Welcome to the forums. Please post in English or French.

You are not logged in.

#1 2011-11-01 19:00:31

JMB365
Member
Registered: 2008-01-19
Posts: 781

ASTK: Launching MPI job in CAELinux2011

Hello,

I decided to configure 2 QuadCore PCs running CAELinux2011 (which came with CA_11.0 compiled for MPI).    I have:

cat /opt/aster110/etc/codeaster/aster-mpihosts
# ubuntu34 slots=4
# ubuntu35 slots=4

cat $HOME/mpi_hostfile 
# ubuntu34:4
# ubuntu35:4

I commented out all the mpi related line of as_run since I am expecting $HOME/.astkrc/prefs to take precedence based on the relevant lines:

# Following values are not used by codeaster-gui (astk)
# but may be useful to as_run.
# (this replace the old ~/.astkrc/config file)
#
editor : /usr/bin/nedit -geometry 132x70+200+30
terminal : /usr/bin/xterm -geometry 132x60+30+30 -e @E
shared_tmp : $HOME/MyFiles/tmp
# proxy_dir : $HOME/MyFiles/tmp
# shared_folders : $HOME/MyFiles/tmp

# ----- MPI commands and parameters
# mpirun
#    available arguments are : mpi_hostfile, mpi_nbnoeud, mpi_nbcpu
#    (use Python string formatting style)
# mpirun_cmd : mpirun -np %(mpi_nbcpu)s -machinefile %(mpi_hostfile)s %(program)s
# mpirun_cmd : mpirun -machinefile %(mpi_hostfile)s -np %(mpi_nbnoeud)s %(program)s
mpirun_cmd : mpirun -machinefile %(mpi_hostfile)s -np %(mpi_nbcpu)s %(program)s

# file which contains list of hosts (REQUIRED even if it is not used in mpirun_cmd)
mpi_hostfile : $HOME/mpi_hostfile

# command called to initialize MPI environment (for example for mpich2)
# mpi_ini : mpdboot --totalnum=%(mpi_nbnoeud)s --file=%(mpi_hostfile)s ; sleep 10
# mpi_ini : mpdboot -n %(mpi_nbnoeud)s ; sleep 10
# mpi_ini : mpdboot --totalnum=%(mpi_nbnoeud)s --file=%(mpi_hostfile)s ; sleep 10
mpi_ini : mpdboot --totalnum=%(mpi_nbcpu)s --file=%(mpi_hostfile)s ; sleep 3

# command called to close the MPI session (for example for mpich2)
mpi_end : mpdallexit

But when I submit and ASTK job I get:

Code_Aster run
<INFO> Command line 1 :
<INFO> ./asteru Python/Execution/E_SUPERV.py -eficas_path ./Python -commandes fort.1 -rep none  -num_job 13305-ubuntu35 -mode interactif -rep_outils /opt/aster110/outils -rep_mat /opt/aster110/NEW11.0/materiau -rep_dex /opt/aster110/NEW11.0/datg -suivi_batch -memjeveux 64.0 -tpmax 54000
Processor ID is not defined !
Processor ID is not defined !
EXIT_COMMAND_13365_00000015=0
<INFO> Code_Aster run ended, diagnostic : NO_RESU_FILE

What am I doing wrong?  I am using CAELinux2011 mostly unchanged (except 'sudo aptitude install mpich2 mpich2-doc', since these do not seem to be installed in CAELinux2011) and the minor modifications to the files I have shown above.  Since I have not messed around trying to compile my own asteru/d MPI versions I was expecting ASTK/CodeAster it to work in MPI mode somewhat 'out-of-the-box', with minor configuration changes.  Of course single processor mode works on the mumps01a.* files that I am testing with.  Any help would be appreciated.  Thanks.

