What is it about?
We created an extension to GDB- a popular open-source debugger- allowing it to debug MPI programs from a single interface. To do this we added a mechanism based on the MPI standard to connect to the various processes even on multiple nodes, and additional commands to make debugging a collection of processes more manageable. We've used it to solve complex distributed systems bugs and show that it can reduce the time to detecting an "out of bounds access" error in a simple parallel program.
Featured Image
Photo by Taylor Vick on Unsplash
Why is it important?
Alternative methods of debugging MPI programs are either expensive or don't provide the same kinds of features.
Perspectives
A lot of HPC centers like the one where I earned my PhD didn't have access to tools for distributed debugging. Having a tool like MPIGDB provides a lot of power to understand distributed programs.
Robert Underwood
Argonne National Laboratory
Read the Original
This page is a summary of: MPIGDB: A Flexible Debugging Infrastructure for MPI Programs, August 2023, ACM (Association for Computing Machinery),
DOI: 10.1145/3589013.3596675.
You can read the full text:
Resources
Contributors
The following have contributed to this page
