Genetic algorithms were developed during the 1960's by John Holland and are based on the theory of Darwinian evolution. A large number of potential solutions to the problem, each of which is represented as a chromosome, are randomly created to form a population. The solutions are then tested to see how well they solve the problem, and their fitness is rated. The fittest solutions are combined to form new solutions and mutated to seek better solutions. The fitness of the new generation is then evaluated, and goes through the same process of selection. Eventually, the solutions meet some predetermined end condition and the algorithm stops the evolutionary process.

The nature of genetic algorithms lends itself to parallelization and subsequent compilation in a clustered computing system. We propose to explore the relatively simple problem of weather prediction, specifically average daily air temperatures, in order to learn to program for a cluster environment, create genetic algorithms to run on the cluster, and explore the merits of different approaches in designing parallel genetic algorithms.