The parallel genetic algorithm (PGA) is a prototype of a new kind of a distributed algorithm. It is based on a parallel search by individuals all of which have the complete problem description. The information exchange between the individuals is done by simulating biological principles of evolution. The PGA is totally asynchronous, running with maximal efficiency on MIMD parallel computers. We compare the PGA with the breeder genetic algorithm (BGA) based on methods used by breeders of livestock. The BGA is more efficient, but less parallel. We will show the power of the PGA with one application-the graph partitioning problem.