Parallel genetic programming: component object-based distributed collaborative approach

We discuss the feasibility of applying the distributed collaborative approach for improving the computational performance of genetic programming (GP), implemented on cost-efficient clusters or the Internet. The proposed approach exploits the coarse grained inherent parallelism in GP among relatively autonomous subpopulations. The developed architecture of a distributed collaborative parallel GP (DCPGP) features a single, global migration broker and centralized manager of the semi-isolated subpopulations, which contribute to quick propagation of the globally fittest individuals among the subpopulations; this reduces the performance demands on the underlying communication network, and achieves dynamic scaling-up features. DCPGP exploits the distributed component object model (DCOM) as a communication paradigm, which as a true system model offers generic support for the issues of naming, locating and security of communicating entities of the developed architecture. Experimentally obtained speedup results show that close to linear speedup characteristics of the prototype of DCPGP are achieved on a network of 8 workstations