Version 1 (modified by mocavero, 6 years ago) (diff) |
---|
NEMO HPC subgroup: Wed 1 Aug 2018
Attendees: Mike Bell (Met Office), Mirek Andrejczuk (Met Office), Matthew Glover (Met Office), Miguel Castrillo (BSC), Oriol Tinto (BSC), Erwan Raffin (Bull), David Guibert (Atos), Clement Bricaud (MERCATOR), Michael Lange (ECMWF), Peter Dueben (ECMWF), Nils Wedi (ECMWF), Marie-Alice Foujols (CNRS), Silvia Mocavero (CMCC)
Apologies: Martin Price, Eric Maisonnave, Marcin Chrust, Andrew Porter
1. NEMO profiling and tiling (MetO)
Maff presented the work done to apply the tiling strategy to the traldf_iso NEMO routine, one of the computationally intensive routines identified by profiling the NEMO code using DrHook?. Three tiling regions have been implemented within the routine and an improvement of ~27% on the elapsed time has been achieved. Caching and compute time have been profiled separately for one of the region. Tiling needs to be moved higher in the call tree, but a refactoring of the code is needed, also taking into account the presence of MPI calls (see attached slides)
Actions: the NEMO code equipped with DrHook? profiling calls and the tiling code will be shared with other groups (e.g. BULL) to be tested on other systems
2. Draft answering to "Ocean Model Performance” questions (Silvia)
Silvia drafted and summarised the answers to the questions and Nils sent some questions/remarks (see the attached file)
Actions: all to check the answers, taking into account Nils remarks. Contributions will be exchanged by email. A new version of the document is planned for the end of August.
3. Next meeting call
Next meeting will be in September. Silvia will send the doodle poll
Attachments (2)
- Ocean Model Performance.txt (7.0 KB) - added by trac 6 years ago.
- NEMO_profiling_and_tiling.pptx (1.2 MB) - added by trac 6 years ago.
Download all attachments as: .zip