A genetic algorithm with parallel steady-state reproduction

Abstract

During the last years, heterogeneous computing became more and more popular. Its advantage is availability, since any network of workstations can be used as a parallel virtual machine. On the other hand, programming these machines efficiently is even more difficult than programming a parallel dedicated machine. Each machine in the network may dynamically change performance, so optimal load balancing is the main challenge. One also has to deal with reliability problems on the network. With these two points in mind, we designed a parallel genetic algorithm featuring a new reproduction operator, called parallel steady-state reproduction. This operator adapts to a heterogeneous and unpredictable environment by allowing some “chaotic” behaviour. This parallel genetic algorithm revealed to be robust and efficient, and is particularly easy to implement.