Characteristics of Models of Parallel Genetic Algorithms on PC Cluster System

In this paper, the characteristics of the typical two models of parallel genetic algorithms; the coarse grained model and the fine grained model, are compared. Especially, the parallel efficiency on PC clusters that have no more than 10 CPUs is discussed. The cluster used in this study has two kinds of network architectures; FastEthernet and Myrinet. Through the numerical examples, these models are examined and discussed. The followings were made clarified. In the fine grained model, the master slave model, the ideal parallel efficiency is not 100%. It is concluded that the fine grained model is not suitable for this kind of cluster. On the other hand, in the coarse grained model, the population is divided into sub populations. The communication does not happen frequently. The coarse grained model is suited for this type of clusters. However, even in the coarse grained model, the data transfer schedule is needed because of the data stacking.