Hello everyone,
I have installed Code Saturne 3.0.1 with PTSCOTCH 5.1.12b.dfsg-1.2 on debian 7.0.0. Installation process have done without any problems. When I started palarell computation of a tutorial (1-simple_junction) with openmpi 1.4.5-1, an error message appeared as attached file (listing.20130521-2008). I also attached two files (listing and error).
I think the error is related to PTSCOTCH. Does anyone have any idea to solve the poblem?
Best regards,
Junichi
Problem about parallel computing of cs-3.0
Forum rules
Please read the forum usage recommendations before posting.
Please read the forum usage recommendations before posting.
-
jun_f_11
Problem about parallel computing of cs-3.0
- Attachments
-
- files.tar.gz
- (6.55 KiB) Downloaded 350 times
-
Yvan Fournier
- Posts: 4208
- Joined: Mon Feb 20, 2012 3:25 pm
Re: Problem about parallel computing of cs-3.0
Hello,
Yes, the crash occurs at the beginning of the PT-SCOTCH partitioning, but I do not know why (I would have expected an initialization issue requiring a specific install option for PT-SCOTCH mentioned in the Code_Saturne install guide, but this is a a true crash.
Were SCOTCH and Code_Saturne packaged ? By the name, I assume SCOTCH was packaged by Debian. If you post your mesh (or mesh_input file in the execution directory) and xml and/or user subroutines, I can see if I reproduced this with PT-SCOTCH 6.0 (the version just after 5.1.12). If I reproduce the error, I'll debug it. Otherwise, it might be a packaging/PT-SCOTCH version issue.
Regards,
Yvan
Yes, the crash occurs at the beginning of the PT-SCOTCH partitioning, but I do not know why (I would have expected an initialization issue requiring a specific install option for PT-SCOTCH mentioned in the Code_Saturne install guide, but this is a a true crash.
Were SCOTCH and Code_Saturne packaged ? By the name, I assume SCOTCH was packaged by Debian. If you post your mesh (or mesh_input file in the execution directory) and xml and/or user subroutines, I can see if I reproduced this with PT-SCOTCH 6.0 (the version just after 5.1.12). If I reproduce the error, I'll debug it. Otherwise, it might be a packaging/PT-SCOTCH version issue.
Regards,
Yvan
-
jun_f_11
Re: Problem about parallel computing of cs-3.0
Thank you Yvan,
I'm using packaged PT-SCOTCH from debian repository site.
I attached a case file (case1.xml), a mesh file (downcorner.des), and a file in execution directory (mesh_input). The case file and mesh file are from Code Saturne tutorial.
Best regards,
Junichi
I'm using packaged PT-SCOTCH from debian repository site.
I attached a case file (case1.xml), a mesh file (downcorner.des), and a file in execution directory (mesh_input). The case file and mesh file are from Code Saturne tutorial.
Best regards,
Junichi
- Attachments
-
- case_files.zip
- (52.52 KiB) Downloaded 344 times
-
Yvan Fournier
- Posts: 4208
- Joined: Mon Feb 20, 2012 3:25 pm
Re: Problem about parallel computing of cs-3.0
Hello,
Your case runs fine on my install (with manual PT-SCOTCH install). I do not have a PT-SCOTCH 5.1.12 install anymore (I replaced 5.1.12 with 6.0), but I suspect the issue is with the package.
I may check this using a Debian package later, but with a non-debug build of ptscotch, I won't go very far. If you replace the ptscotch package with the ptscotch-dbg one, you may get a bit more details in your stack trace, and if you send those to me an I compare them with debian package patches or install setup, I might be able to see where the bug comes from, but I suggest simply doing a manual PT-SCOTCH install (but do not forget to read the section about PT-SCOTCH in the Code_Saturne install manual).
In any case, I probably won't be available to look into this before a week or so.
Regards,
Yvan
Your case runs fine on my install (with manual PT-SCOTCH install). I do not have a PT-SCOTCH 5.1.12 install anymore (I replaced 5.1.12 with 6.0), but I suspect the issue is with the package.
I may check this using a Debian package later, but with a non-debug build of ptscotch, I won't go very far. If you replace the ptscotch package with the ptscotch-dbg one, you may get a bit more details in your stack trace, and if you send those to me an I compare them with debian package patches or install setup, I might be able to see where the bug comes from, but I suggest simply doing a manual PT-SCOTCH install (but do not forget to read the section about PT-SCOTCH in the Code_Saturne install manual).
In any case, I probably won't be available to look into this before a week or so.
Regards,
Yvan