Reducing view inconsistency by predicting avatars' motion in multi-server distributed virtual environments

Abstract

Multi-server distributed virtual environment (DVE) system contains a set of servers to support a huge number of geographically distributed users. It provides a shared virtual scenario where users represented by avatars participate in and interact with each other through networks. However, there are two technical challenges in designing an efficient DVE system with this architecture: view inconsistency caused by network delays and server overloading caused by uneven distribution of users. While the first problem affects users' perception of the VE and causes user disputes, the second problem affects the system response time. We present a new partitioning framework to simultaneously address the two problems, in which the DVE scenario is divided into a set of disjoint grids called virtual cells. The virtual cells are then indexed by a region quadtree and an algorithm is proposed to search the region quadtree so as to equally partition avatars among servers. By using probabilistic models to predict each avatar's future motion and the virtual region its area of interest (AOI) will cover, a density-based method is proposed to help each server construct avatar clusters from its local avatars in parallel. By estimating the possible distances of avatar's future position to the geometry center of each cluster, an algorithm is proposed to repartition non-assigned avatars among servers so as to reduce view inconsistency and preserve workload balancing among the servers. We conducted comprehensive experiments to evaluate our method by comparing it with other approaches.