Dynamic mapping and load balancing with parallel genetic algorithms

The paper presents an approach to dynamic mapping and load balancing of parallel programs in MIMD multicomputers, based on coordinated migration of processes of a parallel program. A program graph is interpreted as a multi-agent system with locally defined goals and actions, operating in some environment. A parallel genetic algorithm (island model) is developed to work out a set of collective decisions concerning processes' migration. Presented experiments show a behavior of the algorithm.<
>