Provisioning Online Games: A Traffic Analysis of a Busy CounterStrike Server

1
Provisioning On-line Games A Traffic Analysis
of a Busy Counter-Strike Server

• Wu-chang Feng, Francis Chang, Wu-chi Feng,
  Jonathan Walpole
• Instructor Dr. Charles Krasic
• Presentation by zhen tan

2
Outline

• The goal of the paper
• About the game
• About the trace
• About the method of the research
• The analysis of the results
• Implication
• Conclusion

3
Goal

• To Understand the resource requirements of a
  popular on-line first person shooter game

4
Why games?

• Because of the Rapidly increasing of games in
  popularity
• Forrester Research 18 million on-line in 2001
• Consoles on-line
• Playstation 2 on-line (9/2002)
• Xbox Live (12/2002).

5
Why FPS?

• Gaming traffic dominated by first-person shooter
  games (FPS).

6
Why CS ?

• a modification to the popular Half-Life game
• one of the most popular and most
  network-intensive first person shooter games
  played over the Internet.

7
About the game...

• In the game Two teams of players competing in
  rounds lasting several minutes
• Rounds played on maps that are rotated over time
• Each server supports up to 32 players

8
About the game...

• Centralized server implementation
• Clients update server with actions from players
• Server maintains global information and
  determines game state
• Server broadcasts results to each client
• Sources of network traffic
• Real-time action and coordinate information
• Broadcast in-game text messaging
• Broadcast in-game voice messaging
• Customized spray images from players
• Customized sounds and entire maps from server

9
The trace the server

• cs.mshmro.com (129.95.50.147).
• Dedicated 1.8GHz Pentium 4 Linux server
• OC-3
• 70,000 unique players (WonIDs) over last 4
  months

10
The trace Summarizes of the trace
11
The trace total bandwidth

• Figure 1 (a) total bandwidth obeserved at the
  server
• Aggregate bandwidth - around 900 kbps

12
The trace total bandwidth

• Figure 1(a)Total bandwidth

13
The trace packed load

• Figure 1 (b) packed load observed at the server
• packet rate - around 800 pps

14
The trace packed load

• Figure 1 (b) packet load

15
About the hurst parameter

• Normalized variance aggregated sequence divided
  by the variance of the initial, unaggregated
  sequence
• Block sizes the number of frames per block.
• Hurst parameter (H) to measure the variability
  of the network
• ß beta is the magnitude of slope of the best
  fit line through the data points
• The relation of H and ß H1- 1/2ß

16
Variance time plot
Figure 2
Normalized to base interval of 10ms
17
The analysis of the results - behavior at varying
time scales

• Periodic server bursts every 10 ms and 50ms
• (a)the clients with state updates about every
  50ms
• (b)aggregating over the interval of 50ms smoothes
  out the packet load

Figure 3 (a)Interval size10ms
Figure 3 (b)Interval size50ms
18
The analysis of the results - behavior at varying
time scales

• Low utilization every 30 minutes
• (c) Server configured to change maps every 30
  minutes
• (d) increasing the interval size beyond the
  default map time of 30min

Figure 3(c)Interval size1sec
Figure 3 (d)Interval size30min
19
The analysis of the results - Finding the source
of predictability

• Games must be fair across all mediums (i.e.
  56kers)
• Aggregate predictability due to saturation of
  the narrowest last-mile link
• Histogram of average per-session client bandwidth

20
The analysis of the results - Packet sizes

• The outgoing bandwidth exceeds the incoming
  bandwidth.
• Rate of incoming packets exceeds that of outgoing
  packets.
• Server taking state information from each client.
• Servers aggregate and broadcast larger global
  updates.

21
Implications

• Routers, firewalls, etc. must be designed to
  handle large bursts at millisecond levels.
• Game requirements do not allow for loss or delay
  (lag).
• Routing devices that are not designed to
  handle small packets will see significant packet
  loss or even worse, when handle game traffic.
• Should not be provisioned assuming a large
  average packet size Partridge98.
• Router designers and vendors often make
  packet size assumptions when building their gear,
  often expecting average sizes in between 1000 and
  2000 bits (125-250 bytes). Thus, a significant
  shift in packet size from the deployment of
  online games will make the route lookup function
  as the bottleneck of the link speed.

22
Implications

• Routers, firewalls, etc. must be designed to
  handle large bursts at millisecond levels.
• there are buffers anywhere, they must...
• Use ECN (Explicit Congestion Notification).
• Be short (i.e. not have a bandwidth-delay product
  of buffering).
• Employ an AQM(active queue management) that works
  with short queues.

23
Conclusion

• The results show that the traffic behavior of
  this heavily loaded game server is highly
  predictable and can be attributed to the fact
  that current game designs target the saturation
  of the narrowest, last-mile link.
• As current routers are designed for bulk data
  transfers with larger packets, a significant,
  concentrated deployment of online game servers
  will have the potential for overwhelming current
  networking equipment.

24
Questions?