A dynamic approach to Consistency Management for Mobile Multiplayer Games ‘Abdul Malik Khan and Insttut TELEG TELECOM & 3 SudParis ‘CNRS SAMOVAR ‘9 1ue Charles Fourier 91011 Evy cedex, France (Abdul_malik khan )@it-sudparis.eu, Sophie Chabridon ABSTRACT. [in reltiplayer pm: where many, players take pt game while communicating through @ network, the players ‘ney lowe an faconatiend iow ofthe girae world became of the comumunieation delays across thie network. Consistency smninionance algorithins mast be aed ta have a waif ‘ee of the gue etl, “The canbority of Cheer wlepeithnes ae used for wired network and they use a fied approach splseh la atlly slapd ba Ube Caegingt cto ane gpa ecvironmest. In these algeritimss. an approach called local Ings seed which delays the presentation of updates xo as to ‘Set far meters Intopcin. “Fos rakes of local lag ally fixed fora game wasion. In wireless networks where the saetwork latencies can varies greatly eral local las appease an cause inconsistencies inthe game state between diferent replicas Ale deal-reckoeing is wed aloog with lcal-ag to liek obec latency, eed Ingreoye wapeeeironeen Rep: ing a fixed dend.reetoning approdch im the face of earn, ‘iettork ed game tavicormnesh eau peodece enpeodictalle Eur behaviour which can ease hata layers to it the fame. In this paper. we prevent a dynamic and adaptable iepprocth for keca-lng cad deacheockuning in whith te a forithm's parameters are changed according to the chang gine csrpecictalie actwurk od are srdiommecit: WOE bbeliew that this adaptable approach can greatly improve conwittency and hence the playability ef the game Categories and Subject Descriptors SDA.66 [Distributed Systems|: Simulation ond Mad. eli. General Terms isi, Algorithm Permisson to make ign or han copies fal oc pan af sis work for Penola lasso ns raed wis a pod that cops me et malece dace fr post or comer antag Sl cope Tolan ede lla atc etpage Tocapy bere epublhipostos ses or a nn as, ues Sp Pesmsiea antl ae COU NOTERE 208 ne 28-27, 208 Ly France Coppi 2s ACM 978. 19595.002. Lom $5.00 ie Chabridon Antoine Beugnard TELECOM Bret re Computer Senco Beparimert seats F-29028 Brest cedex 9, France antoine, beugnard@enst-bretagne ir Keywords Multiplayer Mobile Games, Latency Hiding. Data Synchro nization, Consistency Algorthan INTRODUCTION Consistency maintenance i a comple iarue in networked james becouse eves tore ifleul to adtre when Bla ing on a mobile phune because of the highs latency of nobile ‘ecbaac [f] discumeee the effet of high Inteney on players and ob sarves that a high commumieation delay eam cans the player get the patie, A player amemally Yokes tcions 20m reacizon time from tne systems [10]. Ta hide latency Sand maintain comantency in multiplayer games algorithene such on Dead-Reckmng [1] snd Titi Sate Syeetiromieae tions [5] may he tee. The algortims wes a fied ap povech kn Ube face of miging sectors nd. garne cerioer tment, In the wircles networks whee networks delays are ‘eetesaily wnprecctable, Ue Bead pyre cin be Sout Prone im teres of consistencies aenong diferent gaawe repli- ext. In this paper. we preseat a dynamic ane spdatable ‘approach to consistency toniniesinet and delay ding far multiplayer games un mobile popes which clang it pas rameters not only with the changing network environment tet: nlc! eocating to dices objets wal iagions SE the ame. We, fist. disrus different syichronization agceitions ited their folevenes Gor high Intewry hile xtwerte. Then, in the next section, we propose an adaptable approach te smalniain comintency in mualiplaver gamm. In section 4 vwediscuss how we cam separate these algorithms and insert Use into the Syachremieation Meditm which we previously peoponel Ih ‘One of the frat micthoe see for maining a consistent view of the game among renote tem is Dead Redkoating [2}, Dend-reckoning is used ta rence baadwidth coma tion ad hide network latencies. by seeding update msgs Jose Frequently anal atiunting te ste inormntion etree te uplates. wing the already received information such a powition, velocity and/or acceleration af the object. The predicted value can be dflrent fom the actual alte which received throngh the next update miemae. a thin ce noe comecgence eth con bead Go arstvcat the act value, The importance of demd-recioning in mobike