Sensitivity Of Quake3 Players Sensitivity Of Quake3 Players - PowerPoint PPT Presentation

Sensitivity Of Quake3 Players Sensitivity Of Quake3 Players Sensitivity Of Quake3 Players Sensitivity Of Quake3 Players To Netw ork Latency and Jitter To Netw ork Latency and Jitter To Netw ork Latency and Jitter To Netw ork Latency and Jitter An incomplete, experimental look at the impact of network conditions on a player's choice of server for multiplayer, networked games (Oh, and something fun to do as well....) Grenville Armitage gj_armitage@yahoo.com Nov. 1st, 2001 gj_armitage@yahoo.com 11/1/01 Page 1 SIGCOMM Internet Measurement Workshop (poster) From the incomplete-but-fun-research-department

Introduction ● Qualitative assertion: Low latency and jitter are desirable for real-time, interactive games ● Quantitative assessments: Rare, yet useful to ISPs and game hosting companies – What is the latency radius within which I'll find my primary population of players? ● This project attempts to correlate observed player activity with network conditions – Specific context: Quake III Arena , a networked, multiplayer ' first person shooter ' (FPS) game ● Hope others will embark on similar research – This project is self-funded, donated resources gj_armitage@yahoo.com 11/1/01 Page 2 SIGCOMM Internet Measurement Workshop (poster) From the incomplete-but-fun-research-department

Test Environment Californian London Hypothesis: server server Palo Alto University ● Players will prefer lower 'ping' times College 147ms London to servers ● Server usage patterns will reflect topological locality of players Methodology: ● Establish two QuakeIII servers that Players from everywhere appear identical to client-side on Internet Reality: selection process ● Log players, their IP addresses, and ● Californian server: 600MHz Celeron, in-game 'ping' samples over period 128MB, FreeBSD4.2, T1 link to PAIX (hosted in Palo Alto) of months ● London server: 900MHz Athlon, ● Assess topological locality of 128MB, Linux kernel 2.4.2, 10Mb link to players, and distribution of observed UK net (hosted at University College London) ping values. ● Both servers advertised their location as "Palo Alto, California" gj_armitage@yahoo.com 11/1/01 Page 3 SIGCOMM Internet Measurement Workshop (poster) From the incomplete-but-fun-research-department

Quick Stats.... Duration of Trials: Common server details: ● Californian server: ● Quake III version 1.17 (linux binary) May 17 to Aug 18, 2001 ● Same 6 maps, fixed cycle sequence 5290 unique clients ● 20 minutes per map 338 clients played >= 2hrs each ● Up to 6 remote players 164 'days' aggregate played time ● 2 permanent 'bots' to attract players ● London server: ● Identical registration with master May 29 to Sep 12, 2001 server (clients see latency as only difference) 4232 unique clients ● Server-side 'ping' sampled everytime 131 clients played >= 2hrs each 77 'days' aggregate played time player runs over an object, dies, or kills another player Donated resources: ● Tristan Henderson supported server at UCL ● Brian Reid supported server in Palo Alto Total Played Time on Californian Server Total Played Time on London Server 80 180 160 70 140 60 120 Days Played Days Played 50 100 40 80 30 60 20 40 10 20 0 0 140 150 160 170 180 190 200 210 220 230 240 250 260 125 150 175 200 225 250 Day Of Year Day Of Year gj_armitage@yahoo.com 11/1/01 Page 4 SIGCOMM Internet Measurement Workshop (poster) From the incomplete-but-fun-research-department

Popular Latencies Cumulative Median Ping Median 'ping' per game: 100 ● Each player's 'ping' sampled > 10 times 90 per game 80 ● Median values per player per game 70 California 10 ● Cumulative plot reflects most frequently 60 Total % London 10 appearing median ping values 50 California 1 London 1 40 ● California and London curves similar 30 20 Players who picked up at least 1 item per minute (minimal activity) 10 0 California 1: 80% < ~196ms 0 50 100 150 200 250 300 350 London 1: 80% < ~210ms Median Ping (ms) But what does this prove? Players who picked up at least 10 ● Perhaps nothing! items per minute (reasonably active) .... if most of the Internet is less California 10: 80% < ~158ms than 250ms from anywhere central London 10: 80% < ~182ms ● Need evidence of regional locality... gj_armitage@yahoo.com 11/1/01 Page 5 SIGCOMM Internet Measurement Workshop (poster) From the incomplete-but-fun-research-department

