Traffic Analysis of Avatars in Second Life

1
Traffic Analysis of Avatars in Second Life

• James Kinicki and Mark Claypool

Computer Science Department Worcester Polytechnic
Institute Worcester, Massachusetts, USA
http//www.cs.wpi.edu/claypool/papers/second-life
/
2
Introduction Online Virtual Worlds

• Users interact with world and other users via
  avatar
• For both business and leisure
• Walt Disney to Wells Fargo creating new virtual
  worlds
• Second Life most well-known virtual world today
• Over 12 million users
• Over 950 thousand having logged in in the past
  month

• Online Games
• Forms of interaction are limited to gameplay
  interactions
• Static, preexisting environments
• Low bitrate requirements (small, frequent packets)

• Virtual Worlds
• Flexible forms of interaction with emergent user
  behavior
• Dynamic environments - users add objects, media
• Significantly higher bitrates

What is the turbulence for virtual worlds?
3
Research Questions

• Are results from previous experiments (Fernandes
  et al., NOSSDAV 2007 6) reproducible?
• Does turbulence of Second Life vary with number
  of objects and avatars in zone?
• What is turbulence for teleportation?
• How does turbulence of Second Life compare with
  online games?

4
Outline

• Introduction (done)
• Methodology (next)
• Analysis
• Conclusions

5
Methodology

• Determine avatar actions to study
• Select Second Life zones to visit
• Setup measurement environment
• Gather data
• Analyze results

6
Methodology - Avatar Actions

• Teleporting - user selects new zone on map and
  teleports avatar, standing after arrival
• Standing - avatar remains still (no movement of
  mouse or keyboard)
• Walking - avatar moves in straight line across
  single zone at constant speed
• Flying - avatar flies in circle around the edge
  of single zone at constant speed

7
Methodology - Zones

• Hypothesis is that both objects and avatars
  affect the Second Life network traffic ? find
  four zones
• sparse and deserted
• dense and deserted
• dense and crowded
• sparse and crowded (but this one tough)

8
Methodology - Measurement Environment

• PC w/Windows XP pro on a 2.8 GHz P4 with 1 GB of
  RAM
• Second Life v 1.18.2
• Residential broadband connection
• Based on 6, not a bottleneck
• Cable modem (4.5 Mb/s down, 1 Mb/s up)
• Wireshark
• Capture all traffic to/from Second Life servers

9
Methodology - Data Gathering

• Since single server handles one zone,
• not actually leave zone as avatar moves
• Flying goes in a circle
• Walking goes in a line, can cross zone in about
  30 seconds
• Teleport waits until all objects rendered
• Deserted have no other avatars
• Dense, crowded between 90 and 100 people
• Streaming media was turned off

10
Outline

• Introduction (done)
• Methodology (done)
• Analysis (next)
• Conclusions

11
Analysis - Bandwidth

• Zone
• Dense, crowded requires more than sparse and
  deserted zones
• Upstream/Downstream
• Up 10x less than down
• Fewer correlations (zone, action)
• (down for rest of analysis)

• Action
• Walking, flying require more (8x) than standing
• Flying requires only slightly more than walking
• Teleportation requires similar to flying

12
Analysis Packet Size

• Action
• Standing smallest
• Walking and flying larger
• Teleport always moderately large

• Zone
• Dense and crowded both larger
• Sparse and deserted small

13
Analysis Inter Packet Time

• Frequent packets, regardless of zone or action
• Half the packets arrive back-to-back.
• Highest rate for teleporting

14
Comparison to Online Games

• Second Life turbulence far greater than online
  games
• Bandwidth use 10x-100x
• Packet sizes 15x-20x
• Packets sent 3x-20x
• Large turbulence suggests meeting QoS over wide
  range of networks could be a challenge

15
Conclusions

• Previous results (Ferdandes et al. 6) somewhat
  reproducible
• Yes, zone and avatar affect Second Life
  turbulence
• No, magnitude of bandwidth different
• ? Suggests further study needed before models
• Turbulence impacted by objects and avatars
• Dense, crowded 10x bwidth and psize of sparse,
  deserted
• Dense, deserted 2x bwidth and psize of sparse,
  deserted
• ? Suggests avatars plays larger role than objects
• Teleport most turbulence for all zones
• Second Life more turbulence than online games
  (10x)
• Bandwidth results
• Help users in broadband selection choices
• Help ISPs and virtual world server hosting
  w/capacity planning
• Packet results
• Used for classification and simulation

16
Future Work

• Private zone w/full control over access
• Control exact number of objects and avatars
• Caching, in conjunction with motion
• Quality of Service requirements
• May be similar to third-person online games (or
  not)

17

• Worcester, Massachusetts, USA
• October 21-22, 2008
• http//netgames2008.cs.wpi.edu/
• Game related topics in Networks and Systems
• (Like a NOSSDAV for games!)
• Papers due June 15th!

18
Traffic Analysis of Avatars in Second Life

• James Kinicki and Mark Claypool

Computer Science Department Worcester Polytechnic
Institute Worcester, Massachusetts, USA
http//www.cs.wpi.edu/claypool/papers/second-life
/