Scaling Distributed Virtual Environments using Socially Aware Load Partitioning and Interest Management

Abstract

Massively Multiplayer Online Games are played by millions of people from around the world who play together in large virtual environments. With sometimes even a million gamers playing the same game, the load on the virtual environments they play in needs to be distributed between servers to deal with this many players. Several techniques have been proposed, and are used in practice, to partition the load of these distributed virtual environments. Massively Multiplayer Online Games often try to facilitate as many players in a single virtual environment as possible, but even advanced spatial partitioning techniques cannot handle the load generated by the players when a large group of them gathers in a small area. Players usually have in these virtual environments a limited vision range making sure that game clients are not overwhelmed by status updates for all players in the environment, this approach also proves problematic when crowding occurs. We propose socially aware load partitioning and interest management, with the aim of reducing the load on the virtual environment and the game clients when crowding occurs, which takes the social relations between players into consideration when partitioning the load and deciding who should see whom. In this thesis we look specifically at how we can fairly compare various techniques and how we can effectively use social data to improve the scalability of distributed virtual environments. To evaluate and compare the performance of our socially aware techniques we design and implement an extensible framework which facilitates the evaluation of a wide variety of distributed virtual environment techniques and policies. As input for this framework, a workload is required that includes the components these techniques optimize for. As such we propose a new player behaviour model that supports player movement and interaction. In this framework existing techniques have been implemented in addition to our newly designed socially aware techniques and policies. The experiments using our framework show that socially aware techniques can result, assuming that the interaction probability between players can accurately be predicted, in better performance and scalability than their spatial counterparts. Further experiments with various proposed socially aware policies indicate that all these policies succeed, to varying degrees, in grouping friends together in the same partition. Experiments with socially aware interest management show that the proposed policies can achieve significant performance improvements, the best results are obtained when combining socially aware load partitioning and interest management.