Log In | Users | Register
Edit | Attach | New | Raw | Diff | Print | | Tools
You are here: ICON » Meetings » ICONmeet20180530

ICON meeting May 30th, 2018


Bernhard Enz (BE); Colombe Siegenthaler (CS); Doris Folini (DF); Guy de Morsier (GDM); Katherine Osterried (KO); Philippe Marti (PM); Valentin Clément (VC); Xavier Lapillonne (XL); Dominik Brunner (DB); Thomas Köster (TK); Qing Mu (QM);

Update of each participant:

* Colombe Siegenthaler (CS):
  • The coupled ICON-HAM version is now running, and CS is working on testing and verifying it. MOZ is not yet part of the coupled version- this work is planned for the future (hopefully in a year or two).

* Guy de Morsier (GDM):

  • GDM is working on using the DWDtools program to convert IFS input files for use to drive ICON. It is currently not working when there is a combination of new grib1 and grib2 files. DWD is working on debugging this issue now.

* Thomas Köster (TK):

  • TK is finished with porting the ICON turbulence parameterization vdiffup/vdiffdown (from ECHAM physics) to GPU.
  • TK will work next on parsing the parameterization with the CLAW compiler and comparing to his porting.
  • TK is also working on evaluating the GPU speedup with different compilers and setups. However, there are some issues with getting the code to run with certain compilers on Piz Daint (particularly cray and pgi).

* Bernhard Enz (BE):

  • BE will use ICON-LAM with forcing at the boundaries.
  • BE tried to use the online grid generation tool to generate the model grid, but it only works with rectangular and circular grids. There is a grid generator released with the source code of ICON, but it does not work.
  • BE will next test rectangular grids to see if they are sufficient for his purposes.

* Katherine Osterried (KO):

  • KO reported that the C2SM ICON license is still in progress- recently MPI made some changes to the license, so it needs to be approved again by the ETH legal department.
  • KO is working on a major release of Extpar targeted for the summer which will work for COSMO and ICON (tested at DWD and MPI).

* Dominik Brunner (DB) and Qing Mu (QM):

  • DB and QM have a new ~2.5 year project to do simulations across several model resolutions.
  • QM will begin the simulations with COSMO-ART, and then hopefully switch to using ICON-ART if preliminary results are satisfactory.

* Xavier Lapillonne (XL):

  • MeteoSwiss still has a target date to go operational with ICON for weather forecasting by the end of 2023.

* Valentin Clément (VC):

  • VC is trying to parse the JSBACH part of ICON with the CLAW compiler (which is a source-to-source translator that inserts OpenACC and OpenMP code automatically)
  • JSBACH is now almost completely parsed by the CLAW compiler. Once this is completed, JSBACH can be ported to GPU using the CLAW compiler and tested.

* Philippe Marti (PM):

  • PM has been working on fixing the JSBACH and vdiff standalone models, which broke after the latest ICON update. This occurred in spite of the fact that the ICON releases are tested on Daint. The tests are not sufficient, however, as segmentation faults give false positives.

* Doris Folini : nothing to report

Summary of the discussions:

* CS asks: Which compilers are working on Daint currently with the latest version of ICON?

PM and TK give the following status for different compilers of CPU running:

Compiler Status Comment
PGI Gives segmentation faults. PGI is working on this- hopefully resolved by fall
Cray Works if you run with less than 6 cores. Not clear when a resolution will be given to run on more cores.
Intel Working  
GCC Working (currently the recommended compiler)  

* XL asks if anyone has a flowchart of the workflow of the ICON model. CS shows an example she has from a colleague at MPI. It is agreed that this can be used as a starting point, but needs to be adapted to include NWP physics as well.

* KO reports that the COSMO-CLM colleagues at DWD are interested in adapting the technical testsuite for use with ICON. This would also potentially address some of the issues with ICON releases not working on Daint because the current tests are giving false positives. It is agreed that XL and KO will arrange a meeting with the DWD colleagues to collaborate on this issue.

* DB asks if both physics packages (ECHAM and NWP) are now running properly. It is agreed that several papers have been produced with both physics packages of ICON, so it seems both are running now.

* Will Sawyer mentions by email that CSCS is interested in doing a performance comparison of COSMO and ICON on Daint, particularly on gpu. It is agreed that the best test case for this would probably be the COSMO-1 setup.

To do points:

* CS will obtain the latex version of the ICON flowchart and distribute it to the group. This is done. The latex source code of the flowchart of page 27-30 in Sebastian Rast's icon lecture is here.

* XL will address two issues at the next ICON infrastructure meeting: the broken grid generator, and the false positives given by the current Buildbot testsuite.

Next meeting:

August 2018 (before the next infrastructure meeting in September)
version 1 uploaded by ColombeSiegenthaler on 07 Jun 2018 - 09:39
version 1 uploaded by ColombeSiegenthaler on 07 Jun 2018 - 10:24
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!