Welcome, Guest
Username: Password: Remember me
Forgot your password? Forgot your username? Create an account
Forum
System Modules
ARTEMIS
Artemis and direct solver in parallel (MUMPS)
Flat Threaded

TOPIC: Artemis and direct solver in parallel (MUMPS)

Artemis and direct solver in parallel (MUMPS) 9 years 1 week ago #18925

  • sebourban
  • sebourban's Avatar
  • OFFLINE
  • Administrator
  • Principal Scientist
  • Posts: 814
  • Thank you received: 219
... I presume you know about WinMUMPS: sourceforge.net/projects/winmumps/

Maybe something we can use to build the open source version for windows (not depending on VisualStudio or IntelFortran) ?

Sébastien.
The administrator has disabled public write access.

Artemis and direct solver in parallel (MUMPS) 9 years 1 week ago #18926

  • cyamin
  • cyamin's Avatar
  • OFFLINE
  • openTELEMAC Guru
  • Posts: 997
  • Thank you received: 234
Hello Sebastien,

Changing compiler is not a big issue, although mingw-w64 and the equations suite are both based on gcc and could face the same limitations. On the other hand, switching to MPICH2 from MSMPI would mean that I cannot take advantage of the HPC Pack and its queuing ability. So, it is a step backwards for my installation and I prefer to start from there.

I have come across the WINMUMPS project but it is a Visual Studio/Intel Composer solution, so it is not 'open'. I believe most of the problem is attributed to MSMPI not being entirely compatible to our 'open' building environment.

I have been using MSYS/MSYS2 that provide a windows equivalent to linux environment that seems to work fine in general. I could try MPICH2 just for the compilation and if it works out, it would prove my case.

Regards,
Costas
The administrator has disabled public write access.

Artemis and direct solver in parallel (MUMPS) 9 years 1 week ago #18932

  • cyamin
  • cyamin's Avatar
  • OFFLINE
  • openTELEMAC Guru
  • Posts: 997
  • Thank you received: 234
Hello Sebastien,

Unfortunately, MPICH2 gives the same error, but nonetheless I have made enough progress to have compiled libraries. The built-in tests of each library failed to initialize, but the ones that did (some of LAPACK's) passed.

I did a quick attempt with the ile_para example and I stumbled to this error during the mesh partitioning stage (partel_ARTGEO.log):
 +-------------------------------------------------+
   PARTEL/PARRES: TELEMAC METISOLOGIC PARTITIONER
                                                    
   REBEKKA KOPMANN & JACEK A. JANKOWSKI (BAW)
                  JEAN-MICHEL HERVOUET (LNHE)
                  CHRISTOPHE DENIS     (SINETICS) 
                  YOANN AUDOUIN        (LNHE) 
   PARTEL (C) COPYRIGHT 2000-2002 
   BUNDESANSTALT FUER WASSERBAU, KARLSRUHE
  
   METIS 5.0.2 (C) COPYRIGHT 2012 
   REGENTS OF THE UNIVERSITY OF MINNESOTA 
  
   BIEF 7.1 (C) COPYRIGHT 2012 EDF
 +-------------------------------------------------+
  
  
   MAXIMUM NUMBER OF PARTITIONS:       100000
  
 +--------------------------------------------------+
  
 --INPUT FILE NAME <INPUT_NAME>: 
 INPUT: ARTGEO
 --INPUT FILE FORMAT <INPFORMAT> [MED,SERAFIN,SERAFIND]: 
  FILE FORMAT MUST BE "MED" OR "SERAFIN" OR "SERAFIND" 



 PLANTE: PROGRAM STOPPED AFTER AN ERROR
 RETURNING EXIT CODE:            2

Here is the full output of the command:

File Attachment:

File Name: output_smpd.txt
File Size: 3 KB


Also, the command prompt shows the following error (that is not captured in the text output):
Attempting to use MPI routine before initializing MPI

This error occurs even when I try SOLVER=3, so it seems irrelevant to MUMPS. Direct solver works fine. The ile_para example was taken from the trunk as well.

