Traffic Analysis of a Popular Massively Multiplayer Online Role Playing Game MMORPG

1
Traffic Analysis of a Popular Massively
Multiplayer Online Role Playing Game (MMORPG)
Jong Gun Lee and Sue Moon

• presented by Jong Gun Lee (jglee_at_an.kaist.ac.kr)
• Advanced Networking Lab,. KAIST

2
Breakdown

• Introduction
• Motivation Background
• Problem definition
• Part I Traffic identification
• Heuristic method
• Result of heuristic method
• Part II Basic analysis
• Session duration
• Packet size distribution
• Part III Further analysis
• Feasibility of the use of TCP protocol
• Latency
• Conclusion Future work

3
Introduction

• Game traffic
• Amount
• about 35 of total backbone traffic at 2000 2
• These days probably more
• Characteristic of game traffic
• is different from one of other typical
  applications
• Game categories
• First person shooter games
• Unreal Tournament and Counter Striker
• Real-time strategy games
• Starcraft and Warcraft I, II
• MMORPGs
• World of Warcraft , Lineage I, II, and GuildWar
• Casual games
• Board games, Racing games, and etc

4
Motivation

• So far,
• there are much researches on FPS and RTS games
• But, a little researches on MMORPG
• Why ?
• Until now, the major games are FPS and RTS games
• It is not easy to get the traffic of MMORPG
• because of game company policy and privacy issues
• But,
• MMORPGs are getting more popular
• We also need the characteristics of MMORPGs
• So,
• we collect and analyze the traffic of
• World of Warcraft (WOW), a popular MMORPG

5
Problem definition

• To find the similar and different things
• between FPS games and MMORPGs
• in aspect of traffic characteristics
• To check the feasibility
• of the use of TCP protocol in MMORPGs
• This result is a preliminary result of game
  traffic analysis

6
World of Warcraft

• Typical MMORPG
• Blizzard Entertainment
• At Dec 2004 test version
• Central server architecture
• TCP protocol
• Game server 3724 port

7
Dataset

• Collected the packets at KAIST core router
• 1Gbps fiber link
• Link utilization about 7080 (up to 100)
• Using GIGEMON 4.3GE
• Description

8
How to get the game traffic

• We consider the traffic which are
• long-lived flows (class I) and
• flows generated by normal playing (class II)

9
Heuristic method

• Terminology
• S packet to KAIST, C packet from KAIST, F a
  flow
• port port number, pkts of packets
• size flow size, dura flow duration
• Roadmap
• Step 0 Preprocessing step
• Collect Sport 3724
• Make flows according to 5-tuples
• Step 1 filtering step
• Cport gt 1024 and Cport ! p2p application port
  number (4662)
• Step 2 heuristic step
• Fdura gt 180
• 40 lt (Fsize / Fpkts) lt 70
• Fpkts gt 540

10
Result of heuristic method

• Result with Jun11 (having biggest of flows)
• after filtering step
• Type I error (false positive rate)
• 1.28 (1/352)
• Because we exclude non-class I and II
• Type II error (false negative rate)
• 0.01 (2/16,378)

11
Summary of Part I

• It is difficult to get game traffic from game
  company
• because of company policy and security issues
• Therefore, we make a simple heuristic method
• for game traffic identification
• to use simple flow characteristics
• Simple, efficient, and with good performance

12
Session duration

• About 50 players play within 1.5 hours, and
• about 90 players do within 5 hours
• This session duration is much longer that one of
  FPS
• Curve shape of these traffics is similar to one
  of FPS

FPS game (Counter-strike) 4
13
Session duration

• Fitting session duration(Dec24) with Weibull
  distribution
• Weibull distribution one of the most widely used
  lifetime distributions in reliability engineering
  1
• Good fitting result
• between session duration of WOW and Weibull
  distribution

CDF fitting
PDF fitting
14
Packet size distribution

• Average packet size of inbound traffic
• Diverse packet size because of central server
  architecture
• Average packet size of outbound traffic
• Size of each flow / of packets of the flow
• Mean 52.1271, standard derivation 3.8232
• FPS game mean outbound packet size about
  7080 3

15
Packet size distribution

• Size of each packet in one flow (Jun11)
• Plot packet size distribution of each flow
• Defined packet size according to activity
• such as walk and attack

Average packet size about 70bytes
16
Summary of Part II
17
Retransmission

• TCP protocol has several overheads
• when comparing with UDP protocol
• such as retransmission and longer header length
• We check whether retransmission affect on game
  playing, or not

18
Latency

• Definition of latency
• the time interval between sending a packet
• and receiving the corresponding ACK packet
• Almost all the packets received the ACK packets
• within 60msec
• FPS 50100ms 5

19
Summary of Part III

• Retransmission
• At most only 1 of total packets per flow
• Retransmission scheme may be no problem
• Latency
• Almost all of the packets receive ACK packets
  within 60ms
• When comparing to sensitivity of FPS games
• latency may be no problem
• TCP protocol is proper to MMORPG architecture
• with these result

20
Conclusion

• Simple but efficient game traffic identification
  method
• with several flow characteristics
• On two axis, analyzed the game traffic
• In basic step
• Compare the characteristics of WOW with ones of
  FPS games
• Session duration and packet size distribution
• In further step
• Check the feasibility of the use of TCP protocol
• Retransmission and latency

21
Future work

• More generic game traffic identification
• More in-depth analysis of game traffic
• Other network characteristics
• such as bandwidth and network dynamics
• TCP vs. UDP protocol in game architecture
• Game architecture
• with less delay and
• robustness to the number of player
• Analysis on other genres game

22
Reference

• 1 http//www.weibull.com/lifedatawebcontents.htm
  , 2003
• 2 Sean McCreary, kc claffy, Trends in Wide
  Area IP Traffic Patterns, ITC Specialist
  Seminar, 2000
• 3 Wu-chang Feng, Francis Chang, Wu-chi Feng,
  Jonathan Walpole "A Traffic Characterization of
  Popular On-line Games", IEEE/ACM Transactions on
  Networking, June 2005
• 4 Francis Chang, Wu-chang Feng, "Modeling
  Player Session Times of On-line Games", in Proc.
  of NetGames 2003, May 2003
• 5 Tom Beigbeder, Rory Coughlan, Corey Lusher,
  John Plunkett, Emmanuel Agu, and Mark Claypool,
  The Effects of Loss and Latency on User
  Performance in Unreal Tournament 2003, In
  Proceedings of ACM Network and System Support for
  Games Workshop (NetGames), September 2004.