Transcript ppt - Computer Science - Worcester Polytechnic Institute

Network Characteristics for
Server Selection in Online Games
Mark Claypool
Computer Science Department
Worcester Polytechnic Institute
Worcester, Massachusetts, USA
http://www.cs.wpi.edu/~claypool/papers/game-server/
Introduction
•
Online games growing in popularity
•
Online games growing in variety
•
To support, increasing number of game servers
•
Which servers do players connect to?
– 4 a.m.  310,000 people playing over 100,000 online
games!
– Game consoles & hand-helds are all networkenabled, most games online multiplayer
– Then: few players on a LAN playing a FPS
– Now: thousands of players on a WAN playing a
FPS/RTS/MMO…
– Some run by players (most FPS and RTS games)
– Others run by game companies
MMCN, San Jose, CA, USA
2
January 2008
A Choice of Servers
•
Players often have choices
•
Choice matters
•
•
– Server browser lets player scan and select
–
–
–
–
Some servers may require cheat protection or mods
Some maps or game types may be more fun
Some servers can be full (maximum player capacity)
Latency! Ranging from milliseconds to seconds
Finding the “best” server more difficult when
multiple-players trying to play together
What does this server landscape look like?
MMCN, San Jose, CA, USA
3
January 2008
This Paper
1) Better understand characteristics of game
servers…
– How many and how often are servers up?
– Are there time of day or day of week correlations?
 Query game servers over month long period (oneto-many)
2) Observe if current game server deployments
sufficient for …
– Games of different genres?
– Single and multiple players?
 Simultaneous browsing by many game clients to
many game servers (many-to-many)
MMCN, San Jose, CA, USA
4
January 2008
Outline
• Introduction
• Server Browsing
• Measurement Methodology
• Analysis of Results
• Conclusions
MMCN, San Jose, CA, USA
5
(done)
(next)
January 2008
Game Server Browsing
•
Game company hosts master server
•
Game server starts
•
Game client starts
•
Game client queries each server
•
Player selects server to play
– Persists at well-known IP address and port
– Registers with master server
– Queries master server for list of game servers
– Map, players, game type …
– “ping” time as a measure of latency
– Launches game
MMCN, San Jose, CA, USA
6
January 2008
Typical Game Server Browser
MMCN, San Jose, CA, USA
7
January 2008
Outline
• Introduction
• Server Browsing
• Measurement Methodology
• Analysis of Results
• Conclusions
MMCN, San Jose, CA, USA
8
(done)
(done)
(next)
January 2008
Methodology
•
•
•
•
•
•
Select games (3)
– id Software  Quake 3, Quake 4, Doom 3
Query master servers for selected games
– 1 month for long-term trends
– Determine “permanent” servers
Select servers (20 for each game)
– Permanent and geographically distributed
Emulate game browsing with Qstat
– Emulate browsing of selected games
– Run from command line (easy to automate)
Select clients (25)
– Geographically distributed, on Planet Lab
Control and collect data from WPI
– 1 week for time of day and day of week trends
MMCN, San Jose, CA, USA
9
January 2008
Geographic Location of Servers and Clients
server
client
Outline
• Introduction
• Server Browsing
• Measurement Methodology
• Analysis of Results
(done)
(done)
(done)
(next)
– One-to-Many
– Many-to-Many
• Conclusions
MMCN, San Jose, CA, USA
11
January 2008
Number of Servers - Day of Week
(No correlation)
MMCN, San Jose, CA, USA
12
January 2008
Number of Servers - Time of Day
(No correlation)
MMCN, San Jose, CA, USA
13
January 2008
Servers - Permanent or Ephemeral
(Three regions. Most ephemeral)
MMCN, San Jose, CA, USA
14
January 2008
Number of Players – Percentage Filled
•
Average
– Q3 = 1.3
– Q4 = 0.45
– D3 = 0.93
(Few full. Many totally empty.)
MMCN, San Jose, CA, USA
15
January 2008
Number of Players – Day of Week
(No correlation.)
MMCN, San Jose, CA, USA
16
January 2008
Number of Players – Time of Day
(Correlation.)
MMCN, San Jose, CA, USA
17
January 2008
Latencies – Time of Day
(Correlation. Min latencies good!)
MMCN, San Jose, CA, USA
18
January 2008
Outline
• Introduction
• Server Browsing
• Measurement Methodology
• Analysis of Results
– One-to-Many
– Many-to-Many
(done)
(next)
• Conclusions
MMCN, San Jose, CA, USA
(done)
(done)
(done)
19
January 2008
Latency for Multiple Players
(Lowest average latency may not be fairest.)
MMCN, San Jose, CA, USA
20
January 2008
Maximum Latency
(Curves shift right with players. Knee flattens.)
MMCN, San Jose, CA, USA
21
January 2008
Player Performance versus Latency
MMCN, San Jose, CA, USA
22
January 2008
Acceptable Servers
(Few for First-Person, Many for Third-Person+)
MMCN, San Jose, CA, USA
23
January 2008
Conclusions
•
Correlation for day of week?
•
Correlation for time of day?
•
Server performance depends on …?
•
Servers can support multiple players…?
– Server uptime (NO)
– Player population (NO)
– Server uptime (NO)
– Server performance (NO)
– Player population (YES)
– Game generation (NO)
– Number of players playing together (YES)
– Third-person games (YES)
– First-person games (NO)
MMCN, San Jose, CA, USA
24
January 2008
Future Work
•
Data is public, so additional analysis possible
•
Server selection for ‘opaque’ servers
•
Tools to improve server selection
– Latencies of connected players correlated with
scores, or geography or …
– Geographic location in server selection
– Need help from game companies
– Make it easier
– Reduce network traffic
– Make games more fun
MMCN, San Jose, CA, USA
25
January 2008
Worcester, Massachusetts, USA
October 21-22, 2008
http://netgames2008.cs.wpi.edu/
Game related topics in Networks and Systems
(Sort of an MMCN for games!)
Papers due first week of May!
MMCN, San Jose, CA, USA
26
January 2008
Network Characteristics for
Server Selection in Online Games
Mark Claypool
Computer Science Department
Worcester Polytechnic Institute
Worcester, Massachusetts, USA
http://www.cs.wpi.edu/~claypool/papers/game-server/