Multiplayer games and virtual worlds Part2: Second life

1
Multiplayer games and virtual worlds-Part-2
Second life

• CS 7270
• Networked Applications Services
• Lecture-14

2
Traffic Analysis Beyond This World the Case of
Second Life

• Stênio Fernandes
• Rafael Antonello
• Josilene Moreira
• Djamel Sadok
• Carlos Kamienski

Federal University of Pernambuco Networking and
Telecommunications Research Group
3
Agenda

• Motivation
• SL Architecture
• Methodology
• Performance Evaluation
• Discussion
• Conclusions

4
Motivation

• SL is a VW that has been gaining a lot of
  popularity
• Gaming, business , LSL Script extensions
• The application can generate many different
  traffic profiles (voice, music, video, and data)
• Independent music stream activation
• The world is at the server!
• ISPs are concerned to provide an adequate QoS to
  guaranty a good experience
• Little effort has been devoted to understand
  Second Lifes traffic profile and the
  implications to the traffic management area

5
SL Grid Architecture
SIM routing
XML-RPC/TLS
IM
64.128.0.0 64.129.255.255
Hayok Physics lib.
6
Methodology

• More than 100 hours of experiments (collected
  from January 24 to 29 of 2007)
• Scenarios (SL traffic index used)
• A popular place (Goddess of Love)
• A unpopular place (Menglin II)
• Network connections
• A 100 Mbps network called UFPE
• A residential ADSL connection of 600 kbps called
  ADSL
• Actions and external sources
• 3 different actions (Standing still, walking and
  flying)
• Experiments with and without listening to music

7
Methodology

• Metrics
• Throughput
• Packet size (TCP, UDP and IP headers)
• Packet inter-arrival time
• Traffic volume (UDP/TCP)
• 10-minute experiments
• Metrics were computed for each 10 seconds
  interval ? 60 samples
• 99 of confidence intervals

8
Throughput - Flying
Figure 2 Throughput (Flying, with music)
Figure 3 Throughput (Flying, no music)

• Average throughput is around 300 kbps for the
  ADSL
• Average throughput is around 400 kbps for the
  UFPE

• Average throughput is around 180 kbps for the
  ADSL
• Average throughput is around 210 kbps for the
  UFPE

Requirements beyond those from average Games!!
9
Average throughput Popularity

• Popular places generate around 2.5 more traffic
  than unpopular ones

10
Throughput Standing, Walking, Flying

• For an unpopular place
• when the avatar is standing still, up to 97 of
  the time the throughput is below 20 Kbps,
• whereas with some form of movement it is between
  60 Kbps and 110 Kbps in 87 of the time
• For the popular place
• even when the avatar is standing still, 93 of
  the time, the throughput is between 75 and 350
  Kbps

Throughput distribution (unpopular place)
Note the difference between flying and walking!!
Throughput distribution (popular place)
11
Packet size (Avg. and distribution)
Packet size distribution (popular place)
Packet size (bytes)

• In general, server-to-client packets are bigger
  than the client-to-server ones
• Average packet sizes generated by both the server
  and the client at the unpopular place are similar

Packet size distribution (unpopular place)
12
TCP-UDP Traffic Volumes Packet Interarrival Time
Traffic volume for TCP and UDP
Packet Interarrival time (average)

• TCP packets are only exchanged in scenarios with
  music
• Inter-arrival time is shorter for more complex
  environments (i.e. popular places) and when the
  avatar is moving faster

13
Packet size x Interarrival time
Throughput (average)

• When the avatar is standing still in an unpopular
  place
• the client sends small packets (50 to 150 bytes)
  to the server
• whereas the server updates the client with
  slightly larger packets (100 to 250 bytes),
  mostly below 200ms of interarrival time

Packet Size x Interarrival Time (unpopular, stand
still, no music)
14
Packet size x Interarrival time

• The profile for the client-to-server
  communications remains almost the same
• The server increased the size of the packet
  payload

Packet size x Interarrival time (Popular,
standing still, no music)
15
Discussion

• SL client can generate diverse network patterns
• Users can attach external traffic sources within
  their land boundaries
• SL requires more stringent network parameters
  than others 3D online games
• Using some knowledge of traffic patterns of the
  SL clients, a local ISP could monitor and
  forecast the aggregate bandwidth requirement to
  plan a link capacity upgrade
• Forecast for more than 30 million users soon

16
Conclusions

• This paper takes a first step on understanding
  the SL architecture and profiling the traffic
• SL makes intensive use of network resources
• Some SL places may generate higher and
  unpredictable traffic patterns in the future
• The increase of the simultaneous online users
  number is a potential concern for network
  management
• Can latency compensation or other techniques can
  be applied?