A Neural Network Approach for Dynamic Load Balancing In Homogeneous Distributed Systems

A novel neural-based solution to the problem of dynamic load balancing in homogeneous distributed systems is proposed. The winner-Take-All (WTA) neural network model is used for implementing the selection and location policies of a typical dynamic load balancing algorithm. Unlike most of the previous literature that assumed independent tasks, which is not always true, tasks with interprocess communication requirements are considered. All delays due to any usage of the communications network resource are taken into account. A simulation study was carried out to verify the effectiveness of the proposed approach, results were compared against the no load balancing case. Although performance improvements are dependent on the system overall load, load intensity per node, and nature of tasks, the results suggest that it is always beneficial to use load balancing than not at all.