges. Each edge consis s o a line pa allel o a segmen o he piecewise linea cu e. The dis ance be ween an edge and he app op ia e segmen o he cu e is he equi ed e o h eshold – see Figu e 2a. Nex , he end poin s o he segmen s need o be joined so as o p o ide a closed polygon. I his is implemen ed inco ec ly, he esul ing polygon could be dis o ed, leading o e oneous assump ions abou he en i y’s beha iou . Again, see Figu e 2a. The p oblem is a oided by conside ing wo con iguous segmen s o he piecewise linea cu e in o de o de ine he s a egy polygon e ices. The pa allel segmen s a e c ea ed as de ailed abo e. Howe e , he e ices o he s a egy polygon a e gi en by he loca ions whe e he newly gene a ed pa allel segmen s in e sec . Examples o such in e sec ions a e de ailed in Figu e 2b. Repea ed applica ion o his echnique o all he segmen s o he piecewise linea cu e esul s in a g oup o e ices ha desc ibes he s a egy polygon. This p ocess can be used in o de o de ine mul iple s a egy models. Ini ially, he equi ed numbe o s a egies is c ea ed. Fo he pu poses o his pape , he newly c ea ed s a egies a e always pa allel o ha o he o iginal piecewise linea cu e. The e o e, he p ocess by which he polygon edges a e c ea ed can be used o de ine each indi idual s a egy. The pa allel dis ance be ween each cu e is he equi ed dis ance be ween s a egies as de ailed in Figu e 3. Nex , each cu e is con e ed in o a ull s a egy polygon, using he equi ed e o h eshold. This h eshold is always se o a alue g ea e han hal he dis ance be ween s a egies in o de o p o ide an o e lap be ween models, and he e o e a oid unnecessa y upda es due o excessi e swi ching be ween models. Wi h mul iple models, i becomes impo an o de e mine which model, i any, he en i y’s cu en posi ion esides in. The ‘poin con ainmen es ’ is used o do his [7]. This es is used ex ensi ely in compu e games in a p ocess known as polygon illing [8]. I wo ks by ex ending a ho izon al line om he cu en en i y posi ion and coun ing he numbe o in e sec ions be ween his line and he edges o he polygon. An odd numbe o in e sec ions indica es ha he posi ion is inside he polygon. O he wise he posi ion is ou side i . The E o Th eshold Dis ance be ween s a egies Figu e 1 A sample piecewise linea model. Dead Reckoning Upda es Use T ajec o y Piecewise linea cu e Y- axis X- axis Figu e 3 Mul iple hyb id s a egy-based models Figu e 2(a) Dis o ion can occu when c ea ing he s a egy polygon (b) The polygon dis o ion is esol ed using he in e sec ion poin s o pa allel segmen s Piecewise Linea Cu e S a egy Polygon Edge In e sec ion Poin E o Th eshold (a) (b) a b c O e lap be ween s a egies Piecewise linea cu es decision o make he swi ch be ween models is aken when an upda e om he cu en ly employed model is equi ed. The s a egy model app oach was implemen ed in To que using he algo i hms desc ibed in his sec ion, along wi h a i s o de dead- eckoning algo i hm. The engine was also modi ied so ha use ajec o ies could be eco ded and s eady s a e s a egies iden i ied. The ollowing sec ion de ails he expe imen s pe o med using hese models. 3. Expe imen al Simula ions The expe imen al es pla o m was implemen ed using he To que game engine and a ull clien -se e a chi ec u e was employed. Each simula ion was ca ied ou in a game en i onmen ha consis ed o a unique ou e om a s a o a a ge posi ion. These posi ions emained in a ian h oughou . The use ’ s goal was o na iga e om he s a o he a ge posi ion in he sho es ime possible [6]. The applica ion simula ed he numbe o packe s ansmi ed in o de o main ain global consis ency wi hin a ce ain h eshold e o ole ance. The e o h esholds and sepa a ion dis ance be ween he s a egies is de ined in To que game uni s. Each game uni co esponds o a me e in dis ance wi hin he i ual en i onmen . The i s se o expe imen s measu ed he numbe o packe s ansmi ed as he numbe o s a egies was inc eased and a cons an e o h eshold was main ained. The second g oup o