A Load balancing algorithm for distributed computing systems

Abstract

In this thesis we propose a new model of load balancing for distributed computing systems. The model is based on a phase concept of a distributed application which is defined as "a number of application modules execute in parallel and communicate in parallel." The algorithm developed consists of graph reduction, initial assignment of modules and pairwise exchange techniques. It has been tested on a number of distributed applications and distributed systems generated randomly. The test results showed that the algorithm is successful in achieving suboptimal solutions compared to the optimal solutions. In addition, two enhancements to the algorithm have been incorporated. The first enhancement reduces complexity of the algorithm. The second enhancement improves objective function by means of module reassignment.