The 2021 edition of the Platform for Advanced Scientific Computing (PASC) conference was held 5 to 9 July as a virtual digital event.
It featured 45 distinct minisymposia, several comprising multiple parts, including some carried over from the cancelled 2020 incarnation.
POP organised a two-part mini-symposium "Performance Optimisation and Productivity for EU HPC Centres of Excellence (and other European parallel application developers preparing for exascale)" where sectorial HPC Centres of Excellence were invited to report on their experience collaborating with POP and using POP services as part of preparing their flagship application codes for forthcoming exascale computer systems.
The first part on Wednesday 7 July opened with an introduction to POP and a review of best practice for efficient and scalable application performance by Marta Garcia-Gasulla (BSC), which was also the topic of a virtual poster. The outcome of performance evaluation workshops jointly organised with the Energy-oriented CoE (EoCoE) was then reported by Mathieu Haefele (Universite de Pau). This was followed by an in-depth look at the performance of the SeisSol code for earthquake simulations that was part of a campaign of assessments for the ChEESE CoE focused on Solid Earth, given by Ravil Dorozhinskii (TU Munich). Closing the first session, Peter Coveney (UCL) reviewed work that was done with two of the CompBioMed CoE for computational biomedicine, where the HemeLB and SCEMa simulation codes are being prepared for exascale with the assistance of E-CAM and POP.
On Thursday 8 July the minisymposium continued with a second session. Ricard Borrell (BSC) summarised the collaboration with POP on the versatile Alya computational mechanics code used by six HPC Centres of Excellence. Sebastien Masson (Sorbonne Universite LOCEAN-IPSL) reviewed POP collaboration to improve execution efficiency of the NEMO ocean modelling framework that features in the ESiWACE weather and climate CoE. And in the final talk, Andrea Ferretti (CNR Nano) presented the work of the MaX CoE on materials design where electronic structure community codes have been ported to GPUs after POP performance audits.
A concluding panel discussion with the presenters from both parts discussed their HPC CoE and flagship application preparations for a range of potential exascale computer systems and remaining challenges.
While some significant progress has been achieved getting ready for systems with large numbers of Intel CPUs and Nvidia GPUs, forthcoming exascale systems with other processors and accelerator devices, with different compilers and parallelisation frameworks, require much more attention. It was also emphasised that file I/O is critical for production executions yet this aspect of parallel application performance remains inadequately analysed and addressed.