A P2P genetic algorithm environment for the internet

Abstract

A peer-to-peer (P2P) based genetic algorithm (GA) execution environment that makes use of the computational resources distributed over the Internet for executing time-consuming GA applications is discussed. Object oriented techniques are used as the basis of addressing the flexible GA application programming interface and user customization. As the number of demes is increased from one to 15, the performance is improved and the percentage of correct partitions is increased form 50% to 83% for a 4-bit deceptive function. The simulation shows that a good speedup for the PTP GA system can be achieved if the execution time for fitness evaluation is large and the communication latency among peers on the Internet is small.