expe imen s ocused on inc easing he densi y o s a egies wi hin he same a ea o he en i onmen . This was achie ed by beginning wi h a single s a egy wi h a la ge e o h eshold. This de ined he a ea, and encompassed all eco ded use ajec o ies. The e o h eshold was hen dec eased and he numbe o s a egies inc eased so ha he s a egy models always co e ed he same a ea. A o al o six subjec s pa icipa ed in he expe imen s. All use ajec o ies we e eco ded. The applica ion hen employed his da a o simula e he gene a ion o upda e packe s unde a ious expe imen al condi ions. The simula ed en i y na iga ed he es le el using he posi ional da a eco ded om each pa icipan . When he a ge posi ion was eached, he ollowing da a was eco ded o disk: ype o upda e, he posi ion whe e he upda e occu ed, he o al numbe o upda es ansmi ed o ha a emp , he ac ual en i y posi ion, he modelled posi ion and he s a egy polygon e ices. 4. Resul s He e, esul s a e p esen ed o he wo se s o expe imen s ou lined in he p e ious sec ion. In all expe imen s, bo h he Hyb id S a egy-Based Model (H), and Dead Reckoning (D), a e conside ed. In e e y case, U n deno es a es use . 4.1 Cons an E o Th eshold As discussed in sec ion 3, he i s se o expe imen s concen a ed on he compa ison o pu e dead eckoning wi h he hyb id s a egy-based model app oach as he numbe o s a egies inc eased and he e o h eshold was main ained cons an . Table 1 de ails esul s om wo es use s who a e ep esen a i e o all use s. In his case an e o h eshold o 6 game uni s o bo h dead eckoning and he s a egy model, and a dis ance o 9 game uni s be ween s a egies, is employed. F om his able we can see ha as he use con e ges o a s eady s a e ajec o y, he e is a signi ican educ ion in packe s when compa ed o pu e dead eckoning. We can also no e a educ ion in packe s sen as he numbe o a ailable models inc eases. When a use i s en e s he en i onmen , as exempli ied in ial a emp s 1 and 2 o bo h use s, hey a e ollowing an explo a o y o ansien ajec o y. In hese cases, he use is explo ing he en i onmen , and dead eckoning is used ex ensi ely. Wi h he ansien ajec o ies, as he numbe o models inc eases, he numbe o s a egy packe s also inc eases. This occu s because he e a e mo e models a ailable and he e is inc eased likelihood ha he en i y will c oss back and o h ac oss hem in ge ing o know he en i onmen and accomplish he goal. Once he use has become mo e amilia wi h he es le el, as seen in ial 3, ajec o ies close o ha o he s eady s a e s a egy a e adop ed. This leads o a no able educ ion in upda es, and o e all be e pe o mance om he hyb id model in compa ison o dead eckoning. When a s eady s a e ajec o y is adop ed, he addi ion o models abo e a ce ain quan i y has li le o no e ec on packe upda e a es. In hese cases, he op imum numbe o models o ha ajec o y has been ound. This can be seen in ials 4 and 5 in bo h ables, whe e minimal packe educ ion is no ed a e he use o 3 s a egy models. In he inal a emp s o bo h use s dead eckoning is made comple ely edundan . A se ies o expe imen s we e also conduc ed ea u ing an e o h eshold o 9 game uni s, and a dis ance o 12 T ial DR 1 Model 3 Models 5 Models U1 U2 U1 U2 U1 U2 U1 U2 D D H H H H H H 1 20 30 16 28 16 29 16 31 2 17 30 11 28 12 29 12 31 3 12 19 9 17 5 16 5 15 4 12 11 6 8 5 9 5 10 5 12 12 5 1 5 1 5 1 Table 1 E o Th eshold 6 S a egy Wid h 9 game uni s be ween s a egies. Resul s simila o ha o able 1 we e no ed. Due o space es ic ions, howe e , hey a e no p esen ed he e. 4.2 Va iable E o Th eshold The nex se o esul s ocuses on an inc easing spa ial densi y o s a egy models – see Table 2. In his case, we ini ially s a wi h a single la ge model wi h an e o h eshold o 36, and p og essi ely inc ease he numbe o models ound wi hin he a ea occupied by ha model. Table 2 ein o ces he ac ha he hyb id s a egy- based model pe o ms be e han pu e dead eckoning. As would be expec ed, he numbe o pu e dead eckoning packe s inc eases as he e o h eshold