Any suggestions?

Regards,
Costas
The administrator has disabled public write access.

Artemis and direct solver in parallel (MUMPS) 9 years 1 week ago #18933

  • c.coulet
  • c.coulet's Avatar
  • OFFLINE
  • Moderator
  • Posts: 3722
  • Thank you received: 1031
Hi costas

About the partel problem, I suspect there is a recent change which is not well manage by the script. You could probably bypass it by modifying the partel.par file in the temp directory and run it manually.
Just adding a line with SERAFIN after the name will be sufficient

Regards
Christophe
The administrator has disabled public write access.

Artemis and direct solver in parallel (MUMPS) 9 years 1 week ago #18935

  • cyamin
  • cyamin's Avatar
  • OFFLINE
  • openTELEMAC Guru
  • Posts: 997
  • Thank you received: 234
Hello Christophe,

What is the exact procedure? I never had to do the partitioning manually.

Regards,
Costas
The administrator has disabled public write access.

Artemis and direct solver in parallel (MUMPS) 9 years 1 week ago #18936

  • c.coulet
  • c.coulet's Avatar
  • OFFLINE
  • Moderator
  • Posts: 3722
  • Thank you received: 1031
Hi

My idea is to use the --split function of the scripts
After the partel crash, go in your temp directory, you run partel manually and then you answer the questions.
You could look the required value in the file partel.par as an example.
I suspect partel.par doesn't contain the line about the file format which generate the crash.
If you manually enter the right value (probably serafin) it should works.

After that, you could just run the simulation without the partitionning step.

hope this helps
Christophe
The administrator has disabled public write access.

Artemis and direct solver in parallel (MUMPS) 9 years 6 days ago #18942

  • yugi
  • yugi's Avatar
  • OFFLINE
  • openTELEMAC Guru
  • Posts: 851
  • Thank you received: 244
Hi,

if you are using the trunk version you should not have to change the values of partel.par.
Just running with the option --split should work.

Hope it helps.
There are 10 types of people in the world: those who understand binary, and those who don't.
The administrator has disabled public write access.

Artemis and direct solver in parallel (MUMPS) 9 years 6 days ago #18948

  • cyamin
  • cyamin's Avatar
  • OFFLINE
  • openTELEMAC Guru
  • Posts: 997
  • Thank you received: 234
Hello,

Both the normal execution and the --split option fail during the mesh partitioning stage (with the same error). I didn't manage to configure and feed the partel.par file to partel.exe (I don't know the proper PARTEL.PAR syntax) but, as it emerged it has interactive use as well and did it this way. After manual partitioning, I started the computation with the --run option and got this error (see attached output):

File Attachment:

File Name: output_smpd_2015-11-19.txt
File Size: 5 KB


Is this issue justified by the trunk development or could it be a problem with my compilation?

Regards,
Costas
The administrator has disabled public write access.

Artemis and direct solver in parallel (MUMPS) 9 years 6 days ago #18950

  • c.coulet
  • c.coulet's Avatar
  • OFFLINE
  • Moderator
  • Posts: 3722
  • Thank you received: 1031
Hi Costas

Could you confirm us that this test case work on scalar?

Regards
Christophe
The administrator has disabled public write access.

Artemis and direct solver in parallel (MUMPS) 9 years 6 days ago #18951

  • cyamin
  • cyamin's Avatar
  • OFFLINE
  • openTELEMAC Guru
  • Posts: 997
  • Thank you received: 234
Yes it does:

File Attachment:

File Name: output_smpd_2015-11-19-2.txt
File Size: 8 KB


Costas
The administrator has disabled public write access.
Flat Threaded
Forum
System Modules
ARTEMIS
Artemis and direct solver in parallel (MUMPS)

The open TELEMAC-MASCARET template for Joomla!2.5, the HTML 4 version.