Behaviour and Performance of Interactive Multi-player Game Servers

1
Behaviour and Performance of Interactive
Multi-player Game Servers
Ahmed Abdelkhalek, Angelos Bilas, and Andreas
Moshovos
2
Introduction

• Investigate the scalability of services provided
  by servers
• Multi-player games are enabled by a central
  server
• FPS typically supports under 70 users

3
The First Experiment

• 32 PII-400 MHz computers
• Machines connected on the same LAN
• Quake (FPS game)
• 1 human controlled player with the rest being
  automated players

4
Server incoming network throughput per client
5
Server outgoing network throughput per client
6
Server response size
7
Server received request rate
8
Server response rate (relies/s)
9
Server execution time breakdown
10
Graphics Rendering Time for Clients
11
Server response rate as measured at the clients
12
The Second Experiment

• Each client computer simulates 16 players
• Compare statistics on two maps, a small map and a
  big one
• use PIII-800 MHz computer as server later
• Use EMON to monitor the CPU statistics

13
Server response rate (small map)
14
Server outgoing network throughout per client
15
Server execution time breakdown (small map)
16
Request processing time at the server (small map)
17
Server response size (small map)
18
Server response rate (large map)
19
Server execution time breakdown (large map)
20
Request processing time at the server (large map)
21
Server response rate (800MHz processors, large
map)
22
Request processing time at the server (800MHz
processors, large map)
23
Server execution time breakdown (800MHz
processors, large map)
24
Server CPU Behaviour
25
Conclusion

• Network bandwidth is not an issue
• Incoming bandwidth per player on server is
  constant with increasing number of players
• Outgoing bandwidth per player increases with
  increasing number of players
• Memory requirement is not an issue
• Use server response rate and average response
  time for evaluation

26
Conclusion (contd)

• Can support between 60 to 100 players with the
  equipment
• Server utilization increases linearly with the
  number of players
• Doubling the speed of CPU results in a less than
  double increase in the number of players
• Room for improving server performance at the
  microprocessor architecture level