Quality of Service For Computer Gaming An Evaluation of DiffServ

1
Quality of ServiceFor Computer Gaming-An
Evaluation of DiffServ
04gr832b Erling V. Matthiesen Flemming Olufsen
Jakob K. Larsen June 23rd 2004
2
Agenda

• Presentation of the problem (Jakob)
• Methods used (Erling)
• Results and conclusion (Flemming)
• Demonstration in the IP-lab

3
Problem

• Performance factors
• Delay
• Jitter
• Packet loss
• Bandwidth

4
Problem

• Traffic-classes
• Background
• Interactive
• Streaming
• Conversational

5
Problem

• Gaming traffic
• First Person Shooter (FPS)
• Real Time Strategy (RTS)
• Massively Multiplayer Online Role Playing Game
  (MMORPG)

6
Problem

• Problem statement
• How can a data path through a network be
  optimized to minimize delay and jitter in FPS
  network games?

7
Counter-Strike
8
Counter-Strike
9
Counter-Strike

• Delay
• Low
• Jitter
• Low
• Packet loss
• Some allowed
• Bandwidth
• Low

10
Concept

• Topology
• 1 router
• 4 routers
• Possibilities
• Resource Reservation Protocol (RSVP)
• Differentiated Services (DiffServ)

11
DiffServ

• Classification and marking
• DSCP-field
• Congestion management
• Packet scheduling
• WFQ
• LLQ

12
DiffServ

• Congestion avoidance
• RED and WRED
• Traffic shaping/policing

13
DiffServ
Edge device Classifying Marking
Edge device Classifying Marking
Core routers Per hop behaviour
14
Methods Used
15
Use of DiffServ

• Different possibilities in routers examined
• Two selected for further investigation
• Weighted Fair Queuing
• Low Latency Queuing

16
Weighted Fair Queuing

• Is configured for each interface
• Enabled by fair-queue 64 256 0
• Defines 256 queues
• 64 packets long
• 0 queues reserved for RSVP
• Disabled by no fair-queue

17
Low Latency Queuing

• Requires CEF (Express Forwarding) to be enabled
• Enabled by priority 300 1000
• Gives an maximum throughput of 300kbit/s
• Enables a burst of 1000 bytes exceeding the
  guaranteed bandwidth
• Bandwidth is only restricted if congestion occurs
• Disabled by no priority

18
Network Traffic

• Game traffic
• Cross traffic

19
Game Traffic

• Traffic from one Counter-Strike session were
  recorded
• Measurements made at all hosts

20
Game Traffic

• Game traffic appears bursty
• Two main concentrations in inter-departure times

21
Game Traffic

• Inter-departure times concentrated about 33 and
  50 ms

22
Game Traffic

• Dependance between packet size and inter-
  departure time investigated

23
Game Traffic

• Modeled game traffic
• With a mix of two normal distributions
• With a two state discrete time Markov model
• Other possibilities
• Repeating a time series section as model
• Extended Markov model

24
Game Traffic

• Mix of two normal distributions
• Weights calculated from traffic samples
• Normal distributions fitted with the mean and
  standard deviation of the traffic sample
• Discrete time Markov model
• Transition matrix derived from traffic samples

25
Game Traffic

• Server traffic
• Modeled with a normal distribution

26
Game Traffic

• The server and client traffic generator were
  implemented by us
• The cross traffic used were generated with Iperf

27
Acquiring data

• Routers configured via telnet
• Cisco configuration CLI used
• Scripts to enable and disable features developed

28
Acquiring data

• Traffic dumps taken with tcpdump
• Delay extracted with awk-scripts
• Statistics done with matlab

29
Experiments/Results Conclusion

30
Experiment Execution

• Conducted in the IP Laboratory
• Equipment
• 3 Cisco Routers 3620
• 1 Cisco Router 3631
• 2 Cisco Catalyst Switches 2950

31
Experiment Execution

• 2 Experimental Setups
• Small setup (1 Router)
• Large setup ( 4 Routers)

32
Experiment Execution
33
Experiment Execution
34
Experiment Results

• Experiments
• Small setup, no crosstraffic
• Small setup, 6M crosstraffic
• Large setup, no crosstraffic
• Large setup, 2M crosstraffic
• Large setup, 6M crosstraffic
• Large setup, 8M crosstraffic, CEF enabled

35
Experiment Results

• Experiments
• Small setup, no crosstraffic
• Small setup, 6M crosstraffic
• Large setup, no crosstraffic
• Large setup, 2M crosstraffic
• Large setup, 6M crosstraffic
• Large setup, 8M crosstraffic, CEF enabled

36
Experiment Results
Small setup no cross traffic
37
Experiment Results
Small setup no cross traffic
38
Experiment Results
Large Setup 6M cross
Large Setup 8M cross, CEF enabled,
39
Experiment Results
8M crosstraffic CEF enabled
6M crosstraffic
40
Experiment Results
8M crosstraffic CEF enabled
6M crosstraffic,Y-akse skal laves om
41
Conclusion

• Solid base for further experiments
• Gametraffic generators
• Routerscripts
• Results
• Not conclusive if QoS enhancement not effective
  or not working

42
Next steps

• Different settings of DiffServ
• RED/WRED
• Bandwidth reservations

43
Next steps

• Different setups for the experiments

44
Game over

• The end.