gamesfs that it permite a mobile tersinal wot to be blocked and to cintinne even Hb ont cetring the data fn ere ofa dfisconnection, for iastance. The convergeace tien rst fine Ue tial value and the profited. valet fm srmocth vray, 20 that there is no abrupt effect au the game wser_ [13] proposes a deacbreckoning protocol bined on Use position Inistory of the olgjoct being dead-reckoned. It makes seme to mse the airaple dear reckoning algorithon wher: the path panos iad iteaight ouch no ts che toring puines, and to ue the position-based history protocol when the path is periodically chansia “Time Warp Syecivonization (TWS) (if an ophiminike synchronization meclanism for parallel/distributed sisnal Use. TTWS tals 2 enapebot of the stale nt the reception ff command aad int a rollback to an carlicr state if ih cl cartes Hoa Ua lat eee deme ceived. Ona zollback, the state is fist restore to the previ tus snapshot anf then all the commands tnt geeureel be (ress the seipeln end Ube exocetins time ure meted ‘Two problems ari with this method when used for mobile games. Pint, i necda to store previous states, lor which memory is required. Second, rollback requires proces power which ean be limited in ease of mobile devices. ‘Tratling State Syackraaszation (TSS) [5] lo executes rll tuck when inconsistency is detected. However. i iniple: ments rollhack so a to aveid high memory and processor tverheads demanded by Time Warp Syackronization. lar Stead af keeping stapalots of avery command, TSS keeps treo copies of the anton gaace work etch at a diferent: ‘al simulation times weparated by some synchronization de ly. Whew am icomistency is detected im the state lead ing in time and rollback is required instsad of copying the State fom a snapubot as TWS dows, TSS just simply copies the game state from the tiling state to the lending sine, and then performs all commands between the inconsistency point and the present point again. TSS still does not ac ually solve the rolliack problem originated in TWS but i sell ve better performance than TWS. ‘Porenptive Consistency (PC) [9 provides an tering of sapdlates and avoids potential coats. It introduce the coy cept of legality nod sineltaredty, Tae property of leyality requires that the Latency for a given media instance between feo remote proces mast be kept constant Tike smal: teelly property sats that the physical time etme the payouts of two updates is te samme for all users In the foes of aces racing guene, the airmen property & held if in the cae of a collision between two cars. the two cars sre considered at the same place af a given time for all the were, For a system to be perceptive omistent, it mst Stig) both properties of semntenerty and legality. ‘Tle al: sorithm implementing PC! han three pws In the fmt pile, the algoritl calculates for a given player, the rank: oti communication delay Between thie local player and ‘iD ts syamcte plagets. tp he coed phe, the algerition raleulates the local lag to be fntreduced loeally to order the spats befoes the playent of n retin intnnce, In the thind phase, the message is played out. Ia case of mobile games, fs network delays can be quite longer, the local lag to duces by PC can have bd effects on the players. Hence, deadreckoning can be combined with PC to hide the effect ff cal lag by ding predicted Enterta baton Jin [F] the authors present an. approach called deltarcausality fn while mags atrving late are discarded! and only Haowe saesiages are played oul which renelt from semate sed to Toca nec inn pease Cires Rit‘ dicaoing iy ome because te Ine arriving mesages may violate the causality peperty ofthe lnccmatng operations 2, AMORE DYNAMIC APPROACH Ii pecvions auction ws dicnaeed ifleat nyachecttialtes algorithms sed for consistency msinteaance in multiplayer fnames, Hecate of the rll back ad re procestng of com senea, TSG amd Tine Wasp are wry covtly in totem of memory and procesor usage ancl hence not very suitable for mobile games. ‘The Ioeal lag approacla, combined with deadreckoning is suitable for high latency networks. But because of changing delays aad jitlers in wirelase net works, Sa cal lg ean eens incimninicid fm the ues state arose different nodes, ale, we believe that mage dic ‘carding appraach proseated in [7] can be intereting to com bine with