dec eases. A leas one packe will be equi ed ega dless o he model used; ials 4 and 5 indica e his o he hyb id model. The packe ansmi ed in hese ials e e s o he long- e m s a egy model. In ial 2 o Table 2, we can see ha , as he h eshold dec eases, an inc ease in he spa ial densi y o models occu s, and swi ching be ween s a egies becomes mo e likely. This accoun s o he inc easing numbe o packe s as he densi y o models inc eases. The esul s indica e ha when a single model is used i co e s all s eady-s a e ajec o ies as shown by he single packe ansmi ed in ials 3 o 5 in Table 2. Howe e , he h eshold is e y high and consequen ly consis ency is low. By inc easing he numbe o models wi hin he same space, he alue o he h eshold dec eases wi hou incu ing a signi ican inc ease in packe s sen . T ials 4 and 5 clea ly illus a e his. A igh e h eshold p o ides g ea e consis ency. This is a clea illus a ion o he consis ency- h oughpu adeo [4]. 6. Conclusions and Fu u e Wo k In his pape we ha e desc ibed a no el me hod o cons uc ing mul iple use s a egies using dead eckoning and polygons in a ealis ic DIA known as To que. This was ca ied ou in o de o p o ide a eal wo ld implemen a ion o he hyb id s a egy-based model, and o es he e ec s o a ying spa ial densi y on he amoun o packe s needed o main ain consis ency o a DIA. Using his app oach, we ha e shown ha he hyb id s a egy-based model pe o ms be e han pu e dead eckoning. As he numbe o models inc eases, he e is a educ ion in he numbe o packe s needed o s eady s a e ajec o ies. In hese cases, an inc ease in he numbe o models abo e a ce ain quan i y has a negligible e ec on packe educ ion. The poin a which his occu s is he op imum numbe o models. The ac ual op imum alue will a y wi h he applica ion and he ou e ep esen ed by he s eady s a e s a egy. An inc ease in he densi y o s a egies in one a ea esul s in he minimisa ion o dead eckoning packe s. When he ajec o ies ha e con e ged o a s eady s a e, he hyb id s a egy-based model can p o ide inc eased consis ency using ewe packe s han dead eckoning alone. Fu u e wo k will in ol e he examina ion o he psycho-pe cep ual e ec s o la ency on use s o Dis ibu ed In e ac i e Applica ions o es ablish c i e ia o choosing e o h eshold alues. REFERENCES [1] Humble, R., Inside EVERQUEST, in Game De elope . May 2004. p. 18-23. [2] Bondi, A.B. (2000). "Cha ac e is ics o Scalabili y and Thei Impac on Pe o mance".In Second in e na ional wo kshop on So wa e and Pe o mance, (O awa, On a io, Canada, 2000), ACM P ess, New Yo k, NY, USA, 195 - 203. [3] Bassiouni, M.A., M.H. Chiu, M. Lope , M. Ga nsey, and J. Williams, Pe o mance and eliabili y analysis o ele ance il e ing o scalable dis ibu ed in e ac i e simula ion. ACM T ansac ions on modeling and compu e simula ion, 1997. 7(3): p. 293-331. [4] Singhal, S.K. and M. Zyda, Ne wo ked Vi ual En i onmen s, ed. S. Spence . 1999, New Yo k: ACM P ess, New Yo k, New Yo k. [5] Delaney, D., T. Wa d, and S. Mc Loone (2003). "Reducing Upda e Packe s in Dis ibu ed In e ac i e Applica ions using a Hyb id Model".In 16 h In e na ional Con e ence on Pa allel and Dis ibu ed Compu ing Sys ems, Augus 13-15, (Reno, USA, Augus 13-15), 417-422. [6] Ma shall, D., A. McCoy, D. Delaney, S. McLoone, and T. Wa d (2004). "A Realis ic Dis ibu ed In e ac i e Applica ion Tes bed o S a ic and Dynamic En i y S a e Da a Acquisi ion".In I ish Sys ems and Signals Con e ence, (Bel as , I eland, 30 June - 2 July), IEE, 83-88. [7] Mo enson, M.E., An In oduc ion o he Ma hema ics and Geome y o Compu e G aphics. 1989: Heinemann Newnes, Jo dan Hill, Ox o d. [8] Jenq, J.J. (1999). "Pa allel Polygon Scan Con e sion on Hype cube Mul ip ocesso s".In ACM symposium on Applied compu ing, (San An onio, Texas, Uni ed S a es, Feb ua y 28 - Ma ch 02), 110 - 114. T ial 1 Model 3 Models 5 Models D H D H D H 1 10 6 20 10 24 16 2 8 8 16 14 18 18 3 5 1 11 5 15 7 4 5 1 8 3 11 5 5 4 1 6 1 9 1 Table 2 Va iable E o Th eshold Resul s o 1 Use