Log In | Users | Register
COSMO Web
Edit | Attach | New | Raw | Diff | Print | | Tools
You are here: ICON » Older_meetings » ICONmeet20180315

ICON meeting March 15th, 2018

Participants:

Bernhard Enz (BE); Colombe Siegenthaler (CS); Doris Folini (DF); Guy de Morsier (GDM); Katherine Osterried (KO); Philippe Marti (PM); Sylvaine Ferrachat (SF); Valentin Clément (VC); William Sawyer (WS); Xavier Lapillonne (XL)

Update of each participant:

* Colombe Siegenthaler (CS):
  • CS is working on interpolating HAM emission files on the ICON grid for feeding in ICON-HAM.

* Guy de Morsier (GDM):

  • GDM can now successfully compile and execute ICON on Kesch. It runs only with small nproma values (<16). With higher nproma, ICON crashes with a segmentation fault in hdf5.
  • GDM plans to work more with ICON in the next months.
  • GDM plans to report on the COSMO/ICON dynamical core comparison experiment in September.

* Sylvaine Ferrachat (SF):

  • A new PhD student has started in Ulrike's group (Zane Dedekind). He started working with COSMO, but will work afterwards with ICON. He will look at the cloud microphysics interaction with orography and compare ICON with COSMO.

* Valentin Clément (VC);

  • VC is trying to parse ICON with CLAW. VC will now work on using CLAW in JSBACH.
  • JSBACH is using lots of FORTRAN 2003 and 2008 features. VC sends pull requests to the OMNI compiler developers to get those features supported by the compiler, because the CLAW compiler is based on the OMNI compiler. Until now almost all pull request were accepted.

* Philippe Marti (PM):

  • PM extracted vdiff and JSBACH in the standalone tool.
  • All the routines around JSBACH (between vdiff and JSBACH) are now ported to GPU using OpenACC.
  • PM will now work on JSBACH which will be ported using CLAW together with Valentin.

* Bernhard Enz (BE):

  • BE will use ICON-LAM with forcing at the boundaries.
  • BE uses a nonstandard domain avoiding the Mexico Bay but including the mountains.
  • BE could talk to Florian Prill and H2CP people for help generating his grid.

* Katherine Osterried (KO):

  • KO is going next week to Offenbach to work with Jürgen and Luis on EXPAR. Since she is the COSMO source code administrator, she will make sure that the new ICON specific developments do not break anything for the COSMO use.

* William Sawyer (WS):

  • WS did a performance comparison between COSMO and ICON. A short report of this comparison can be found in the copy of WS's reporting email and the attached excel sheet.
  • WS, PM and Thomas Köster are using the FortranTestGenerator tool of Christian Hovy (Univ. Hamburg) to generate standalone test cases. WS has generated a “velocity_tendencies” standalone test which is somewhat representative for the dycore as a whole.
  • There is a new diploma student from the university of Lugano (Thomas Köster) who is working on the porting of vdiff. His diploma project lasts for 6 months. Thomas already validated the CPU version with 4 compilers. He is beginning with GPU porting now. Thomas is really fast, hopefully he will have a fully ported version of the turbulence parameterization by the end of his diploma thesis.
  • WS, PM, XL are again working on the OpenACC dynamical core by restructuring it for very large nproma (essentially equal to the full number of vertical columns). Once it is done, WS will do a performance analysis.
  • WS is responsible of the merging of ENIAC work into the CSCS git repository.
  • WS is part of the ICON infrastructure meeting. Next meeting will be a teleconference.
  • WS encourage people which have problems with compilers (pgi,cray,intel) to open a ticket at CSCS. WS or a colleague will be assigned to the ticket and report it back to the compiler people.

* Doris Folini : nothing to report

Reports of the last ICON meetings

* ICCARUS

* ICON user meeting

  • There are 2 papers with the ECHAM physics in AMIP simulation which are submitted to JAMES. ICON with ECHAM physics has been registered for CMIP6 simulations.
  • The people in DKRZ are working on the parallel I/O routines; however they have problems with the vertical levels in JSBACH.
  • YAXT will have a new interpolation method.
  • For the moment it is not possible to do nesting or LAM mode with the ECHAM physics.
  • The data structure for ECHAM and NWP physics are completely different, stored in a different place, with different units and variable names.

* ICON infrastructure meeting

  • It has been decided to do work on the configuration of ICON.
  • mh-linux file will be broken up into one file by institution.
  • buildbot will be cleaned up.
  • The package manager SPACK was chosen for ICON. There are tutorials available. Sergei Kosukini (?) at MPI is working on using SPACK with ICON. He will maintain it. SPACK is flexible enough to run on your laptop, if needed. Sergei is having trouble getting SPACK working on CRAY machines, particularly on Daint and the DWD machine.
  • There were discussions about splitting the monolithic ICON code into more libraries. The final decision was to do nothing for the moment and to charge Luis to define a plan.
  • Buildbot is giving some times FALSE negatives and FALSE positives. You can get red points even if there is no problem (just need to resubmit the test and it turns green). Be cautious using buildbot. This also happens with Jenkins.

Summary of the discussions:

  • KO asks if it will be possible to add tests to the buildbot system. WS says yes definitely. WS will add a performance test. When people really start working with ICON, KO will collect representative tests for COSMO and COSMO-CLM and come back to WS.
  • KO asks if ICON will be used for regression tests at CSCS. WS says for the moment not, but they will.
  • DWD will buy a new machine in the next year or two, so they did probably already did performance analysis with ICON. It would be nice to have access to these analyses, maybe Philippe Steiner could ask?
  • New MeteoSwiss operational forecast machine planned for 2020. Replacement of Piz Daint probably around 2022.
  • SF have seen a comparison of package managers easybuild and SPACK. You can find it here.

To do points:

* WS will try to debug ICON with DDT.

* CS will try to run a newer version of ICON on 18 nodes (12 MPI proc per node).

* WS will send around a document where he compared the performance of ICON against COSMO.

* WS will ask for an invitation at the infrastructure meeting for Xavier and Katie.

* SF will send around the link to a comparison of package managers easybuild and SPACK.

* GDM will ask Philipe Steiner to ask DWD people for their ICON performance analysis.

Next meeting:

June 2018
Attach
doc
version 1 uploaded by ColombeSiegenthaler on 15 Mar 2018 - 15:43
xls
version 1 uploaded by ColombeSiegenthaler on 20 Mar 2018 - 11:00
txt
version 1 uploaded by ColombeSiegenthaler on 20 Mar 2018 - 11:22
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!