Genetic Adventures in Parallel: Towards a Good Island Model under PVM

Keith Vertanen

Oregon State University, 1998.

Genetic algorithms (GAs) have proved to be a very useful and flexible way to solve difficult combinatoric problems. Arriving at high quality solution however often involves a very large number of evaluations and consequentially is quite computationally demanding. Evaluating GAs in parallel is thus desirable, but specialty parallel computers are not available to many who might benefit from parallel GAs. In this paper we will seek a parallel GA implementation under the PVM (parallel virtual machine) environment. After looking at the various parallel models available, we will choose the island model as the most appropriate for use with PVM. From a simplistic and inefficient starting implementation, we will develop a better implementation that provides close to optimal speedup.