A Visibility-Driven Approach to Managing Interest in Distributed Simulations with Dynamic Load Balancing

Abstract

Distributed simulations that support a massive number of users typically divide the virtual world into zones that are managed by separate servers to evenly distribute resources and achieve scalability. However, such zoning restricts cross-zonal interactions and exposes the division of the world to the participating parties. Problems such as crowding one zone among others defeats the very purpose of interest management and makes geographic partitioning inefficient for modeling interactions. In this work, we have designed and implemented a visibility-driven approach to make the partitioning transparent to users. The effectiveness of this distributed architecture is tested through a prototype implementation. We also introduce a novel idea to dynamic load balancing that can be achieved in real-time without modifying the communication architecture. By increasing the granularity of the partitioning and providing a layered approach to zoning, transient crowding can be handled by adoptively dispersing parts of the crowded zone to adjacent servers.