Algoritmo aleatorizado basado en distribuciones deslizantes para el problema de planificación en sistemas Grid

In this paper we present a randomized algorithm for the online version of the Job Shop problem where jobs are composed of processes with precedence constraints and processors are organized in a Grid topology. The proposed randomized algorithm is based on a new technique that we have denominated as sliding distributions, which aims at combining the advantages of the deterministic approximation algorithms with those of the Montecarlo randomized algorithms. The objective is to provide an algorithm that delivers ρ-approximated solutions with high probability, but at the same time, is able to investigate an extended neighborhood of such solutions so that it can escape from local extrema. We formally characterize the temporal complexity of the proposed algorithm and show that it is correct. We also evaluate the performance of the proposed algorithm by means of a series of simulation-based experiments. The results show that the proposed algorithm outperforms the traditional state of the art algorithms for scheduling in Grid systems. The performance metrics are average delay, maximum delay, and Grid utilization.