The Effects of Latencyon User Performance in
Warcraft IIINathan Sheldon, Eric Gerard, Seth
Borg, Mark Claypool, Emmanuel AguComputer Science DepartmentWorcester Polytechnic Institute
Worcester, MA, USAhttp://www.cs.wpi.edu/~claypool/papers/war3/
May 2003NetGames'03, Redwood City, CA, USA2
Why Study Network Games?•In 2000, U.S. economy grew by 7% while
computer game industry grew by 15%
•60% of Americans age 6+ play computer games
•221+ million computer games sold in 2002– 2 games for every household in America
“Top Ten Industry Facts”, Interactive Digital Software Association, May 2003.
“Economic Impacts of the Demand for Interactive Entertainment Software”, Interactive Digital Software Association, 2001.
May 2003NetGames'03, Redwood City, CA, USA3
Why Study Warcraft III?
•Top selling computer game genres– Strategy (27.4%)– Children’s (15.9%)– Shooter (11.5%)– Family Entertainment (9.6%)
•Warcraft III set sales record – Fastest to sell 1 million copies
“Warcraft III - Shatters Sales Records Worldwide...”, Blizzard Press Release, October 2002
“Top Ten Industry Facts”, Interactive Digital Software Association, May 2003.
May 2003NetGames'03, Redwood City, CA, USA4
Network Games and Latency• Latency degrades performance of interactive
applications– Web-browsing – seconds– Audioconference – 100’s of milliseconds– First Person Shooters (FPS) – 100’s of milliseconds
•Real-Time Strategy (RTS)? • Knowing effects of latency useful for
– Building better network games– Building better networks to support games (QoS)
Effects of Latency on Warcraft III (RTS)
May 2003NetGames'03, Redwood City, CA, USA5
Outline
•Introduction
•Experiments
•Analysis
•Conclusions
May 2003NetGames'03, Redwood City, CA, USA6
Warcraft III Overview
RTS User Interaction Components:• Exploration• Building• Combat
May 2003NetGames'03, Redwood City, CA, USA7
Exploration Map
Performance?• Time(to reach end)
May 2003NetGames'03, Redwood City, CA, USA8
Building Map
Performance?• Time(to build tech-nology tree)
May 2003NetGames'03, Redwood City, CA, USA9
Combat Map
Performance?• Games Won• Unit Scores
May 2003NetGames'03, Redwood City, CA, USA10
Controlling Latency
•Warcraft III uses client-server– Set computer B as server (also a client)– Set computer C or D as client
•NIST Net on computer A – Induce latency [0 ms to 3500 ms]
May 2003NetGames'03, Redwood City, CA, USA11
Outline
•Introduction
•Experiments
•Analysis– Application Level – Network Level– User Level
•Conclusions
May 2003NetGames'03, Redwood City, CA, USA12
Building and LatencyBuild Time vs. Latency
R2 = 0.0516
7:40
7:55
8:09
8:248:38
8:52
9:07
9:21
9:36
0 500 1000 1500 2000 2500 3000 3500 4000
Latency (ms)
Build
Tim
e (m
:s)
Time To Build Linear (Time To Build)
May 2003NetGames'03, Redwood City, CA, USA13
Exploration and LatencyExplore Time vs Latency
R2 = 0.6334
3:50
4:04
4:19
4:33
4:48
5:02
5:16
5:31
0 200 400 600 800 1000 1200
Latency (ms)
Expl
ore
Tim
e (m
:s)
Time To CompleteLinear (Time To Complete)
May 2003NetGames'03, Redwood City, CA, USA14
Combat and Latency (1)Unit Score Difference vs. Latency
R2 = 0.0138
-3000-2000-1000
0100020003000
0 200 400 600 800 1000 1200 1400 1600Latency (ms)
Uni
t Sco
re D
iffer
ence
May 2003NetGames'03, Redwood City, CA, USA15
Combat and Latency (2)
May 2003NetGames'03, Redwood City, CA, USA16
Outline
•Introduction
•Experiments
•Analysis– Application Level – Network Level – User Level
•Conclusions
May 2003NetGames'03, Redwood City, CA, USA17
Bandwidth
3.8 Kbps
4.0 Kbps
6.8 Kbps
May 2003NetGames'03, Redwood City, CA, USA18
Inter-Packet Times
May 2003NetGames'03, Redwood City, CA, USA19
Payload Distributions
May 2003NetGames'03, Redwood City, CA, USA20
Bandwidth and Latency
•Battle.net games had ~100 ms latency•Trace combat maps with induced
latency
May 2003NetGames'03, Redwood City, CA, USA21
Payload Distributions and Latency
May 2003NetGames'03, Redwood City, CA, USA22
Commands and Latency•Pilot studies suggest 6 bytes of
overhead per command•Remove 6 bytes from each packet
payload•Add up remaining command payloads
May 2003NetGames'03, Redwood City, CA, USA23
Outline•Introduction
•Experiments
•Analysis– Application Level – Network Level – User Level
•Conclusions
May 2003NetGames'03, Redwood City, CA, USA24
User-Level Analysis•0-500 ms latency, users could easily adjust•800+ ms, game appeared erratic
– Degradation in gaming experience•500-800 ms degradation depended upon
– User •More skilled were more sensitive
– Strategy•Micro managers were more sensitive•Combat managers were more sensitive
May 2003NetGames'03, Redwood City, CA, USA25
Conclusions• Typical Internet latencies do not significantly
affect user performance in Warcraft III– Some effect on exploration– No statistical effect on building or combat
• RTS game play emphasizes “strategy” (which takes 10s of seconds or minutes), not “real-time”
• RTS games less sensitive to latency than are FPS– RTS in QoS class similar to that of Web browsing
• At the network level:– Small packets with low bandwidth– Command aggregation at higher latencies
May 2003NetGames'03, Redwood City, CA, USA26
Future Work
•Effects of latency on user strategies•Other RTS games
– Age of Empires– Command and Conquer
•Effects of latency on other genres– First Person Shooter– Role Playing Game
•Effects of loss
The Effects of Latencyon User Performance in
Warcraft IIINathan Sheldon, Eric Gerard, Seth
Borg, Mark Claypool, Emmanuel AguComputer Science DepartmentWorcester Polytechnic Institute
Worcester, MA, USAhttp://www.cs.wpi.edu/~claypool/papers/war3/