Applications of Parallel Platforms and Models in Evolutionary Multi-Objective Optimization

Parallel Multiobjective Metaheuristics for Inferring Phylogenies on Multicore Clusters

“…We will design mixed mode implementations based on the industry standards OpenMP/MPI, a widely-used strategy to parallelize scientific applications on SMP clusters [27]. For this purpose, we will study master-worker implementations, as this model represents one of the most efficient choices to tackle optimization problems with compute-expensive objective functions [3].…”

“…This fact gave rise to the application of multiobjective optimization (MOO) to bioinformatics, with the aim of finding a set of Pareto solutions which optimize simultaneously two or more objective functions [2]. This formulation brings about new issues due to the additional complexity required to resolve a multiobjective optimization problem (MOP), mostly motivated by the high processing times required to find and evaluate solutions according to multiple criteria [3]. The availability of cluster architectures composed of multicore symmetric multiprocessors (SMP) represents an opportunity to address such compute-intensive problems.…”

Parallel Multi-Objective Evolutionary Algorithms: A Comprehensive Survey

“…Throughtout the years, several pMOEAs have been proposed (see [14] and [15] for a condensed list of approaches), aiming to advance the state-of-the-art or to tackle a specific problem. However, a constant factor has been the use of the four main types of parallel models: master-slave, island, diffusion, and hybrid models [16,13] (see Figure 1). The master-slave model is one the of the simplest ways to parallelize MOEAs.…”

“…To date, several tutorials and surveys on pMOEAs have been published, aiming to provide the basics on the matter and describe the most recent approaches [17,14,18,1,19,16,15,13]. In 2003, Van Veldhuizen et al [17] published a journal paper which not only gives useful guidelines for the design and implementation of pMOEAs, but also makes observations about parallel architectures, benchmarks, performance metrics, and estimations of their computational time.…”

“…Nevertheless, the review of approaches is exclusively focused on reporting their main features (according to the taxonomies) without a discussion of properties and performances. On the other hand, López Jaimes and Coello [16] presented a review of modern parallel platforms and the way in which they can be exploited to implement pMOEAs. Additionally, an explanation of the four parallel models and a review of five approaches from 2000 to 2007 is included.…”