Attendies:
- ACH VTK team: F. Granberg, J. Nurminen, K. Heljanko, I. Maarala
- ACH EPFL team: N. Varini, M. Peybernes
- TSVV team: P. Tamain, H. Bufferand, P. Ricci, M. Wiesenberger, A. Stegmeir, A. Nielsen
Agenda:
- Code developers: round-table review of the needs of the codes in terms of porting to modern development tolls
- ACH VTK: introduction to ACH team, views on best practices and best way of deploying them
- Open discussion on priorities and near term actions
Needs of codes:
Comment from ACH team: there is a solution (from AI) for archiving / labelling of simulations, could be investigated
Question from ACH team: should we try to develop common solution for all EUROfusion codes => PT: yes if possible
- GBS (P. Ricci):
- already on Gitlab
- need help with setting-up CI/CD
- maybe with testing / building dependencies
- FELTOR (M. Wiessenberger):
- on public Github repo
- automated testing not set-up due to hardware limitations (needs to be tested on CPU, OpenMP, GPU, MPI+OpenMP, MPI+GPU) => idea on how to do that?
- have not only unit-tests but also unit-benchmarks to test not only that the code
is sound but also performs as expected => everyone agrees it would be a nice additional feature
Comment from ACH team: could be done at Finish super-computing center (CSC)
- GRILLIX (A. Stegmeir):
- gitlab repo at MPCDF
- essentially already have a CI/CD environment (unit-tests with pfunit4)
- start to work with GPUs so raises the problem of multi-platforms
- suggests to use backbone of environment used by Grillix (like we did for the EBC)
- new aspect is unit-benchmarks, very interesting to analyse best practises in industry for this
ACH team presentation (F. Granberg)
Team composed of Fredric (PI), Jukka Nurminen (Prof of computer science), Keijo Heljanko (Prof of computer science), Ilari Maarala (PhD) => Ilari is the one who will actually be doing the work
Generic question whether weshould adopt fully new tools or be conservative and capitalize and share
already used best practices? => rather avoid exotic tools and adopt approaches already used in the community unless strong incentive
Discussion (all)
- Portability / compilation
- one option = containers, but very very complex to set-up
- using cmake is probably a lower hanging solution
Action (interested developers): each code needing help on this should get in touch with Ilari to set-up access to source code and launch work
- CI/CD environment, for unit-tests why not unit-benchmarks, could be built upon skeleton of what is used for GRILLIX (similarly to what was done for the EBC)
Action (Ilari M.): get in touch with Andreas S. to extract skeleton of CI/CD used for GRILLIX and paste it in other codes repositories if wanted
Action (ACH team): survey best practices in industry and evaluate pros and cons compared with what is used in GRILLIX. Report on recommendations (adopt or not new tool?) at later meeting.
- Simulations labelling and archiving:
- useful but probably difficult and complex to set-up
- KH: this is common practise in genomics and AI, we could survey what they do in these fields
Action (ACH team): survey possible solutions (as well as possible needs for other EUROfusion codes) and re-discuss in a few months
Next meeting
A priori at first task 9 regular meeting, but a dedicated meeting could be called before if deemed necessary.
There are minutes attached to this event.
Show them.