A Reversible Debugger for MPI Applications
Year: 2024
Type: conference-paper
Abstract: Cluster machines are gaining importance, for example in high-performance computing and eScience. For this, programs need to be parallelized and run with appropriate tools, typically MPI (Message Passing Interface) in a scientific context. Since writing programs for parallel computation is significantly more difficult than programming for sequential execution, debugging tools, which are considered a necessary part of the toolset of software developers, are of even higher importance there. Reversibility, providing the ability to progress backwards in the program execution in some form, has been added to some debuggers and is a useful feature for debugging MPI applications as well. This paper presents a debugger for MPI applications which offers reversible debugging commands. This is done using a checkpoint-restore mechanism. We demonstrate the viability of this approach to enable reversible debugging for parallel computation.
Institution: University of Tartu
Language: English
Subfield: Software
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4402527248