Genetic Algorithms On Distributed Systems

Abstract

A distributed implementation of Genetic Algorithms (GA) to solve Job Shop Scheduling problems is discussed The initial part of the paper introduces the highlights of GA and their application to Job Shop Scheduling problems A few words are also spent in describing the PVM (Parallel Virtual Machine) system that was adopted to implement the distributed application on a workstation network. The main discussion of the paper is devoted to the solution of the problems encountered in the GA implementation. The strategies adopted for improving the performance of our solution, both in terms of goodness of the numerical results and in terms of parallelism efficiency, are also discussed. As far as the distributed solution is concerned, the paper reports a set of tests to evaluate the relative performance indexes of the used workstations To this end, the response times of the whole distributed application on each individual machine, when varying the number of the parallel searches, are considered. Finally, the results of the distributed searches, with the co-operating processes distributed among all the available workstations, are presented. These results are compared with those of the previous calculations to give a performance evaluation of parallelism in terms of relative speedup and efficiency.