locallag. This is because in wireless networks, there are mewages whick arsive Inte dae to the meter jit te are bas et be tourna Unc eke ering mee ‘cages can cane inconsistencies. In this section me present ‘scm wpecpnc iy whieh Ube ceteteaey tmatndenee algoritiun caanges its parameters according to different fac= tor of the eaviranment: eg. network land. type of object et the virtual easement, location ofan object im the virtual ‘world ete” We Brat isc lie conditions tnder ics these dliflerent parameters are dynamically changed and then wr ‘combine Ubese diferent approaches in the form of an algo fn. 2.1 Observations Whale playing a euleiplayer game, the iconsistencion 2 ‘cur due to communication delays across the network. The ane pices catia Chess delays ta code ba en peminte for the Inte acriring. meninges been) remite wieo Becarwe of the jitters specially in the wire networks, thee lags inn vary greatly: ‘Thescive be bv eccniry to folmerve the network load during the game play and com> penile for Unie dears accordingly Ia a rich multiplayer sgunes, there: are many types of objects in the virtual euvi- Fonment. The velocitivs as ditections of tise olsjocts wy coop I Whats mace. Bor Gnanele, '6 beanie the spore of the tenis ball greater than the speed of the hires. Alo a player ks ts toot dha Us Ut sateen (fm ball. Therefore, theve different type of objects im the game world as diferent types of consistency requirement. ‘Phe algarithons scopenetile or tsiateacy meingeeed es to react not only to the varying latencirs of the underlying, actwork ldvantzicture, Wat ales ln te deal with elfen types of objects of the gue diferently, We also observe thu the consistency requirement of ax objeet mot unl de pets wpe ta sped, bet aloo upon et postie tn the wi tual word. For example, in a car racing game. we nee! strict comsintency management when to car are very to cach other and they both are very near to the finishing, line. Based upon these observations, we belie that a eox- seterey eclntemnt geritheon tc ele ne scent the ‘context ofthe same along with network latencies. 22 Anadaptable local lag When « mosageabont an object i reesived from a remote saan; this ject as gertain diathermy omy Bee the ate objects, destination, ony other partant eetityfn the game word, called pivot. Weeehange the valle of the local lng scoring te thes factors, 4. Lf the ahjert mtose mange we have received Grea the remote user is roming doses towards the peat we ms doce the walae of locablng, [fee chfoct i gol awry from the pivot, we increase the value of loeal lag upto (: cortala Bel called! Focmbdeg.- Tha focremes ean bbe continous with respect to the motion of the ob {eet or ean be dierote fumed on sons a i [12]. The Tate of the change of tse wale of boral tng ia arth Cation dependent, and the progeaminer mist specify 18 during the development of the game. Is the next section, we present aa approach on how a programmer can speedy these values to a eomponent responsible ie Goenisncy ising 2 The value local lag also changes according to the load fon the network. When the nimber of meages ariv- fg Bates Alien cortala wait tine ineseapen 0 cesta ruber Ng, we increase the value of local lag. Tis Increner in the number of menage arriving late ean be due to jitters in the network. Note that the value cf local ng is proportional to actrork latency, but we caemgt ioceeeee amore than a cote it borane ‘wil Lave a bad effect om the rexponsicenina. Tie ‘upper limit cam be dependent on the pace of the game ‘and if te game or am object in the game deauands igh ‘responsivene, we mt! not increase Use lal lag alae (ove m exci Heit ‘This lie can be pected bry te ‘pune developer. SL Wee sel diffrent lnc Ing walacs Far diferent types cf ebjects noc to thet portame fa Une Ene For example we can have amaller value of lacal lag for ‘he ball ned igh value for the players in a. tennie game because the responsiveness of the ball mist be {il to satisty the expectation requirement in (10) ‘Agni eel en fnberfoce provides ey the comepoment ‘implementing the algorithin so that game developer can specify the relative wales af Yoeal lage Sor diferent dbjects in the virtual envionment 2.3 Message Disearding ‘As proposed in 6), messages arcing later han a certain tte limit, Beate of the network jitters ci ene incon