Comparing Parallel Algorithms for Van der Waals Energy with Cell-List Technique for Protein Structure Prediction

• Daniel R. F. Bonetti
• Gesiel Rios Lopes
• Alexandre C. B. Delbem
• Paulo S. L. Souza
• Kalinka C. Branco
• Gonzalo Travieso

DOI: https://doi.org/10.5753/wperformance.2018.3322

Resumo

This paper compares the runtime of three distinct parallel algorithms for the evaluation of an ab initio and full-atom approach based on GA and celllist technique, in order to minimize the van der Waals energy. The three parallel algorithms are developed in C and use one of these programming models: MPI, OpenMP or hybrid (MPI+OpenMP). Our preliminary results show that van der Waals Energy are executed faster and with better speedups when using hybrid and more flexible parallel algorithms to predict the structure of larger proteins. We also show that for small proteins the communication of MPI imposes a high overhead for the parallel execution and, thus the OpenMP presents a better relation cost x benefit in such cases.