Regards, JMB

PS: BTW cpi does work as shown below:

mpdboot -n 2 -f $HOME/mpi_hostfile (or 'mpiexec -machinefile $HOME/mpi_hostfile -np 2 /tmp/cpi')
mpiexec -n 2 /tmp/cpi
# Process 0 on ubuntu35
# Process 1 on ubuntu34
# pi is approximately 3.1416009869231241, Error is 0.0000083333333309
# wall clock time = 0.001418

Last edited by JMB365 (2011-11-01 19:21:17)


SalomeMeca 2021
Ubuntu 20.04, 22.04

Offline

#2 2011-11-01 19:18:00

Thomas DE SOZA
Guru
From: EDF
Registered: 2007-11-23
Posts: 3,066

Re: ASTK: Launching MPI job in CAELinux2011

You did not fill in the way to find the rank ID of a MPI process :

# shell command to get processor id
# LAM/MPI : echo $LAMRANK
# OpenMPI (1.2) : echo $OMPI_MCA_ns_nds_vpid
# OpenMPI (1.3) : echo $OMPI_MCA_orte_ess_vpid
# OpenMPI (1.34 : echo $OMPI_COMM_WORLD_RANK
# Mpich2  : echo $PMI_RANK
mpi_get_procid_cmd : echo $PMI_RANK

TdS

Last edited by Thomas DE SOZA (2011-11-01 19:18:25)

Offline

#3 2011-11-01 22:11:46

JMB365
Member
Registered: 2008-01-19
Posts: 781

Re: ASTK: Launching MPI job in CAELinux2011

Hello TdS,  It worked!  Thank you VERY much!  Regards, JMB


SalomeMeca 2021
Ubuntu 20.04, 22.04

Offline

#4 2012-03-07 14:35:03

frotscher
Member
From: Aachen, Germany
Registered: 2010-11-17
Posts: 124
Website

Re: ASTK: Launching MPI job in CAELinux2011

Hi!

I have a similar problem as JMB365 but unfortunately the proposed solution doesn't work for me (as I want to use the preinstalled OpenMPI if possible).

The only thing that I changed in my fresh CAELinux 2011 installation was adding _DISABLE_SCOTCH to the DEFS line of the config-file because SCOTCH used to cause problems in older installations. The reason that I didn't change anything else in the installation was my assumption that everything in the fresh installation should work without any modifications.

Please find attached my config-file and the output file of the computation run that I started via ASTK. In the output file there are error messages that I am not able to interpret. I suppose that there are some library confusions in the system.

Therefor I kindly ask for your help correcting these errors because I really need Code_Aster running in parallel smile

Thank you very much in advance,

Ralf


Attachments:
files.tar.gz, Size: 6.79 KiB, Downloads: 576

Ubuntu 64bit, SALOME 7.2, Code_Aster 12.1 (sequential and parallel), ASTK 1.13.1

Offline

#5 2012-03-13 08:35:24

frotscher
Member
From: Aachen, Germany
Registered: 2010-11-17
Posts: 124
Website

Re: ASTK: Launching MPI job in CAELinux2011

Hi at all again,

I was able to solve this problem myself. I compared the config-file of the Code_Aster installation coming with the current CAELinux with an older one and changed the MAKE and BIN* lines in it:

MAKE           | build   | -     | debug nodebug 
BIN_NODBG      | bin     | 11-04 | asteru_mpi 
BIN_DBG        | bin     | 11-04 | asterd_mpi 
BINCMDE        | bin     | 11-04 | mpi/commande 
BINELE         | bin     | 11-04 | mpi/elements 
BINPICKLED     | bin     | 11-04 | mpi/cata_ele.pickled 
BINLIB_NODBG   | bin     | 11-04 | mpi/lib/libaster.a 
BINLIB_DBG     | bin     | 11-04 | mpi/lib/libasterd.a 
BINLIBF_NODBG  | bin     | 02-05 | mpi/lib/libferm.a 
BINLIBF_DBG    | bin     | 02-05 | mpi/lib/libfermd.a 
BINOBJ_NODBG   | bin     | 11-04 | mpi/obj/aster 
BINOBJF_NODBG  | bin     | 11-04 | mpi/obj/ferm 
BINOBJ_DBG     | bin     | 11-04 | mpi/obj/asterd 
BINOBJF_DBG    | bin     | 11-04 | mpi/obj/fermd

Now the parallel computation using MUMPS works in general for simple test cases.


Ubuntu 64bit, SALOME 7.2, Code_Aster 12.1 (sequential and parallel), ASTK 1.13.1

Offline