sibtencio in the ganie and bewce ts betes to discard tbe Is [they peop to wary the wie of dellxcninalty Le tine threo for discard according to the network lac 20 ata keep the mumber af discarded tssages under 3 cert ath init. Becatne ia a virtual eovirsement all the ebjects do not behave ia the sate way. we belicye that diferent chars Bf objects can be assigned iflorent value of detaentsalty the value of deluweansality must not be wery large 50 that (he peti of the Ball beste tment dpe a the oe diction than the real value. We believe that combining the focal lag approach with mesage discarding can yield better rill 24 Adaptable DeadReckoning ‘The vale of dend-recholng sould be tual aide ‘sunmeat obpesepenent 1 u Figure 1 Area argund the base line constitute a critical region in Tennis game 24.1. Critical Regions We define Critical Region as a region in the game, where wwesneel strict consistency so that all the players tn the game bare coininat view of the game is that tagien Aertend region ic the one in which inconsistencies can violate the fairnens of the game or cin annoy the player because his ‘expectations are violated. Therefore we progsice to tee real nupdate messages instead of prediction in thee regions. For ‘eznmple im tbe coe. o'w tennin puateh, of te’ bl it by fone player is touehing the gmound near the base line on the ‘ther aie ofthe court, aad the ppement player lx quit tar from the ball, we stup the dead-rerkoning 20 as to ducrease Use fairness between players. The hall wil tauck the groted ‘ely whee the erigeal mecmge bs roocived eel hee the result of the score will be corset The alcalation of the “deciiel whetheer to nacidhadbeechocing tr mot tan areal be dome through soane exsy arithmetics. Far example, in Figure 1 let ey and be the current speed of the ball mae tse masimum spend of the opponent player respectively. Let (be tha ist ance covered by the ball for the centre of the ‘hart tothe polet where te tomclas the gronnd nd he thie distance of the player from that point. The ball will reach the gronnd in l/, Uanes uit and daring that time (he player an hes ictnnes of ep later anal, We > 1, l/c, then the player eanmot reach the ball. ancl sre stop dearlecchimieng Bein sea the a brew Hil he ‘calculation. OF come, the ball will stop and jump it the lle for a short period OF tine ene the entre of the court but it wil aot affect the playability and faimess of the game becanse we know that the opponent wouk! not reach the ball many cave Note that if wee do sot tne the idea of eitheal region atid contimur with deachreckoming we may peeict ‘eros porition of the ball Loach the ground near the base line and hence that wreng decision may cause one player paint which he otherwise would have won. ‘Apeet even (or what? Hes bests) critical copie deal- reclaming.s alo dependent upon the network Joa and the ature of the object. Por example we can have aliferent dead: reckoning thredild for eilferent objects according to Uae movement and importance in the game world. This Mibestald arent be ageclted by the gyune developer to the ‘component responsible for comsistency maintenance.25° The Algorithm In this setion, we combine the iden disused above in the form of a alain. 1. Calculate the focal lag at the beginning of the game for ears claw: of objects according to network Lateacy sand responsiveness requirement of the objects). 2 Change the loraf lag ifthe network load has changed cr the location of an abject is changed ‘2. Buter te mcseige nccoiting to their (aol fag Yabo before plaving oat de Did he ig tiny ace arg Mes ‘3. Caleulate the number of mesoages discard 6. If the sumber of meseages discarded exon = cota Unreal for that claw of abject © Ther readjust the value of local lay apply DR (Change the threshold value for DR if the object is entered the ertial regs the metcore load has camped 3. SYNCHRONIZATION MEDIUM Ti this section we diarus how we ean weparate these con Skpzoey tmnlctenance algeetiline fom the game loge mn how we can peovide the game programmer the interfaces to specify different parameters ofthe slgoritn|=) We previouly prevented the notion of Syachronization Masini (0) Based Go Ue cient of wade ciigelly pecstuied in A.A syechetesietics weliens © a commence ation cemponeat which handles cominteney uanagement » satel commie