Waffle Mixer: A Library for Distributing Genetic Algorithms

We describe a toolkit that enables simple parallelization of genetic algorithms. Genetic algorithms are favored for optimization problems because they lend themselves to both continuous and discrete combinatorial problems and are less susceptible to becoming trapped at local optima in comparison to gradient search methods. Unfortunately, GAs are typically computationally expensive. The toolkit enables writers of genetic algorithm applications to choose between local (one machine) operation or distributed computation at runtime. The distributed approach achieves the same result or better than a local computation, but significantly diminishes the required computation time. An analysis of the system’s performance is described.