Comparative Study of a New Problem Decomposition Method for Solving Global Optimization Problems on Loosely Coupled Systems

Loosely coupled computing systems is an emerging class of parallel computing systems. They are capable of solving large computationally expensive problems at a relatively low cost. During the computational process one or more computing nodes can be turned off resulting into loss of data. In global optimization problems this loss of data can lead not only to increasing the computation time but also to decreasing the solution quality. This paper presents a new problem decomposition method for loosely coupled systems that splits the search domain into multiply connected subdomains. Such an approach allows minimizing the negative impact of node termination. Results of the comparative experimental investigation with a use of benchmark functions are presented in this paper which demonstrate the increase in solution quality comparing to the traditional decomposition methods.