Log In | Users | Register
COSMO Web
Edit | Attach | New | Raw | Diff | Print | | Tools
You are here: ICON » Older_meetings » ICONmeet20180807

ICON meeting August 7th, 2018

Participants:

Bernhard Enz (BE); Carlos Osuna (CO); Colombe Siegenthaler (CS); Philippe Marti (PM); Will Sawyer (WS); Xavier Lapillonne (XL); Zane Dedekind (ZD);

Update of each participant:

* Philippe Marti (PM):
  • PM works in the ENIAC project which consists of porting ICON on GPU using OpenACC directives. Standalone parts are needed. Inside ENIAC, a tool was written to automatize the extraction of standalone parts. The tool is not working anymore for the release candidate because of the last merge (of another group) in the release candidate one week ago. The problem comes from a part of code which has been rewritten in Fortran 2008 features (data structures using abstract derived types).
  • PM is working on porting the graupel parameterization in GPU. PM needs reasonable data for testing. MPI cannot deliver data because they are not finished in coupling the graupel parameterisaton to the ECHAM physic. DWD gave data, but since DWD is not using the official ICON version (DWD is using icon 2.3.00 + modifications), their graupel data leads to a segmentation fault with the release candidate.

* Colombe Siegenthaler (CS):

  • CS opened a new GitHub organization called "C2SM-ICON" aimed to host ICON and companion codes repositories for the C2SM community. For the moment, only the DWD Tools are there. C2SM also aims to set-up Jenkins on the C2SM-ICON repositories in order to be able to run our own tests. CS will prepare a more specific plan of the organization of the different repositories and their relationship for the next meeting.
  • CS is trying to get the atm_amip script of ICON on Piz Daint. CS get an HDF5 error with atm_amip_test while the nh35_tri_jws script runs properly. A suggestions form the other participants was to try launch the script from the scratch system and not from the home directory.

* Will Sawyer (WS);

  • WS refactored the dynamical core. If few loop orders were changed, most of the changes concerns the locations of OpenACC directives. The refactoring was reviewed and accepted by people, including Günther (Zängl) and Florian (Prill). WS pushed the refactored dynamical core up to MPI Git repository. It is in the release candidate. WS had lots of problems from the last commit one week ago (because of Fortran 2008 features).
  • XS asked WS to keep the socket to socket performance comparison because it is usefull for MeteoSwiss.
  • Thomas (Schulthess) wants to have a global ICON simulation in high-resolution, similar to the COSMO-global run. Thomas wants to have a comparison of benchmarks of different models: COSMO-global, IFS, ICON, .... For the moment, it would be a CPU run since there are problems with the standalone (cf PM's report above). For setting up this simulation, WS will need help form Luis (Kornblueh) and people at Meteoswiss who did the COSMO-global run.
  • CO reported that he knows Philipp Neumann and Joachim Biercamp at DKRZ who runned ICON globally at high resolution (3 or 5km) on DKRZ machine. CO remember that 2km resolution had some problems, but 3 or 5 kms was definitively done. People in DKRZ were targeting scientific comparison.

* Bernhard Enz (BE):

  • BE managed to construct the grid over his domain, boundary conditions and initialization files.
  • BE needs now to figure out how to update the SST during a run in LAM mode (without ocean coupling).
  • The simulation domain has 1-1.5 million cells. WS suspect this will not fit on none node.
  • The planned resolutions are 13 km, 6.5 km, and 13km containing 1-2 nests of 6.5 km.
  • BE works on an algorithm tracking hurricanes in the ICON grid. In case anyone is interested, contact BE.

* Zane Dedekind (ZD);

  • ZD currently using COSMO, but plans to use ICON in his thesis.
  • ZD is interested in the relation of ice crystal concentration and ice nuclei particle. The simulations will be over the alpine domain.
  • Axel Seifert is developing a new microphysical scheme using superdroplets in ICON. Axel wants to have the new scheme finished in 3D by the end of the year. For the moment 1D and 2D are tested.

* Carlos Osuna (CO):

  • CO is not working directly with ICON yet. CO is writing up the road-map and planning for the transition form COSMO to ICON in MeteoSwiss (planned for 2022).
  • CO is part of the new priority project IMPACT. The role of CO is to link HPC/enginering projects like ENIAC, PASCHA with the COSMO consortium.
  • CO is involved in the renewal proposal for a new machine dedicated to MeteoSwiss. This new machine will come in 2019 and should be optimized for COSMO. The next machine will come in 2022.

* Xavier Lapillonne (XL):

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

Summary of the discussions:

* Points to bring up at the next ICON infrastructure meeting in September (PM and WS will attend):

  • The testing should work. No push should be possible as long as buildbot gives false positive.
  • WS would like to have a status of ICON ocean.
  • Someone could be hired to do the release management and take care of buildbot. This person would be responsible of formal reviewing of the merges (eg. be sure that all builldbot tests are actually passed before code is pushed, insure that the merge are squashed, check that huge commits are divided into smaller pieces, etc).
  • For the moment, nobody is responsible in supporting a good collaboration between groups.
  • Review/establish a formal procedure for checking in the official master.
  • New codes should have some days staggering to let people test before the next commit of another group happens.
  • Change the relative sub-modules into absolute sub-modules, or make the sub-modules public. This problem will anyway have to be resolved when the COSMO consortium will switch to ICON.
  • It should be possible to run each buildbot test manually on every machine where ICON is installed. This would request to have the data on pool and the scripts to be written in a generic way (most of the buildbot tests are PBS scripts).
  • Impose some convention so that all ICON programmers are using the same kind of functionality. The problem is not using fancy things, but that everyone is doing things differently.
  • Ask to go back to the previous official release candidate (i.e. canceling the last commit).
  • If something comes up to your mind, send an email to WS or PM.

* XS will officially report to Philip Steiner that there are problems in the committing procedure and collaboration between groups.

* WS ask if someone would be interested in having access in the push up workflow icon-eniac repository -> icon-cscs repository -> icon-release repository. For the moment, WS is the only person to have this access. XS said he is interested.

* CS asked if a clean recipe of getting code in the official version using git commands exists (her idea is to use/write up such a document for the management of C2SM? ICON repositories). WS took one recipe written by Luis : https://code.mpimet.mpg.de/projects/icon/wiki/Group_Gatekeeper_Merge_Cookbook and adapted it. XS will review the WS's guidline. CS will wait till XS WS's document is reviewed.

Next meeting:

November 2018 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!