DLBEM: A Predictive Approach to Dynamic Load Balancing

In this paper we propose a predictive approach for dynamic load balancing. This approach involves predicting unbalanced load distribution using Estimation Maximization (EM) algorithm, and migrating local jobs based on the estimation average. This strategy is an improvement to the existing approaches because by using EM algorithm, we only extract sample workload information which is assumed to follow mixture Gaussian distribution without keeping track of the whole workload information in the system. This method can give relatively accurate results and greatly reduces the communication efforts in distributed systems. After getting the estimation average load, a local job migration policy is invoked to balance the load distribution without oscillation. Simulation results show that our approach can get load distribution information Expectation Maximization Algorithm, Dynamic Load Balancing, Parallel Computing