New URL for NEMO forge!   http://forge.nemo-ocean.eu

Since March 2022 along with NEMO 4.2 release, the code development moved to a self-hosted GitLab.
This present forge is now archived and remained online for history.
WorkingGroups/HPC/Mins_2020_11_04 – NEMO
wiki:WorkingGroups/HPC/Mins_2020_11_04

Version 3 (modified by epico, 4 years ago) (diff)

--

NEMO HPC WG: Tue 04 Nov 2020

0. Attendees

Jesus Labarta (BSC), Oriol Tinto (BSC), Clement Bricaud (Mercator), Daley Calvert (Met Office), David Guilbert (Atos/Bull?), Francesca Mele (CMCC), Kristian Mogensen (ECMWF), Maff Glover (Met Office), Maicon Faria (BSC), Mario Acosta (BSC), Martin Shreiber (TUM), Michael Lange (ECMWF), Miguel Castrillo (BSC), Mirek Andrejczuk (Met Office), Nils Wedi (ECMWF), Sam Hatfield (ECMWF), Xavier Yepes (BSC), Italo Epicoco (CMCC).

1. Presentations

  • NEMO Computational Perfomance Analysis (Jesus Labarta - BSC)

(see attached slides)

2. Discussion

Computational performance analysis focused on the inner behaviour of the code and the use of the hardware resource. NEMO 4.0.2 has been analyzed in ORCA1 configuration. The main outcomes and suggestions are:

  • the north fold communication introduces a delay which propagate over the time loop
    • Merge sequence of waits for Isends to waitall
    • Reorder, if possibile, the Isend and Recv
    • Assign less load to north processes to take the internal communications out of the critical path
    • Investigate if reducing the number of processes at north could bring some benefit
  • Investigate the use of OpenMP task parallelism

Some comments arise during the discussion

  • a further investigation on the level of vectorization could be interesting
  • this anaysis could be executed again on NEMO 4.2 to evaluate the impact of the new optimizations that will be introduced in the next release and it can be extended also on configurations with SI3 and it could include also the evaluation of I/O operations.

Attachments (1)