Characterizing Online Games

I read this paper as a reference for a recent paper about live game streaming we submitted to the Workshop on Mining Social Network Dynamics, that will be part of the WWW Conference this year. When this paper was written, online gaming was something new and the authors provided an extensive characterization of several aspects related o online gaming, including player and server data. The objective is to give guidelines for game workload management in order to maximize player satisfaction and minimize costs. The authors of the paper Zipper Interactive (game development company), Portland State University,  IBM Research, and Networked Alternate Reality Creations (game development company).

The datasets used in the study are:
GameSpy: Real-time player population data on individual games.
MSN games: Similar to GameSpy, but focused on casual games (i.e. simple games downloaded directly from portals).
CS server: traces of a busy counter-strike server.
Eve online: complete server session history for a popular game.

Provisioning Resources:

Game population sizes varies a lot, what make them difficult to predict. Moreover, game popularity follows a power-law distribution.The authors found consistent daily and weekly variations in player activity. The application of a fast Fourier Transform to the date shows that the cycles occur in 24h and 168h (1 week) intervals. Instantaneous load changes between identical points in time of consecutive weeks follows a t location-scale distribution and most of the variations week-to-week variations are under 15%. For different games, usage patterns are very similar and even the usage patterns of commercial websites are similar to those found for games, which reduces the benefits of resource sharing. Finally, games present strong diurnal geographic patterns, which motivates geographic shifts of resources.

Player Characterization:

Players do not try to reconnect many times, the number of accepted reconnects follows a negative exponential distribution. Most of the players try the game for a short time and quit and the number of sessions of a player, before quiting, follows a weibull distribution. The authors present an interesting discussion about why the ability of MMORPGs to attract new players decrease with time, since new players come at a great disadvantage. My brother thinks that updates are the usual solution for this problem, since they put newcomers and 'professional' players close again, but the results disagree with him, showing that updates have small impact over the player population and playing time. The paper also  investigates how players that might be losing interest in the game can be identified automatically. Players that are quitting reduce their playtime.  Moreover, they play shorter sessions and have increased intersession time.

 This paper was very easy to read and understand. It is good that the subject of the paper is games and not video/audio, because I'm getting tired of these characterization papers. The authors seem to be specialists on the subject and some discussions are really interesting.

Link: http://www.thefengs.com/wuchang/work/cstrike/ton09_characterizing.pdf