To Infinity and Beyond: Scaling Massively Multiplayer Games
Authors
Cohen, Ron ; Ejlersen, Anders ; Kristensen, Rasmus
Term
4. term
Education
Publication year
2010
Abstract
Denne afhandling undersøger, hvordan man fordeler beregningsarbejdet i massivt multiplayer online-spil (MMOG’er) ved at opdele spilleverdenen mellem flere servere. Vi fokuserer på geografisk partitionering – det vil sige at dele kortet i regioner, som hver server håndterer. Vi analyserer tre eksisterende to-dimensionelle metoder: statisk partitionering, matrix-partitionering og JoHNUM-partitionering, og udvider dem, så de kan bruges i tredimensionelle virtuelle miljøer. For at afprøve metoderne byggede vi prototypespillet Rock Pounder og evaluerede de udvidede metoder i to scenarier, der afspejler almindelige MMO-spillemønstre. Resultaterne viser, at statisk partitionering fungerer godt, når arbejdsbelastningen er stabil. Ved dynamiske belastninger klarer den dynamiske JoHNUM-tilgang sig bedst, især fordi den bruger en mere grundig foranalyse af verdenen før opdelingen. Overordnet set skalerer JoHNUM pænt, så flere spillere kan understøttes i takt med at antallet af servere øges.
This thesis examines how to distribute computing load in massively multiplayer online games (MMOGs) by splitting the game world across multiple servers. We focus on geographical partitioning—dividing the map into regions assigned to servers. We analyze three existing two-dimensional methods: Static partitioning, Matrix partitioning, and JoHNUM partitioning, and extend them to work in three-dimensional virtual environments. To test these methods, we built a prototype game, Rock Pounder, and evaluated the extended approaches in two scenarios that reflect common MMO play patterns. The results show that Static partitioning works well when the workload is stable. For changing, dynamic workloads, the dynamic JoHNUM approach performs best, largely because it relies on a more thorough pre-partition analysis of the virtual world. Overall, JoHNUM scales well, supporting more players as more servers are added.
[This abstract was generated with the help of AI]
Documents