Exploiting multi-objective parallel extremal optimization features in dynamic load balancing

Abstract

This paper is focused on the methodology for using the parallel multi-objective Extremal Optimization in load balancing algorithms for distributed systems. In the proposed approach, parallel multi-objective Extremal Optimization algorithms define task migration as a means for processor load balancing. In the studied algorithms three objectives relevant to distributed processor load balancing are used as global fitness functions: the function dealing with the computational load imbalance in execution of application tasks on processors, the function concerned with the communication between tasks placed on distributed computing nodes and the function concerned with the task migration number. Internal properties of the proposed multi-objective Extremal Optimization algorithms have been discussed. A number of such algorithms with different composition of global and local fitness functions have been presented and verified by simulation experiments. The performed comparative experiments concerned execution of distributed programs represented as macro data flow graphs. Their parallel execution speed-up was discussed based on different best solution search methods such as compromise approach, lexicographic approach and hybrid approach. The obtained results have shown that the parallel multi-objective Extremal Optimization algorithms used in load balancing have visibly improved the quality of execution of the tested program graphs.