Green Parallel Metaheuristics: Design, Implementation, and Evaluation

Abstract

Green parallel metaheuristics (GPM) is a new concept we want to introduce in this thesis. It is an idea inspired by two facts: (i) parallel metaheuristics could help as unique tools to solve optimization problems in energy savings applications and sustainability, and (ii) these algorithms themselves run on multiprocessors, clusters, and grids of computers and then consume energy, so they need an energy analysis study for their different implementations over multiprocessors. The context for this thesis is to make a modern and competitive effort to extend the capability of present intelligent search optimization techniques. Analyzing the different sequential and parallel metaheuristics considering its energy consumption requires a deep investigation of the numerical performance, the execution time for efficient future designing to these algorithms. We present a study of the speed-up of the different parallel implementations over a different number of computing units. Moreover, we analyze and compare the energy consumption and numerical performance of the sequential/parallel algorithms and their components: a jump in the efficiency of the algorithms that would probably have a wide impact on the domains involved.