POP contributed to a Performance Analysis Workshop Series organised by the DiRAC consortium and Durham University in England over 5 weeks in April and May 2024.  The full-day training sessions addressed core-level, intra-node and inter-node runtime efficiency of MPI and OpenMP/multithreaded application codes, including presentations of tools from UVSQ, JSC and others each.

Time was allocated in each session for participants to apply the tools to their own application codes on the local DINE cluster.

The series started with an introduction to the POP methodology by Marta Garcia-Gasulla of BSC, and concluded with sessions on performance regression testing and participants summarising lessons learned. It followed similar events from previous years, complementing preceding series' of workshops on "Debugging, Testing and Correctness" and "GPU Programming" primarily addressed to Research Software Engineers (RSEs) and simulation code developers.  There were again over 30 participants for this hybrid training event, mostly in teams working on the same application code, and from as far away as Colombia and Indonesia.

 

In the wrap-up session, each of the presenters reported benefits from participation and use of one or more tools, along with certain issues and limitations encountered.  While in-person presence and assistance was highly valued, a relatively low number of participants used this option. RSEs reported that the one day per week workshop format best suited their schedules, allowing time to work applying the tools between sessions, whereas others find dedicated full-week workshops preferable where there are fewer distractions and they can fully focus.