On the Geographic Distribution of On-line Games Servers and Players

Wu-chang Feng Wu-chi Feng

Summary, Class Discussion and Slides (16th March 2005)

CPSC 538A - Topics in Computer Systems

Presented By: Abhishek Gupta {agupta@cs.ubc.ca}

A) Paper Summary

This paper describes preliminary work upon the estimation of world-wide distribution of online game servers and game players. The motivation behind this study comes from the fact that game vendors are viewing online gaming as a lucrative opportunity and plan to replace individual hosted game servers by centrally hosted game services. The idea behind doing this is to obtain a tighter control over game quality, user experience and also to prevent cheating.

The intent of this study is to find a correlation between geographic distribution of game servers and the geographic distribution of game players. If a strong correlation exists between the two variables then game vendors would need to place game servers in many geographic locations, otherwise it may be possible to place clusters in a few strategic locations in the world. The other thing to notice is that information about the distribution of game players can be used to properly place and size clusters for hosted gaming services.

There are two quantities to be measured: a) The distribution of game servers, b) The distribution of players. For estimating the distribution of game servers a simple ‘ping’ command based scheme is used. In general all online games provide a centralized registry server with which every game server has to register before providing service. When a client wants to connect to an online gaming service, they contact the centralized gaming agency to download a list of potential game servers. After obtaining the list of game servers, each server is pinged to calculate the lowest RTT delay from the client and then the server with the lowest RTT is used to establish a session. Given this it is relatively easy to obtain the server IP addresses per game at a particular time. The authors rationalize that estimating the distribution of players is not an easy task because you need to coordinate with game publishers in order to obtain player authentication data. However, it can be easily done for a particular game server hosted by the authors themselves. This particular game server mshmro is located at OGI and hosts the counter-strike game. In this study it has been used to estimate the distribution of its players world-wide. Another important requirement of this study is the mapping of IP addresses to geographic location. The authors have been using a commercially available tool for doing that; however, it was able to map only 60% of the IP addresses. This inexactness in the mapping process may invalidate the entire results in this paper.

The evaluation results from the paper suggests that about 40-45 % of the game servers reside in North America and Europe, with about 10% lying in the southern hemisphere and 90% lying towards north of equator. A surprising result was observed during the characterization of the distribution of players. It was observed that a considerable number of players (45%) mshmro were from Europe and East Asia. So, this in effect nullifies the results from previous studies that suggest that players usually tend to connect to geographically closer servers in order to reduce the latency and have better gaming experience. Several explanations for this phenomenon are cited in the paper:

Disparity between geographic location and network topology: It has been reiterated in the paper that geographic and network proximity are not necessarily same. Although continuous development in network technology is bridging the gap between the two concepts however, there have been incidences in which network traffic between two machines has traveled across wider geographic area than their actual geographic distance. In short network proximity depends a lot upon the way the network traffic is routed.

Application server delays dominate network delay: This is again a possibility although the authors reject this idea in the case of mshmro trace as the server was very high quality hardware with excellent configuration and network infrastructure.

• Server selection mechanism for popular games is broken: Several gaming clients use some sort of Quick Connect utility which automatically contacts the centralized server agency and determines the best game server for this session. It is possible that these mechanisms are broken.

• The number of players on a server determines desirability over delay: Many online games are simply not fun because there are lesser number of players playing. For such cases players may choose to ignore a locally available well configured server and instead connect to a geographically remote server because it has more users connected to it.

• A shortage of servers overseas: This is an obvious effect because day time varies over the entire world and therefore during peak hours of playing in Europe, some players may decide to connect to game servers in the US in order to get a better gaming experience and similarly during peak hours in the US some of the players may decide to connect to servers in Europe.

It turns out that the distribution of players for a game is in fact dependent upon time of the day. It was observed that for mshmro, during early morning hours and afternoon there were more players from Europe and Asia, and then later as the day progressed, the number of players from Europe and Asia decreased and the number of players from the US increased. A week long trace showed that in fact this effect is periodic in nature.

Another interesting study which the authors do is a comparison between the distribution of players for their departmental web-server, the mshmro game server and the mshmro Counter Striker game forum over a period of time. They observe that users of the game server and the forum were much more correlated geographically than users accessing the departmental website.

Future work in this area is aimed at improving upon the accuracy of the results and expansion of the characterization to more completely capture geographic distributions of servers and players. Also, planned are the study of evolution of population as gaming applications change and their distribution around the world.

B) Class Discussion

• The authors performed the measurements for Counter-Strike, which is one of the most popular online games and their server in particular is one of the most popular ones to connect to, because of its high quality infrastructure.

• The authors evaluated several tools for mapping IPs to geographic locations before they zeroed in on the commercial one and indeed it performed far better than the open source ones.

• Quick Connect functionality in game clients have intelligence embedded in them, and this can make them potentially biased towards particular servers which are maintained in their history information as high quality.

• An aspect to note is that in the mshmro player distribution graph several parts of the world are sparse. This does not necessarily mean that there are no players in the sparse areas. Instead, it might be the case that those areas have their own local high end servers with support for their local languages and hence players from those areas never connect elsewhere.

• Cheating is a tough issue because of the simple fact that it is extremely hard to predict accurately if a player is cheating or she is actually very good at the game.