Evolutionary Algorithms A Tool for 0 ization with
Jiirgen Wakunda, Andreas Zell · 1997
that there can be several relatively isolated populations (island model) matches the granularity of parallelization on a MIMD computer. 2. overview of E~A We describe the EVA sofiare package which consists of parallel (and sequential) implementations of genetic algorithms (GAS) and evolution strategies (ESs) and a common graphical user interface. We concentrate on the descriptions of the two distributed implementations of GAS and ESs which are of most interest for the future. We present comparisons of different kinds of genetic algorithms and evolution strategies that include implementations of distributed algorithms on the Intel Paragon, a large MIMD computel; and massively parallel algorithms on a 16384 processor MasPar MP-I, a large SIMD computel: The results show that parallelization of evolution strategies not only achieves a speedup in execution time of the algorithm, but also a higher probability of convergence and an increase of quality of the achieved solutions. In the benchmark functions we tested, the distributed ESs have a better performance than the distributed GAS.