Parallel Evolutionary Algorithms on Message-Passing Clusters

Parallel Evolutionary algorithms have been developed to reduce the running time of serial Evolutionary algorithms. Two major paradigms for parallel programming, Message Passing and Shared Memory, are implemented and their performance observed. Message Passing Interface (MPI) and TreadMarks runtime libraries are chosen to implement parallel Evolutionary algorithms, based on a synchronous master-slave model, on a cluster of workstations. We then modify some parameters of the algorithms to observe their effects on the performance. Our objective is to show that, despite many believes that Message Passing scheme should give a better performance, Shared Memory scheme results in a similar performance in some conditions and thus can be considered as an excellent alternative for those who want to implement parallel Evolutionary algorithms.