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

This paper is focused on the methodology for using the paral-lel multi-objective Extremal Optimization in load balancingalgorithms for distributed systems. In the proposed approach,parallel multi-objective Extremal Optimization algorithmsdefine task migration as a means for processor load balanc-ing. In the studied algorithms three objectives relevant todistributed processor load balancing are used as global fitnessfunctions: the function dealing with the computational loadimbalance in execution of application tasks on processors, thefunction concerned with the communication between tasksplaced on distributed computing nodes and the function con-cerned with the task migration number. Internal propertiesof the proposed multi-objective Extremal Optimization al-gorithms have been discussed. A number of such algorithmswith different composition of global and local fitness functionshave been presented and verified by simulation experiments.The performed comparative experiments concerned execu-tion of distributed programs represented as macro data flowgraphs. Their parallel execution speed-up was discussed basedon different best solution search methods such as compromiseapproach, lexicographic approach and hybrid approach. Theobtained results have shown that the parallel multi-objectiveExtremal Optimization algorithms used in load balancinghave visibly improved the quality of execution of the testedprogram graphs.