Log In | Users | Register
COSMO Web
Edit | Attach | New | Raw | Diff | Print | | Tools
You are here: ICON » Older_meetings » ICONmeet20181129

ICON meeting November 29th, 2018

Participants:

Carlos Osuna (CO); Colombe Siegenthaler (CS); Guy de Morsier (GM); Katie Osterried (KO); Philippe Marti (PM); Sylvaine Ferrachat (SF); Valentin Clement (VC); Will Sawyer (WS); Xavier Lapillonne (XL);

Update of each participant:

* Colombe Siegenthaler (CS):
  • CS is able to compile and run the ICON-ENIAC code version on Piz Daint in the setup generated for Bernhard's PhD project.
  • CS did some benchmarking of the ICON code for a CSCS project proposal, and shows a scaling graph. WS notes that Cray uses only O1 optimization level for reproducibility reasons, and that different values of the nproma namelist variable can affect performance a lot. Optimization of the ICON code is a topic scheduled for discussion at the next infrastructure meeting. Additionally, you should check the exact compilation flags in the ICON-ENIAC version before using them for production runs.

* Xavier Lapillonne (XL):

  • XL works on ENIAC (cf. PM's report).

* Carlos Osuma (CO) and Guy de Morsier (GM):

  • CO and GM attended the ICON-LAM training course in Langen. They worked together with the ICON developers on getting ICON-LAM running with IFS forecast 2.3.0 data which is needed for MeteoSwiss operational forecasting. They experienced several issues with getting the code running with IFS, including problems with the weighted interpolation for the radial basis functions used for the triangular grid of ICON. The code is now running with the IFS input data.
  • CO and GM also report that the grid generation and 2 way nesting features seem to work well.

* Sylvaine Ferrachat (SF)

  • SF is involved in the ENIAC project with PM and XL. Due to other obligations, she is not able to currently work with ICON, but plans to begin by spring of next year.
  • SF explains that the new CSCS proposal includes the ICON-HAM code (ICON coupled with the HAMMOZ aerosol and chemistry modules), which is running on Mistral but is not yet running on Daint.

* Katie Osterried (KO)

  • KO reports that C2SM is still working on obtaining a license for it's members for ICON. The process is complicated by the need to retain the IP rights of Gridtools at CSCS. It's not clear how much longer the process will take.
  • KO and CS propose a scheme for hosting the various ICON versions (ICON-AES, ICON-NWP, ICON-COSMO, ICON-CLM, etc.) on Github. The details of the proposal can be found here. However, it is not yet clear from the ICON license discussions if C2SM will have permission to distribute any repositories besides the official ICON repository hosted at MPI.
  • KO attended the ICON project group meeting at the COSMO-CLM assembly. The CLM community has developed an ICON-CLM version based on the ICON-NWP code that works with restarts and has adapted a climatological testsuite to test the code. They plan to calibrate and evaluate the ICON-CLM version against observational data, but this effort probably will not be completed for several years.
  • KO, as the Extpar Source Code Administrator, has released version 5.0 of Extpar which is fully compatible with the ICON versions used at DWD and at MPI.

* Valentin Clement (VC):

  • VC is now able to parse pieces of the ICON code using the CLAW compiler, and is working on transforming the JSBACH soil model for gpu. WS has contact with Reinard, who is ready to help if needed.
  • VC and WS discuss that it may be difficult to incorporate the CLAW compiler into the current build mechanism of ICON. Valentin will contact Sergey from MPI to discuss this, as Sergey is developing a new build mechanism for ICON.

* Will Sawyer (WS)

  • WS reports about a PRACE proposal in development with the target goal to have ICON fully running on gpu by September 2019.
  • WS is now able to run and validate the ICON code with the dynamics running fully on gpu.
  • WS is now working on the advection.
  • WS is participating in the Dyamond project, which is aiming to run ICON globally at very high resolutions. They are currently running a 2.5km (90 vertical levels) global resolution run, and hope to go to 1.5 km resolution.
  • There is some interest in DKRZ for gpus, since they might get a gpu machine. WS is trying to get them help implementing YAXT on gpu.

* Philippe Marti (PM):

  • PM is now able to compile the ICON-ENIAC code on cpu with PGI 18.10 (new compiler version) and run it. It does not run yet on gpu because of deep copies.
  • PM is also now able to compile and run the AMIP test case on gpu (with no output), which means that it is now possible to set up a gpu test case on Buildbot and start putting the work already done porting the standalone physics codes into the official ICON version.

Summary of the discussions:

  • XL proposes having an ICON GPU Hackathon to get more people interested in using ICON on gpu. WS will propose it at the infrastructure meeting.
  • There should be an ICON Developer's meeting in the winter at some point, but there has not been a date released yet. (Note: we have since been informed that the ICON developer meeting will be during the week from 11.2.2019 to 15.2.2019.)
  • Report from the last infrastructure meeting (attended by WS and XL):
    • False positives have been fixed in Buildbot, however now several tests are failing.
    • No one person has responsibility for Buildbot.
  • WS and PM and KO will sit together to look at the Buildbot structure and try to implement a new gpu test there.
  • KO will begin in January to adapt the technical testsuite of COSMO for use with ICON.

Attach
pdf
version 1 uploaded by KatherineOsterried on 10 Dec 2018 - 11:16
pdf
version 1 uploaded by ColombeSiegenthaler on 11 Dec 2018 - 09:35
spacer
This site is managed by the Center for Climate Systems Modeling (C2SM).
This site is powered by FoswikiCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors. Ideas, requests, problems? Send feedback!