aapeci: t altows to,se foes the gume cleats, the Iatency compensation and synchro sieation mechanioms This, a synchramization medium fn abstraction wf the communication for data consistency evens aetosk. "That hin the bdbeitngs ta ocr ie doe reopen syecirobination trol Unt can be, picked Gp. aae) tae directly (Onc-f the main heer he apace ln died is to lewen the burden on the game developer. With the synchronization medinen at hand, the game developer cast ‘acetate tip the gave lope only and i roland of the ‘consistency and synchronization concerns. Nevertheless, ax ‘ve, titioned fn the sects 4, the value of Nocabogg be ‘hflerent for diferent objects. Also the use of a certain dea- revlon terlnique depends report the olpjects i the virtual environment and the lovation of that object ic. critical region. So both loeal lag and deud-reckoning technique are ‘gate depescent. To this end, the syachonieation adem fmnst provide coustracts to the game developer to specify these value for Wifeent objects and regions at the time of tha gpme ceretopeend "Tee in is shee tn Figure 2. ‘A synchronization curmpouent is logieally a single com ponent which be physically didibalos senses the ietwork Seployed on different devices. Encl distrited part of the irecleembestion' mations ta callid Plager Mfaraiger so sides on a player's node, and Server Manager if it resides fon a server node. ‘Two components with the Player ata ‘Administrator soles interset with the taediam, The Plager role comesponds to the game client reading on the mobile ternal aad tines the Iplayer MedeumSersices services of. fered by the medium. The [Player MediummServices inter~ face offers services like provision of information regarding a cw lager i atin ot tno the ygene semen eel he reception of update messages from remote players These services are implemented through fanctions sels ae geiNew Player and updatePlayer. "The adrainestrator role is the game server whicl mies Une [AdgainsMedium Services aor See coed the malic "The iy reba edeninteatine services such ae momitoring the game bo present Getting and lo charge fos dom te eee Tee eee faay need some services to interact with the player These services are ollered by the IplayerCompanentSerrices iss the game data bet city te a ealect of the cimen Apest fom bcie canemunication services, the medina alo offers Sjecheteaton ecrvioes it woeiatichey mtabenante sachs i deetgetening and becnb bg breed appriechen ‘Thee Service are internal services abel are bide from (he ame ica hich sil lee bey Une mine transparently to the player. Tn pamimeterta: the algoitins sed bm he ‘iesRien ta ten offic extern Etats Ks Unt ne developer to fix the value ofeifereat parameters for difer- lent objects and alsp to specify different regions of the game. ‘These parameters ave game dependeat ane the gpm dowel oper uns specify them according to the requirenpent of the fine, Tho medic cles ioe ervices thivogl Us extarenl Imterfuce 1SetParamctersServices, 4. DISCUSSION "The reason for proposing dyaanic and adaptable ap- proach for michile multiplayer game is that the latency of ‘wireless networks is highly unpredictable and hence we cane fot apply a fard appronc. Apart from this, in Fert Person Shooter, guy, there ere objects i the- gyms wold wheat [ater is fast ax compa to other lyjects tm the same gare, By having different values of threbold for Boca ly med dene reckoning for diferent objects, we can seduce the muaber of ‘update mesiges for slow moving objects and ereate a space (the deren fo the tember of tectinges. for fart fate ing ob cts ta increase comistency and responsiveness. Also Us cheedehiaty teqatrcment varies ncccrding othe pion ff the game objets in the virtual world. ‘There are areas fm the game where we eat relax the comistency without fnaving effect on respousivencss and fares, and there are ther ertical regions where stsit consistency nec to be fnintaiaed, We believe tliat by eatifying these restore during the game developeeent and specifying therm to the synchronization suedians will greatly simplify the peablem cdematency canines 5. RELATED WORK We have alzrarydiscused the Joeal lag approach [2 in sectina 2. We Earther cabarets taco by dynamically shongiag the valtc of local lag during the execution of the Pia dee LL alent ‘he movement of the object