Comparative study of parallel vs. distributed genetic algorithm implementation for ATM networking environment

The objectives of this effort are: (1) to provide a faster (parallel) implementation to a previously implemented ATM network design simulator, and (2) to compare two different parallel and distributed computing (PDC) architectures in order to select one to be the architecture for this new parallel version. To improve the performance of the previously implemented GA, two parallel versions are developed using two different approaches. The first uses message passing interface (MPI) function calls embedded in C++ programs on a network of workstations. The second version is developed for a multiprocessor system using a C/C++ parallel compiler. This paper describes the approaches used to develop the parallel versions, the design of parallel applications and the results of all the experiments that were done to study the effect of all the design parameters on system performance