Characterizing On
Transcription
Characterizing On
Characterizing On-line Games C. Chambers, W.-C. Feng, S. Sahu, S. Saha, D. Brandt Presented by: Vladislav Perelman JUB-2011 1 Outline Motivation Experimental Setup Predicting Game Workload Player Characterization 2 Motivation Everybody likes games Online gaming = Business Maximize profit + Minimize cost Player satisfaction Provisioning resources 3 Setup GameSpy trace (550 games): Casual Games trace (110 games): 4 Setup cs.mshmro.com trace: EVE online trace: Data includes games of different genres. More than 2 years of gaming data. Millions of connections and users. Sampling data every 10 minutes. 5 Game Workload On a large scale: Highly diverse player populations – hard to predict long-term behavior 6 Game Workload On a log scale: Casual Games GameSpy The distribution of popularity is similar to power-law distribution: f x=ax 7 Game Workload Short-term behavior: Half-Life players load FFT of year's data samples Figures demonstrate the cyclic nature of short-term gaming workloads 8 Game Workload Short-term behavior: Top 5 GameSpy games: Instantaneous load changes between identical points in time for consecutive weeks closely fit a t location-scale distribution: Most week-to-week load variations are under 15% It is feasible to model week-to-week load variations 9 Game Workload Share servers workload? Between games With web servers Player loads are very similar no matter what the game is Workload depends on the time of the day for games and the web Very limited benefit can be achieved by sharing servers 10 Game Workload Diurnal geographic patterns Regular daily activity patterns tied to working hours Lack of overlap between different geographic regions High degree of synchronization between loads on games and web servers 11 “ WAKE UP” (easy) Question! Name the Game! 12 “ WAKE UP” (easy) Question! 13 Know your players Patience is a virtue? Patience = willingness to reconnect to the server Hard to track due to “ Quick Start” feature Patience decreases exponentially Acceptable reconnections on cs.mshmro.com Not unexpected at all (YouTube 'angry german kid' to see live proof) 14 Know your players Hard to keep new players – many quit right after joining Game's ability to attract people decreases in time Disbalance in new and old players' abilities worsens the situation 15 Know your players “ Your update is ready to be installed...” Providing updates keeps players interested...at least in theory No large impact on the overall subscription Slight increase in average playing time per player 16 Know your players Individual approach Detect when the player is losing interest Decrease in number of sessions as well as duration of each session Deliver new content to keep the player from quitting 17 “ WAKE UP” (difficult) Question! Name the Game! 18 “ WAKE UP” (difficult) Question! 19 Looking back Game Popularity follows a power law Short-term game workloads are predictable Game workloads are synchronized with each other and with the web Gamers are impatient Player churn is increasing over time Gamers reduce the amount of time spent playing before quitting altogether 20 21 For those who are interested Games, in the order of appearance: EVE Online (title slide) ➢Counter Strike ➢Neverwinter Nights ➢Battlefield 1942 ➢StarCraft ➢Call of Duty ➢FullTilt Poker ➢Need for Speed ➢Super Mario Bros. ➢ 22