How network topology affects dynamic loading balancing

Abstract
Previous research has proposed several, different load balancing strategies and measured their performances on either a distributed system or a multiprocessor network of specific topology. This paper addresses the performances of five dynamic load balancing strategies. A simulator has been developed to allow performance comparisons to be made across a range of network topologies: a 2D-mesh, a 4D-hypercube, a linear array, and a composite Fibonacci cube. The Fibonacci cube is one of the more recently proposed novel interconnection topologies. Results show that the performance of a dynamic load balancing strategy is affected by both the average processor distance and average node degree of the network.

This publication has 10 references indexed in Scilit: