Log In | Users | Register
COSMO Web
Edit | Attach | New | Raw | Diff | Print | | Tools
You are here: ICON » ICONmeet20190328

ICON meeting March 28, 2019

Participants:

Dominik Brunner (DB); Valentin Clement (VC); Remo Dietlicher (RD); Saskia Drossaart van Dusseldorp (SD); Bernhard Enz (BE); Sylvaine Ferrachat (SF) - partially; Doris Folini (DF); Xavier Lapillonne (XL); Philippe Marti (PM); David Ochnser (DO); Will Sawyer (WS); Colombe Siegenthaler (CS);

Reports:

* Saskia Drossaart van Dusseldorp (SD):
  • New PhD by Ulrike. Will work on hail prediction in ICON (NWP physics).

* Colombe Siegenthaler (CS):

  • CS began to write python scripts which clone the official ICON organization from MPI repository and push the tags on C2SM Github organization.

* Xavier Lapillonne (XL):

  • XL is organizing an ICON GPU Hackathon week in Mai (20-24th of Mai 2019). The location of the meeting will be the main building at ETHZ. The goal is to push the porting of ICON on GPU using OpenACC commands. The attendants will organizing themselves into teams, which will have tasks. XL will send a short OpenACC tutorial before-head. Some GPU experts will be in the room. One person from PGI compilers company will be present.

* Sylvaine Ferrachat (SF):

  • Is working on another project will be finished at end of April. Afterwards will study OpenACC to be ready for ICON GPU week XL was mentionning.
  • Ulrike has another student, Colin Tully, who will soon work with ICON-HAMMOZ and help debugging.
  • C2SM wrote a technical equipment proposal for compute nodes on Euler and storage. The proposal was accepted.

* Doris Folini (DF):

  • DF has nothing to report yet.
  • Interested in both 200km resolution to have long term projections and high resolution, maybe using nesting facility.
  • SF has been told by Sebastian Rast (MPI) that 2 ways-nesting is now also working with the climate physics.

* Valentin Clement (VC):

  • Valentin is working on JSBACH. 50% is done.

* Philippe Marti (PM):

  • All the climate physics is ported on GPU, instead of radiation scheme.
  • The interface to the dynamical core is not ported. At the interface, all the fields are hardly copied, no pointer is used. However MPI is about to rewrite the interface, maybe it will not be the case anymore.

* Remo Dietlicher (RD):

  • RD is porting ICON-LAM on GPU in the project IMPACT.
  • Meteoswiss is interested in LAM version (equivalent to 1-way nesting), the driving model will be IFS.

* Bernhard Enz (BE):

  • nothing to report, just comes back from Japan.

* David Ochnser (DO):

  • nothing to report. Did not work with ICON yet but is interested in ICON-ART.

* Dominik Brunner (BD):

  • DB is would like to port many development they are using in ICON-ART.
  • First has to sign ICON and ART license. XL says you get and implicitly accept an individual ICON license when cloning the ICON code from MPI Git server.
  • ART is not running on GPU. WS says that KIT is interested in porting ART on GPU.
  • DB is not using COSMO-ART on GPU , but other passive tracer parametrization which is on GPU.

* Will Sawyer (WS):

  • WS is trying to get ready for Xavier's Hackaton.
  • WS is trying to transition 'ICON on GPU' from a 'development' project to a 'production' project, i.e. that people in the community use ICON on GPU.
  • WS is interested to know the plans of people in the community concerning the use of ICON.
  • QUBICC project: A group in MPIMet (Hamburg) wants to get allocation of half of Piz Daint to run 2 years of simulations at 2.5 km, 200 vertical levels globally. They want to study the quasi-biennial oscillation in a changing climate. They will use a similar set-up as the Diamond++ simulations (cloud microphysics from NWP branch, other parametrisations are echam physics, but in high resolution, deep convection resolved). They are doing tests in a Preparatory projects in order to submit a PRACE production project in the coming May. This will be the 1st big configuration which runs on a large GPU allocation.
  • Status ENIAC :
    • The coupling is mostly done (excluding radiation). They are still bugs in the interface, diffusion, etc.
    • The dynamical core gives zero differences between CPU and GPU with intel compiler.
    • horizontal/vertical advection and diffusion are now tested on GPU.

Topic of common interest:

* ICON license for ETHZ members (CS for C2SM):
  • ICON and GridTools licenses are linked together and have to be signed at the same time.
  • On ETHZ side, the license for the official ICON and the one for GridTool are fine.
  • GridTool license is now on DWD's table. They want to check that the GridTool license is not toxic for ICON.
  • As soon as DWD does say that the GridTool license is fine, both licenses can be signed.
  • The above mentioned license is for the official ICON version. Concerning the NWP, COSMO and AES ICON branches, there will be other one-to-one licenses, one with DWD for NWP and COSMO branches and one with MPI (for the AES branch).

* Radiation scheme:

  • PSRad : The current radiation scheme in the climate physics is PSRad . PSRad is not able to run on GPU.
  • RRTmpg : Another radiative scheme is RRTmgp. It simulates the same physical processes than PSRad but is implemented differently. Gustavo is porting RRTmgp in echam physics of ICON. Gustavo is leaving, so he will not have the time to port RRTmgp on GPU. However, most of the scheme has been already ported independently of Gustavo by people in NCAR.

* Outcome developer & infrastructure meeting last February in DKRZ, Hamburg:

  • The transfer of all ICON branches on a GitLab server hosted by DKRZ was accepted at the last ICON developer meeting in Hamburg last February. The GitLab wiki facility will be used. However the current ICON wiki on Redmine will not be transferred automatically to the wiki of GitLab, only the specific pages some user ask will be transferred. The reason is that lots of wiki pages are outdated.
  • The ICON developer meeting next winter will be part of ICCARUS at DWD, Offenbach.

Questions:

* DB: ICON (both NWP and echam physics) are using the non-hydrostatic dynamical core. Is ICON conserving dry or moist air?

* WS: Is there some plan to run ICON on Euler? If yes, who will install it?

  • Answer CS:
    • Yes, ICON will also be run on Euler.
    • Sylvaine and Colombe will set ICON up on Euler.

Next meeting:

June-July 2019 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!