Architecture and enhanced-algorithms to manage servers-processes into network: a management system

Abstract

This work investigates minimizing the makespan of multiple servers in the case of identical parallel processors. In the case of executing multiple tasks through several servers and each server has a fixed number of processors. The processors are generally composed of two processors (core duo) or four processors (quad). The meaningful format of the number of processors is 2k, and k ≥ 0. The problem is to find a schedule that minimizes the makespan on 2k processors. This problem is identified as NP-hard one. A new network architecture is proposed based on the addition of server management. In addition, two novel algorithms are proposed to solve the addressed scheduling problems. The proposed algorithms are based on the decomposition of the main problem in several sub-problems that are applied to develop new heuristics. In each level of the generated tree, some results are saved and used to decompose the set of processes into subsets for the next level. The proposed methods are experimentally examined showing that the running time of the proposed heuristics is remarkably better than its best rival from the literature. The application of this method is devoted to the network case when there are several servers to be exploited. The experimental results show that in 87.9% of total instances, the most loaded and least loaded subset-sum heuristic (MLS) reaches the best solution. The best-proposed heuristic reaches in 87.4% of cases the optimal solution in an average time of 0.002 s compared with the best of the literature which reaches a solution in an average time of 1.307 s.