Princeton Research Computing (PRC) acknowledged the input of a UQ Research Computing Centre (RCC) staffer to open high-performance computing tooling in a talk at an advanced research computing conference in July.
Joshua Maloney – a Research Computing Systems Administrator at RCC at the time of his contribution, who has recently left The University of Queensland following the end of his contract – said he was surprised but grateful to be acknowledged for a small code improvement he made to Jobstats, a HPC tool developed by PRC as part of an open-source project on software developer platform GitHub.
“This is a nice reminder that contributions to open HPC tooling – even small ones – get noticed by the wider community. I'm looking forward to finding more ways to contribute,” said Joshua.
“I was surprised that they [PRC] picked up on such a small development, but these things are expensive and every 1% matters.”
RCC Director Jake Carroll said: “It encourages me to see RCC contributing to global supercomputing projects. such as these. UQ and Princeton have separately built up a significant recent history in research data storage architecture collaboration.”
PRC is the supercomputing centre of private Ivy League research university Princeton University, based in New Jersey, USA.
PRC acknowledged Joshua in a talk titled "Getting the Most Out of Your GPUs", presented by PRC Research Software and Computing Training Lead Jonathan Halverson at the ACM Practice and Experience in Advanced Research Computing (PEARC) conference held in Minneapolis, USA, from 26-30 July.
Other acknowledgements included US universities Yale, Harvard, Brown, Princeton, Iowa State, and the University of Illinois Urbana-Champaign, as well as Germany’s Leibniz University Hannover and Max Planck Institute for Neurobiology of Behavior, and technology company NVIDIA.
The PRC talk covered Jobstats and Job Defense Shield, two PRC-released HPC tools involving auto-cancelling idle GPU jobs*, GPU fractionalisation with Multi-Instance GPU (MIG), and detailed utilisation metrics for HPC clusters.
“The two tools are a great example of the kind of practical, cross-institution tooling that keeps research computing centres running efficiently,” said Joshua.
Jobstats is a free, open-source job monitoring platform designed for CPU and GPU clusters that use the Slurm workload manager.
“I made a small code improvement to make the tool more efficient, convenient and maintainable,” said Joshua. “This is intended to help future authors improve the software, for example by integrating AMD GPU support (which we currently have in the RCC branch of Jobstats). In general, this work is monitoring GPU activity and reducing idle time.”
Job Defense Shield is a software tool for identifying and reducing instances of underutilisation by HPC users.
University of Queensland HPC Bunya uses RCC tool Jobcop to do automatic job cancelling instead of Job Defence Shield. Removing idle jobs frees up resources for jobs actively making the most of the significant investment in Bunya.
One of Joshua’s tasks at RCC was to assist in building tools to monitor and reduce GPU idle time on Bunya.
*An HPC "job" is a request you submit to a shared supercomputer asking it to run your program using a certain amount of computing power for a certain amount of time, and then you wait in a queue until that power becomes available.
