Increasing efficiency of task parallelism in GPU clusters for scientific workloads

• Grant number: 20/08475-1
• Support Opportunities: Scholarships in Brazil - Doctorate
• Start date: August 01, 2020
• End date: February 29, 2024
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Principal Investigator: Guido Costa Souza de Araújo
• Grantee: Gustavo Leite
• Host Institution: Instituto de Computação (IC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
• Associated research grant: 13/08293-7 - CCES - Center for Computational Engineering and Sciences, AP.CEPID

Abstract

The task-parallel programming paradigm is more robust than its data-parallel counterpart because it permits the programmer to express himself more freely. However, using task parallelism incurs new challenges of managing dependencies between tasks, synchronization and scheduling. For this reason, in this work we look to identify, characterize and optimize bottlenecks in the LLVM´s implementation of the OpenMP tasking runtime. Besides, we look forward to use a scientific application from the field of molecular dynamics, called HPCCS, to guide our study. Finally, we intend to deploy HPCCS on top of our runtime running on a cluster of GPUs such that the execution can be carried in the most efficient way possible, all the while requiring minimal changes in the original application. (AU)