This paper presents an adaptive dynamic load balancing scheme employing a genetic algorithm which includes an evaluation mechanism of fitness values in stochastic environments. A sender-initiative task migration algorithm continues to send unnecessary requests for a task migration while the system load is heavy, which brings much overhead before the migration finishes. In a genetic-based dynamic load balancing scheme we propose, a small subset of computers to which the requests are sent off is adaptively determined by a learning procedure to reduce unnecessary requests. The learning procedure consists of stochastic learning automata and genetic operators applied to a population of strings each of which stands for a subset of computers to which task migration requests are sent off. We implement the proposed algorithm on an actual distributed system which consists of UNIX workstations. We show the effectiveness of our approach through empirical investigations on the distributed system.
展开▼