while entering the ertieal rion. ‘They put an upper limit on the velacity of ebjects in the fame due to which it docs not look bike a real world one. Tnatese im our approarlh, we use the evtieal region just 13 calculate wHicther we teed Deaelrschetag or not 12) discusses about the distance betwen objects and an dbjects distance from a pivot, whieh can be an object, dos ination of any other important entity in the game woz, ‘They propose ta set different dearer of comistency foe i fers Chjects according to thelr ditancs fom the pet Hlomever they do nat deal with changing the value of pe rameters during the exccation of the ge 6. CONCLUSIONS AND PERSPECTIVES I thi paper we argu or these a of mone rman ap poeech bs cistency eonintena Sey tasktptarer gamed oe ‘mobile phones. We hope tut this approaci wil greatly belp sebntain the Neve af cniniteny Hetero sect plage a vrellas improve the playability of the game. As future work, ‘we intend to implement this approach ting syachronis lon miei and sing a car racing gate as a prototype. We intend to evaluate this appronrh and compare it with ‘her algerie. The game will be implied om the top of the GASP [11] framework, a middleware dedicated to the development of multiplayer gam om mobile plone A. ‘seropaiative mealies of thes abgertthen brah in peop tive: In future we woulel alsa ll to quantity the welalles ‘af epochal ie err of seca Uh llr ‘on game decelopment and its re-usallity. 7. ACKNOWLEDGMENTS ‘The work isa part of the project JEMTU which aims to design slutions for the technological and poychowocialogial ‘ues that ow down the developaneat of multiplayer games ‘om enobile plone (http /proget terry fe/projects/JEAITU/)- 8. REFERENCES [Application protocols. Ln IEEE Standard for ‘Dittrited interactive Simulation. IEEE Std ata Lams, 1905, [2 N. Boulet: Bast event ordering and perceptive ‘coat, bn ne meltve desetbeed one ipaycs game In 7th International Conference on Computer ‘Games (CCAMES'2005). pagew 186-152. 2005- [) E. Carioa, A. Beugnard, and JoM. Jéaéguel. Am archictecture ane! a procs for implementing ‘Sistritytel collaborations. la The 6th [EEE ternational Enterprise Distributed Object Computing Conference (EDOC 2002). Ecole Polyteciique Pélérale de Lausanne (EPEL), Switzerland, September 17 20 2002, [0] Ket. Chea, P- Hang, and CL. Lei, How sensitive arconline gamers to network quality? Cammum. ACM, 49(11):84-98, 2006. (5) E. Cronin, B. Fultrup, AR. Kure, and S. Jamin, Aw lieth apecctelnetins eibchitan for vero me fchitectuires In NetGames "02: Proceeding of the Int workshop on Network and aystem suppur? for game, Pose 67-79, New York. NY, USA, 2002. ACM Prox [7 Y- lebibet, "Teka, ane 'V."Tactlbmna. Adaptive ‘cammality and media synchronization contral far petworlod multimedia applications. In LON, pages ‘25-241. IEEE Computer Society, 2001. [4] AML Kin, §. Chabridoe, ane A. Bengaacd Syuconiztion medium: a consistency maintenance ‘ajeouie S mls iuleyas gett T NetGames Gt: Proceedings of the 6th ACH SIGCOMM workshop on Network and system supportfor games, yen 90-10, New Vor, NY_ USA, 2007 soll 1B) ML Mune, J. Vor, V. Hit, and W. Eelaberg Tecabag iad Tear Boovtling Comat oe eplicated Continous Applications. EEE Transactions on Mulvedan, (0)-7- 87, Fes, 2004 [10] LPantel amt LC. Wo Oo the impact of delay on ‘tabnwe multiplayer games. In NOSSDAV "2 Proceedings of the 12th ternational warkshop an Network and operating spelems support for digital studio ced ide pages 23-2, New Vor, NY, USA, 20 ACM Pr [MY] RL Paern, PDelpinno, B. Grenier Soodan, and M Simatic Gas A middleware or multiple {es in mabe pace networks (st reel). Uidled ‘i: Proedngs of the tn French speaking tomferener om Mabity end wibutycoenpting, Pag 164, New York, NV, USA. 205, ACM Proms [na] N- Santos L. Vega, al Ferreira Vector eld cesitey he no gg Tn Coren ad TCH Campbel, ltrs, Middewar, wulume 4534 of Lectare Noles in Computer Science, page 80-100, Springer, 207 [nS] SK Singhal and D. fe Cheriton. Using position ibeny teed prtecel for datetiied obet visualzation, Techcal Report C3-TR 1505 Skanfbed University Satlord, CA, USA, 100, [1] S. Zou and H. Shea. A comsitency modal fr highly Inlerattior mplayer lon ganna ANGE ‘OP: Proceadings ofthe 40th Annual Simlation Symposinan, pgs 318-328 Washingt, DC, USA, 1300 IEEE Computer Society.