Fit to programme

This task has been identified by the working groups as part of the agenda behind WP 1.1.

The task number is 006.

Description

CASTEP is one all-time classic in HPC benchmarking and one of the workhorses on UK supercomputers. The goal of this task is to take CASTEP and to curate an example run that allows an independent team to assess the code’s performance. On the long term, we’d expect any team submitting codes to SHAREing to provide us with such an example run before we start. This will give us a repository of codes to play around with when we study performance assessment approaches.

However, we initially have to create a few examples ourselves, so people know what they need to deliver when they submit their code, but also to enable us to apply our performance analysis methodologies to codes of actual relevance.

Obviously, the team taking up this task might want to go beyond the mere prep of benchmarking and actually do some benchmarking and analysis themselves. However, the core delivery of this task has to be a documentation of best practices on our webpages plus a working example.

Outcomes

  • Brief experience report on the SHAREing webpages.
  • Recommendation checklist how people can prepare a benchmark before they actually hand it over to someone else for a performance analysis.
  • Package of CASTEP which we can use to

The recommendation checklist should be integrated into the performance assessment workbook that we curate on SHAREing’s webpage as living document. However, it also should be distilled into a separate checklist that people can download/open on SHAREing’s webpage and then work through.