Comparisons of Load Balancing Strategies for Ray Tracing on Network Clustered Environments

Abstract

Load balancing has been a major issue in network computing for ray tracing and other applications. In this paper we present a new load balancing strategy, called global distributed control to load balancing in the network processors. The algorithm not only has the ability to dynamically adjust to the load, but also has the fault tolerance ability. In order to study the performance of the algorithm we design a common experimental protocol to evaluate the performance. The parameters used in the protocol include environment configuration, problem size, data type, number of workstations, and algorithm selection. We compare the algorithm with the traditional load balancing schemes such as a master slave, and interleaved, and a proposed hybrid algorithms. Experimental results show that the GDC performs the best among four algorithms.