Push notifications in your browser are not yet configured.
You are not logged in, you may not see all content and functionalities. If you have an account, please login .
wrong clm output on mistral
Hello,
i do have some problems in simulating total precipitation in the subtropical region (South East Asia) on the new machine mistral.
In the past i did some experiments on blizzard which worked fine (see attachment: Total_rainfall_blizzard.jpg). Now i wanted to recalculate those experiments on mistral with the same INT2LM & CLM Version but something seems wrong in the output. Vertical stripes occur in the output and i don’t know what i am doing wrong. It seems that it comes from the parameterization (convective_rainfal_mistral.jpg) because the grid scale rainfall looks fine for me.
I am using the INT2LM 1.10_14 and the CLM 4.8_17 version.
Please let me know if you need further information.
Cheers,
David
Hello David,
our COSMO colleague Davide from Bologna lately had a similar problem. And it is not a problem of the name “David”, but he traced it back to a possible compiler bug in the Intel compiler:
Quoting from his Mail:
“we noticed that recent version of Intel compiler
introduces a bug in src_conv_tiedtke of COSMO , producing strange
convective precipitation bands at MPI subdomain boundaries (see
attachment). This has been noticd in operational version 5.0, but I
assume the convection scheme did not change recently.
This is the compiler version:
[dcesari0@node166 ~]$ ifort —version
ifort ( IFORT ) 15.0.2 20150121
Copyright © 1985-2015 Intel Corporation. All rights reserved.
Qnd these are the compiler flags used:
-fno-math-errno -opt-subscript-in-range -align all -axavx -xavx
-heap-arrays -O3
By disabling all the optimizations only on the Tiedtke module (-O0) the
anomaly disappears. The colleagues at Cineca will try to better isolate
the bug (which flag and which code region); it seems that there is no
appreciable feedback on any other model field in atmosphere, so I think
the problem is confined to the prr_conv array. Has anybody noticed this
problem?”
And he showed a figure from convective precipitation, which I attach.
Up to now nobody else complained about that problem, but maybe you are the next one. Are you using the Intel compiler?
Ciao
Uli
Hello Uli,
i hope it is not a problem of the name “David”. Nevertheless i tried your recommendation concerning the optimizations on the Tiedtke module and it worked perfectly. The strange vertical stripes disappeared and the results are comparable to those i calculated on blizzard.
Yes, i am using the intel compiler.
Thank you very much!
Cheers,
David
Hi David,
sorry for the late answer.
Already some weeks ago, I repeated a decadal simulations for Africa on MISTRAL , which I also performed on BLIZZARD within the MIKLIP project.
I could (and can) not see any stripes in the results.
But I did not use the COSMO _4.8_clm17 version, but a more recent one (4.21_clm2, which is not available via the redmine! I also don’ see them when using COSMO _5.0 or COSMO _5.1).
Of course, I used the Tiedtke scheme in any case.
However, there remain some questions:
- which intel compiler version did you use? I guess, it was intel/16.0. I used an older one: intel/15.0.3. Could be interesting to recompile with intel/15.0.3 and to repeat your run, interesting not only for you and me, but perhaps also for the DKRZ people
- which MPI did you use? Bull or INTEL? I used the BULL MPI, see also my recommendations on Redmine—> CCLM at DKRZ
- did you follow my recommendations?
- which compiler options did you use for your first try?
- did you also try to use COSMO _4.8_clm19? This version is available from Redmine. It would be interesting to see whether the stripes also appear when using this version.
Hans-Jürgen
Hello Hans-Jürgen,
It seems that since MISTRAL is in operation now, the “Redmine–> CCLM at DKRZ ” at
https://redmine.dkrz.de/projects/hlre-3-public/wiki
is not publicly accessible anymore.
I have read your recommendations before, but forgot to write them down.
Do you have them saved anywhere else?
Thanks.
Regards,
Weidan
Hi Weidan,
you used the wrong redmine!
The correct one is
http://redc.clm-community.eu/
There, go to “ CCLM at DKRZ ”
Hans-Juergen
Hello Hans-Jürgen,
I’ve used the intel/16.0 compiler. Nevertheless I also tried your recommendation to compile with the older version 15.0.3 but the problem insists. For the MPI i used also Intel but i will try it now with Bull. The compiler flag was set to -O3. After setting it to -O2 the problem with the stripes disappeared.
Like Weidan said i can not see your recommendation anymore and the redmine you are referring to i can not find any topic with CCLM at DKRZ .
Regards,
David
Hi David, hi Weidan,
you are right. Presently, only a handful “selected” members have still access to CCLM at DKRZ .
I don’t know why.
We are working on it.
Hans-Juergen
Hi David, hi Weidan,
please, try again whether you now have access to CCLM at DKRZ .
It should work now, not only for you, but for every member of the CLM -Community.
The reason why you could not see this project for a while was rather simple.
Some time ago an e-mail had been sent to every CLM -Community member by Burkhardt, asking whether he/she is interested in this “ CCLM at DKRZ ” project.
Only a few replied, and if the answer was positive the access was enabled.
For all others no interest was assumed, access was disabled, and as a further consequence they did not get any further e-mail announcing changes related to CCLM at DKRZ
Thus, for every CLM -Community member the lesson form this should be: don’t ignore such e-mails .
If you ignore it, please, don’t complain afterwards
Best regards
Hans-Juergen