Evidence of Locality # 1 Daily Use Cyclical usage patterns: 5 4.5 ● Usage patterns peak at different times, California 4 London different demographics 3.5 % Played Time 3 ● Peaks reflect afternoon and evening of 2.5 their respective locations 2 1.5 ● London 8 hours ahead of Palo Alto 1 ● Servers attract regional players 0.5 ● Supports hypothesis that clients prefer 0 0 2 4 6 8 10 12 14 16 18 20 22 24 'closer' server, other things being equal Time of Day (Californian time) W e e k l y U s e 1 . 8 % P l a y e d T i m e C a l i f o r n i a 1 . 6 L o n d o n 1 . 4 1 . 2 1 0 . 8 0 . 6 0 . 4 0 . 2 0 0 0 . 5 1 1 . 5 2 2 . 5 3 3 . 5 4 4 . 5 5 5 . 5 6 6 . 5 7 D a y o f W e e k ( C a l i f o r n i a n t i m e ) Sunday Saturday gj_armitage@yahoo.com 11/1/01 Page 6 SIGCOMM Internet Measurement Workshop (poster) From the incomplete-but-fun-research-department

Evidence of Locality #2 Using active players who picked up at least 10 items per minute during each game: The Origin of Players: Rank Calforinia Calforinia London London Games/Time(min) Origin Games/Time Origin ● Based on reverse lookups 1 323 / 3005 .ed.shawcable.net 108 / 1027 .pit.adelphia.net 2 192 / 2072 .cruzio.com 73 / 690 .Uni-Mainz.DE on each player's IP 3 124 / 1383 (RogersEAST/@Home) 75 / 679 .upc-d.chello.nl 4 119 / 1246 .018.popsite.net 50 / 606 (telnordia.se) address: 5 118 / 1221 .tx.home.com 53 / 604 .dyn.optonline.net 6 150 / 1200 .mediaone.net 44 / 565 (Rogers EAST/@Home) Californian server: mostly 7 132 / 1178 .pit.adelphia.net 35 / 463 .dyn.optonline.net 8 115 / 1151 .socal.rr.com 53 / 448 .dialup.tiscalinet.it North America 9 87 / 980 .pa.home.com 34 / 430 .pa.home.com 10 93 / 938 .sfba.home.com 20 / 288 .tx.home.com London server: mostly 11 69 / 799 .hsia.telus.net 24 / 273 .btinternet.com Europe and US East Coast () bracketed origins involved looking up 'whois' database after .in-addr.arpa failed. ● Since each server was otherwise identical, latency Table above shows origins of top 11 players on each seems plausible as the client- server. Outside the top 11, the Californian server observable metric on which a also saw dedicated players from ".jp" while the player chooses their server London server saw dedicated ".nl" and ".uk" players. There is also cross-over by players equidistant from either server. gj_armitage@yahoo.com 11/1/01 Page 7 SIGCOMM Internet Measurement Workshop (poster) From the incomplete-but-fun-research-department

Player effectiveness The aim is fragging there is no other reason to play... ● Skill and response time Frag(kill) Rate vs Median Ping 3.5 influence a player's ability to 3.25 frag (kill) others in the game Californian 3 London 2.75 ● Response time has human and 2.5 Frags/minute network components 2.25 2 1.75 ● Average frag rate vs median 1.5 1.25 ping hints at the negative 1 impact of high latency 0.75 0 50 100 150 200 250 300 350 400 ● A player with 45ms ping could Median Ping (ms) average 1 frag/min better than player with 200ms ping ● "Well, duh? " gj_armitage@yahoo.com 11/1/01 Page 8 SIGCOMM Internet Measurement Workshop (poster) From the incomplete-but-fun-research-department

Concluding thoughts.... Why is Jitter missing? Learn anything useful? ● Testbed's ping sampling too ● Players will tend to self-select coarse (10+ samples/minute) servers within 200ms 'radius' ● Lacked resources to deploy ● Two servers (separated by 147ms, distinct timezones and regional player populations) revised sampling method (20+ appear to validate this conclusion samples/second) ● Caveat: server ping estimates are only approximates ● Jitter impact may be significant ● Helps identify potential player (hand-eye co-ordination adapts better population relative to server(s) to constant latency) Looking forward.... ● Move to Half-Life or CounterStrike, dump QuakeIII ● Instrument servers to track packet loss and jitter ● No resources: I need multiple sites to host new servers with more accurate ping sampling gj_armitage@yahoo.com 11/1/01 Page 9 SIGCOMM Internet Measurement Workshop (poster) From the incomplete-but-fun-research-department