Exploiting multi-level parallel metaheuristics and heterogeneous computing to boost phylogenetics

Highlights

• We study heterogeneous multi-level optimizers to address complex phylogeny analyses.

• Accurate parallel strategies at the algorithm/iteration/solution levels are proposed.

• Configuration profiles are identified to attain performance and multiobjective goals.

• Performance-memory tradeoffs are revealed for two heterogeneous interaction schemes.

• Speedups of 396x and relevant energy savings are achieved, preserving solution quality.

Abstract

Optimization problems are becoming increasingly difficult challenges as a result of the definition of more realistic formulations and the availability of larger input data. Fortunately, the computing capabilities of state-of-the-art heterogeneous systems represent an opportunity to deal with the main complexity factors of these problems. These platforms open the door to the definition of robust metaheuristic solvers, in which parallel computations of different nature can be efficiently mapped to the most suitable architectures and hardware resources. This work investigates the combination of multi-level parallelism and heterogeneous computing to address an important multiobjective problem in bioinformatics: phylogenetics. A parallel metaheuristic approach, based on the joint exploitation of parallel tasks at the algorithm, iteration, and solution levels, is proposed to tackle computationally intensive inferences on CPU+GPU systems. Different heterogeneous design alternatives are also discussed, in accordance with the way the interactions between CPU and GPU are handled. The experimental evaluation of the proposal on real-world biological datasets points out the benefits of using multi-level, heterogeneous strategies, reporting accelerations up to 396$\times$ over the baseline metaheuristic as well as significant energy savings with regard to other parallel approaches, without impacting multiobjective solution quality.