HPC Paramal Malleability Traceability

Abstract

This project aims to develop a set of software tools to generate and analyze the execution of high-performance computing (HPC) malleable applications. The software consists of two applications. The first application, the DMRTRACE is a library for recording traces in CSV format. It is designed to be integrated into the Barcelona Supercomputing Center’s (BSC) malleability library, the DMR. The second application, DMRTRACEPARSER, is a utility to convert the CSV trace files to PRV trace files, to analyse traces using Paraver, the standard analysis tool in the BSC for HPC performance analysis. The first part of this document explains the research and development process of the trace analysis tools. The second part is a step-by-step guide about how to use the developed tools to trace and analyze several HPC malleable applications. The software tools have been developed to be used in the BSC. So that they have been developed using BSC standard development tools, procedures, and methodologies. The applications developed are open source, programmed in C++, and tested and validated in the BSC’s supercomputing infrastructure, the MareNosendendtrum5. The source code is placed in the BSC’s GitLab repository.