The Effects of Latency on
Online Madden NFL Football
James Nichols and Mark Claypool
Computer Science Department
Worcester Polytechnic Institute
Massachusetts, USA
NOSSDAV 2004 - Kinsale, County Cork, Ireland
1
Motivation
• Why study network games?
• In 2003, despite economic downturn,
games only industry to grow [ESA - 2004]
• 400,000 of the 1.3 million PS2 owners
that bought SOCOM play online
regularly [EGM - 10/03]
• Demanding application in terms of
network requirements
NOSSDAV 2004 2
Motivation
• Why study sports games?
• No one has yet…
• Are there different user or network level
requirements for sports games?
• Top selling genres in 2002 [“Essential Facts About the
Computer and Video game Industry”, The Entertainment Software Association, 2003]
• Computer Games
• Strategy - 27.4%
• First-Person Shooter - 11.5%
• Sports - 6.3%
• Console Games
• Strategy - not reported! (small)
• First-Person Shooter - 5.5%
• Sports - 19.5%
NOSSDAV 2004 3
Latency and Network Games
• Effects of latency on interactive
network applications well studied
• Web-browsing: seconds
• Audio conference: 100’s of milliseconds
• Games are highly interactive!
• First-person shooters: 100’s of ms
• Real-time strategy games
• What about sports?
• EA Sport’s Madden NFL Football Series
NOSSDAV 2004 4
Test environment
NOSSDAV 2004 5
Dumb Client Model
NOSSDAV 2004 6
Client-side prediction
NOSSDAV 2004 7
Simple Experiments
• Experiment 1
• Hypothesize that online Madden Football uses
client-side prediction to compensate for lag
• Large induced delay from Beta to Alpha
(1500ms)
• Alpha on offense
• Alpha puts man in motion
• Results:
• Beta sees player move on his local display first!
• Player on Alpha’s display is lagged
NOSSDAV 2004 8
Results
9
Results
10
Another experiment
• Experiment 2
• Hypothesize that online Madden
Football uses dumb-client model
• Large induced delay from Alpha to Beta
(1500ms)
• Results:
• Alpha sees player move on his local display
first!
• Player on Beta’s display is lagged
NOSSDAV 2004 11
One more simple experiment
• Experiment 3
• Hypothesize that symmetrical delay results in
roughly synchronized presentation
• Large induced delay in both directions
(1000ms each way)
• Results:
• Player movements are delayed on both displays, but
are synchronized
• Now we can suggest a lag compensation
model for online Madden Football
NOSSDAV 2004 12
Symmetrical Latencies
NOSSDAV 2004 13
Asymmetrical Latencies
NOSSDAV 2004 14
Asymmetrical Latencies
15
Effect of latency on running
• First, isolate running
• What about NPC interactions?
NOSSDAV 2004 16
17
NOSSDAV 2004 18
Experiment
• Hypothesize that offensive production
(Yards/Attempt) will decrease as
latency increases
• Play 3 full games at 8 different
latencies (0-2000 ms rtt)
• One team runs over and over again
• Other team punts the ball right back
NOSSDAV 2004 19
Quantitative Results
NOSSDAV 2004 20
Qualitative Results
• At higher latency, player movements
are lagged momentarily behind user
input
• Hard to get a feel for the timing of
when to spin, juke, stiff-arm, etc.
• Make “mistakes” at high latencies
• Run into defender
• Run out of bounds
NOSSDAV 2004 21
Example
Player is pressing left Player is pressing up
Running back goes out of bounds
NOSSDAV 2004 22
Conclusions
• Experiments suggest that Madden uses a
prediction of round-trip time to delay user
input to compensate for latency
• Latency does effect user performance
• About 30% of the range we evaluated
• Typical latencies don’t have an effect
• Future work is to evaluate the effects of
packet loss and investigate other sports
games
NOSSDAV 2004 23
The Effects of Latency on
Online Madden NFL Football
James Nichols and Mark Claypool
http://perform.wpi.edu
NOSSDAV 2004 - Kinsale, County Cork, Ireland
24