Leo van Kampenhout
thanks for your reply. I checked the config_pes.xml file this has not been changed. In fact I still set the number of processors by hand in the following sequence.
The reason being that I still perform the occasional load-balancing test on this machine and want to be able to easily change processor counts.
Since the safety system in config_pes.xml is only called once and by create_case this is most likely the source of the problem.
However it is not forbidden to change processor counts by hand so won’t you agree the safety system should have been implemented in cesm_setup instead?
On 24 Jun 2014, at 17:00, Bill Sacks wrote:
I am unable to reproduce this problem: when I try to set up a similar case on our machine, it correctly gives me 1 GLC task. I'm wondering if you might have modified config_pes.xml (in scripts/ccsm_utils/Machines) in a way that caused it to give different behavior. In particular, note this important section in config_pes.xml:
Order matters in this file, so if you added anything after that section, it could have messed things up.
By the way: I have been asked to direct questions like this to the CESM forums in the future, rather than handling this by email. So if you have any further questions, related to this or other issues, it would be great if you could post them to the appropriate section here:
In particular, note that there are a number of sub-forums under "Ice Sheet Modeling with CISM".
On Jun 24, 2014, at 2:42 AM, "Kampenhout, L. van (Leo)" wrote:
That would be something like:
./create_newcase -case bg.r01.BG1850C5CN.f09_g16.016 -compset BG1850C5CN -res f09_g16 -mach cartesius -compset_file my_compset.xml
The private compset XML is attached. CESM version is 1.1.2
On 23 Jun 2014, at 15:35, Bill Sacks wrote:
Usually that is done automatically. Can you send me the create_newcase command that you used?
On Jun 23, 2014, at 7:33 AM, "Kampenhout, L. van (Leo)" wrote:
thanks, good to have this confirmed. However, it’s too bad that the CESM configure script did not have this knowledge as it could have saved us some time.
On 23 Jun 2014, at 15:12, Bill Sacks wrote:
Hi Miren and others,
It sounds like this problem has been solved, right? It is correct that NTASKS_GLC should be set to 1, since the version of CISM you're using cannot be run in parallel.
On Jun 23, 2014, at 6:14 AM, Miren Vizcaino wrote:
I have a collaboration with Utrecht University (Jan Lenaerts-CC'ed-, Michiel van den Broeke) and they are setting their runs
with the same code I am using for the BGC5 runs.
Leo (CC'ed) reports problems when running in the super-computing facilities in Amsterdam (surfSARA).
Can you help him with this or point to someone who can help?
(Leo, Bill is the software engineer of the Land Ice Working Group of CESM)
Begin forwarded message:
From: "Kampenhout, L. van (Leo)"
Date: June 23, 2014 12:12:47 PM GMT+02:00
To: Miren Vizcaino
Cc: "Lenaerts, J.T.M. (Jan)"
Subject: GLC mpi implementation broken
I’ve been having some troubles with the GLC component.
It gave segmentation faults like the one in the attached logfile.
I was wondering if you have had similar experiences.
The solution for me was to set the number of cores (NTASKS) to one, i.e. do not run in parallel.
Leo van Kampenhout
Institute for Marine and Atmospheric Research (IMAU), Utrecht University
BBG room 672
Princetonplein 5, 3584 CC Utrecht, The Netherlands
+31 30 2533271