Overview Motivation Fairness Approach Implementation Client-site - - PDF document

Swinburne University of Technology 1

• S. Zander, I. Leeder, G. Armitage, J. But

<szander,garmitage,jbut@swin.edu.au> <i_leeder@hotmail.com>

Centre for Advanced Internet Architectures Swinburne University of Technology

Achieving Fairness in Multiplayer Network Games through Automated Latency Balancing

SIGCHI ACE 2005, June 15th-17th

http://caia.swin.edu.au szander@swin.edu.au Page 2 SIGCHI ACE 2005, June 15th-17th

Overview

Motivation Fairness Approach Implementation Client-site Bots Evaluation Results Conclusions and Future Work

Swinburne University of Technology 2

http://caia.swin.edu.au szander@swin.edu.au Page 3 SIGCHI ACE 2005, June 15th-17th

Motivation

Multiplayer network games have become very

popular and have evolved into some kind of sports

Competitions and leagues are very popular and

comparable to sporting competitions

Professional gamers earn their living just by gaming; they

have fans and TV shows

Many people playing on amateur level take it seriously

Games requiring fast player reactions are very

sensitive to the Quality of Service (QoS) of the underlying computer network(s)

Fairness is very important

Game design (we do not talk about this) Network QoS differences

http://caia.swin.edu.au szander@swin.edu.au Page 4 SIGCHI ACE 2005, June 15th-17th

Motivation cont’d

Focus on fast-paced games e.g. first person

shooters where fast player reactions are crucial

Focus on latency/delay (also called lag)

Influence of jitter has not been sufficiently studied Influence of packet loss is much smaller

Previous work has shown that

Efficiency of players decreases with increasing latency Latency differences cause unfairness

Latency differences are caused by

Network access technology Distance between client and server (propagation delay) Congestion in the network

Swinburne University of Technology 3

http://caia.swin.edu.au szander@swin.edu.au Page 5 SIGCHI ACE 2005, June 15th-17th

Fairness Approach

Implement tool that automatically equalizes

players latency by adding artificial lag

Evaluate effectiveness of approach using

human or computer players

Compare ‘objective’ performance metrics (e.g. kill

rate) for players (player groups) with different latencies

Use hypothesis testing to determine if differences

are significant

If differences are significant there is unfairness Eliminate factors other than delay

http://caia.swin.edu.au szander@swin.edu.au Page 6 SIGCHI ACE 2005, June 15th-17th

Implementation

Self-Adjusting Game Lagging

Utility (SAGLU)

Game independent proxy-

application between game clients and server

Extensible multithreaded C++

implementation

Retrieves player information

from the server (IP address, port and latency)

Equalizes player latencies by

adding fake delay

SAGLU G ameServer Gam eType Player Comms Half-Life Q uake2 Q uake3Arena EnemyTerritory n TrafficShaper D ummyNet n

Swinburne University of Technology 4

http://caia.swin.edu.au szander@swin.edu.au Page 7 SIGCHI ACE 2005, June 15th-17th

Implementation cont’d

Delay adjustment algorithm

How to determine amount of additional artificial delay? How to add the delay? How frequently to measure player's network delay and

adapt the additional delay?

Implemented simple algorithm

for (i in 1:#Players) P[i].NetDelay = getNetDelay() for (i in 1:#Players) P[i].AddDelay = min(max(P[1:#Players].NetDelay), MaxTolerableDelay) – P[i].NetDelay if (P[i].AddDelay > 0) setAddDelay(P[i].IPAddress, P[i].Port, P[i].AddDelay) sleep(AdaptationIntervalTime)

http://caia.swin.edu.au szander@swin.edu.au Page 8 SIGCHI ACE 2005, June 15th-17th

Client-side Bots

Usability trials with human players

Necessary for conclusive evaluation Human responses are highly unpredictable (very difficult

to eliminate all unwanted factors)

Resource and cost intensive (time, equipment, money)

Client-side computer players (bots)

Easy to eliminate unwanted factors e.g. bots behave

identical, do not get tired, do not change playing style etc.

Far less resources needed Bots are different from humans

Incapable of complex navigation (only line of sight) Very effective delay compensation (movement prediction) But send real network traffic and therefore should be

affected by network delay

Swinburne University of Technology 5

http://caia.swin.edu.au szander@swin.edu.au Page 9 SIGCHI ACE 2005, June 15th-17th

Evaluation

FreeBSD PC with 2.4GHz and 1.25GB RAM Emulate network delay using dummynet Static Dynamic (changing every second with exponential distribution) Small simple map without obstacles (e.g. lava pits, elevators)

and powerful explosive weapons

4 bot players (same configuration) SAGLU adaptation interval of 5 seconds Experiments How do bots react to delay? Do bots experience unfairness? Can SAGLU balance unfair games? Average results over 15 games (15 minutes duration)

http://caia.swin.edu.au szander@swin.edu.au Page 10 SIGCHI ACE 2005, June 15th-17th

Evaluation Results

How do bots do react to delay?

100 200 300 400 0 .0 0 .2 0 .4 0 .6 0 .8 1 .0 1 .2 Delay [ms] N o rm a liz e d M e a n K ill R a te Quake 2 Bots static Quake 2 Bots dynamic Quake 3 [5] Quake 3 [6] 100 200 300 400 2 0 4 0 6 0 8 0 1 0 0 Delay [ms] K ills [% ] Shotgun Blaster Machinegun Grenade

Kill rate decrease (bots & humans) Weapons used for kills (bots)

Swinburne University of Technology 6

http://caia.swin.edu.au szander@swin.edu.au Page 11 SIGCHI ACE 2005, June 15th-17th

Evaluation Results cont’d

100 200 300 400 2 4 6 8 1 0 Delay [ms] M e a n K ill R a te [1 /m in u te ] Non-lagged Lagged 100 200 300 400 2 4 6 8 1 0 Delay [ms] M e a n K ill R a te [1 /m in u te ] Non-lagged Lagged

Dynamic delays without SAGLU Dynamic delays with SAGLU

Do bots experience unfairness and can SAGLU

balance the games?

http://caia.swin.edu.au szander@swin.edu.au Page 12 SIGCHI ACE 2005, June 15th-17th

Conclusions and Future Work

Client-side bots behave similar to humans

Kill rate decreases and weapons with area effects become

more effective with increasing delay

Experience unfairness because of delay differences But performance (kill rates) cannot be directly compared

SAGLU effectively balances the game

(http://caia.swin.edu.au/genius/tools/saglu-0.1.tar.gz)

Usability trials with human players in real networks

Refine delay adjustment algorithm Optimize parameters (e.g. adapt. interval, tolerable delay)

Measure performance and overhead

Swinburne University of Technology 7

http://caia.swin.edu.au szander@swin.edu.au Page 13 SIGCHI ACE 2005, June 15th-17th

Thanks for your attention!