Design of a parallel genetic algorithm for the Internet

This paper proposes a parallel implementation of the genetic algorithm (GA) on the Internet which will improve the algorithm's performance. It is motivated by the possibility of aiding research into complex search and optimization problems that use the GA. Requirements and constraints regarding parallelization of the GA are identified. A parallel GA is developed for an ideal PRAM architecture and is shown to have an asymptotic running time of O(log n), an improvement over the sequential GA. A parallel GA is also designed for a Unix network and has an asymptotic running time comparable to the ideal system. The algorithm is a decentralized, asynchronous, and fault-tolerant design that matches the characteristics of the network. The GA population is divided into colonies that are distributed among processors. Trade policies are executed for the exchange of genes.