Implementation and Optimization of Task Offloading in Heterogeneous Clusters using OpenMP Cluster

• Grant number: 24/04232-8
• Support Opportunities: Scholarships in Brazil - Master
• Start date: September 01, 2024
• End date: February 28, 2026
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Principal Investigator: Hervé Yviquel
• Grantee: Jhonatan Cléto
• Host Institution: Instituto de Computação (IC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
• Associated research grant: 19/26702-8 - Trends on high performance computing, from resource management to new computer architectures, AP.TEM

Abstract

The OpenMP Cluster (OMPC) is a parallel and distributed programming model that enables the execution of OpenMP tasks on remote nodes of a High Performance Computing (HPC) cluster. With OMPC, it is possible to develop applications capable of scaling from a single machine to a robust supercomputer only using OpenMP directives. This characteristic makes OMPC a promising alternative to combining different parallel and distributed programming models and SDKs, a common practice for HPC application development. However, OMPC applications currently cannot offload tasks to available hardware accelerators in the cluster, such as GPUs on a node, even though OpenMP supports such operation. Therefore, the objective of this research project is to implement and optimize support for task offloading to accelerators in OMPC. With the inclusion of this new functionality in OMPC, we provide a development tool that, solely using the OpenMP syntax, can transfer tasks to any accelerator present in a heterogeneous HPC system, maintaining application portability across different hardware architectures.