TELEMAC-MASCARET Forum: Problem as running partel with OpenMPI (1/1)

Welcome, Guest

TOPIC: Problem as running partel with OpenMPI

Problem as running partel with OpenMPI 10 years 2 months ago #14236

brahman

Hello Everyone,

For last two days, I have tried to compile Telemac version v6p3r2 with OpenMPI (1.6.5) and Metis (4.0.3) by gfortran (4.8.2) via Python (2.7.4). The installation seems be ok, i.e. it always run and finish gracefully without any error. However, when I ran a test case, the partel command always result error with the following message:

... modifying run command to PARTEL instruction
 partitioning:  T2DGEO
    +>  /home/tqviet/apps/telemac/v6p3r2/builds/py_gfortran_openmpi-165_metis-403/bin/partel < PARTEL.PAR >> partel_T2DGEO.log
At line 439 of file /home/tqviet/apps/telemac/v6p3r2/sources/utils/partel/partel.F (unit = 10, file = 'T2DGEO')
Fortran runtime error: End of file
... The following command failed for the reason above
/home/tqviet/apps/telemac/v6p3r2/builds/py_gfortran_openmpi-165_metis-403/bin/partel < PARTEL.PAR >> partel_T2DGEO.log

It's really weird for me to understand since I tried many times with both gfortran (4.8.2) and ifort (12.1.2), not only with metis 4.3.0 but also with 5.0.2 and 5.1.0 versions as well. After I failed on Arch Linux (my PC), I looked for a luck on Centos, but ... failure again. Both of cases showed the same error on the screen

Even when I had checked openMPI (1.6.5) carefully before the installations, e.g. it works properly with OpenFOAM.

Could you please help me to explain how to fix the error ? Herein I include:

systel.cfg
partel_T2DGEO.log
output of the screen
list of files in folder
cas file

Thank you in advance :)

PS: the example that I tried with is in folder telemac/v6p3r2/examples/telemac2d/tide/

Attachments:

The administrator has disabled public write access.

Problem as running partel with OpenMPI 10 years 2 months ago #14237

brahman	Oh, no ... I'm really sorry. I did upload the wrong files. It totally made me crazy Here I'd like to upload them again. Attachments: systel_2014-09-16.cfg (1KB) output_of_...9-16.txt (5KB) partel_T2D...9-16.txt (5KB)
	The administrator has disabled public write access.

Problem as running partel with OpenMPI 10 years 2 months ago #14240

yugi OFFLINE openTELEMAC Guru Posts: 851 Thank you received: 244	The problem lies with your installation of metis since v6p2 we are using metis 5.x. 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. The following user(s) said Thank You: brahman

Problem as running partel with OpenMPI 10 years 2 months ago #14286

brahman

Hi yugi,

Thank you for your tips.

So far I've installed Telemac v6p3r2 successfully in HPC cluster.

This parallel version goes with Metis 5.1.0. Regarding to Metis's installation, I adopted cmake 3.0.2. Both of Telemac + OpenMPI 1.6.5 + Metis 5.1.0 were compiled by Intel compiler version 13.0.1.

I have just checked the installation with some examples. Especially with examples/telemac2d/seiche/t2d_seiche.cas, the code is running with 8 cores fluently

However, some other test cases were failed and I still dont know why

Anyway, thank you again :)

PS: I think Python scripts is still remaining some wrong things, because it could not handle outputs properly. For instance, after running, the code releases a number of file .sortie, such as t2d_seiche.cas_2014-09-22-23h41min34s_p00001.sortie. There's no any folder for output ???

The administrator has disabled public write access.

Problem as running partel with OpenMPI 10 years 2 months ago #14289

jmhervouet

Hello,

Maybe it is not relevant to your problem but on Intel compiler we have the following problem with Metis called by partel : Metis requires explicitely an unassociated pointer, whereas Intel with some checking options refuses it, considering it is a Fortran error. So we have to compile partel with a limited checking. Note that these checking options considerably slow down the execution time and should not be set for standard runs, we use them here just for finding bugs.

Regards,

JMH

The administrator has disabled public write access.