Parallel late acceptance hill-climbing algorithm for the google machine reassignment problem

Abstract

Google Machine Reassignment Problem (GMRP) is an optimisation problem proposed at ROADEF/EURO challenge 2012. The task of GMRP is to allocate cloud computing resources by reassigning a set of services to a set of machines while not violating any constraints. We propose an evolutionary parallel late acceptance hill-climbing algorithm (P-LAHC) for GMRP in this study. The aim is to improve the efficiency of search by escaping local optima. Our P-LAHC method involves multiple search processes. It utilises a population of solutions instead of a single solution. Each solution corresponds to one LAHC process. These processes work in parallel to improve the overall search outcome. These LAHC processes start with different initial individuals and follow distinct search paths. That reduces the chance of falling into a same local optima. In addition, mutation operators will apply when the search becomes stagnated for a certain period of time. This further reduces the chance of being trapped by a local optima. Our results on GMRP instances show that the proposed P-LAHC performed better than single threaded LAHC. Furthermore P-LAHC can outperform or at least be comparable to the state-of-the-art methods from the literature, indicating that P-LAHC is an effective search algorithm.