Expanding Service Capacities and Increasing Service Reliabilities for the Grid-Based Utility Computing

A non-polynomial (NP)-hard combinatorial optimization problem that is associated with expanding service capacities and increasing service reliability in grid-based utility computing is investigated in this paper. The considered problem is decomposed into master and slave subproblems, with theoretical justification, and a computationally efficient two-level iterative method that is used in solving it is proposed. To solve the slave subproblem, an ordinal optimization-based n-stage method, associated with an approximate model for objective value evaluation, is employed. To solve the master subproblem, a bisection method is used. Under some conditions, the solution obtained using the proposed iterative two-level method is optimal. The validity of the proposed method is tested by ten cases on a 12-node 17-link computing grid. Five of the ten cases are randomly selected, and the solutions that are obtained in these cases are optimal, rather than “good enough.” The average CPU time required by the proposed method in obtaining the optimal solution of the considered NP-hard combinatorial optimization problem is 2.278 h, when executed using a Pentium IV PC with a 2-GB RAM. Additionally, the computational efficiency of the proposed method greatly exceeds a genetic algorithm with an exact model.