We present a network software architecture for solving the problem of scaling large distributed simulations. The motivation for our effort is to expand the capability of virtual environments to serve large numbers (more than 1,000) of simultaneous users. The fundamental idea of our approach is to logically partition virtual environments by associating spatial, temporal, and functionally related entity classes with network multicast groups. Furthermore, we exploit the actual characteristics of the real-world large-scale environments that are simulated by focusing or restricting an entity's processing and network resources to its area of interest via a local Area of Interest Manager (AOIM) and a persistent object protocol. We first discuss related work in the area of networked virtual environments and the problems of developing scalable VEs. The dissertation also provides a taxonomy for discussing VEs in terms of communication methods, data, processes, and views. Moreover, we describe the Distributed Interactive Simulation (DIS) efforts and the limits of DIS today. Finally, we present our theory and the results of simulations using the AOIM. We used data from the U.S. Army National Training Center and the Janus combat model to show how the movement rates and densities of thousands of combat systems allows the use of the AOIM by an military entity to limit network traffic and simulation computation, maintain acceptable reliability, and minimize the effects of latency.
展开▼
