Welcome, Guest
Username: Password: Remember me

TOPIC: V7p0 run in papallel mode using Windows Cluster

V7p0 run in papallel mode using Windows Cluster 9 years 5 months ago #17350

  • joysanyal21
  • joysanyal21's Avatar
Hi,

I successfully compiled and ran V7.0 64 Bit gfortran in windows using python.

I would like to test run it on 2 windows machines that are connected. I have compiled TELEMAC on these two machines and also mapped a drive (M) which is accessible from both PCs where I intend to place the geometry, cas etc.

I also mentioned the name of the two PCs under 'mpi_hosts' while compiling for this test run

Can anybody help me to launch the simulation using multiple Windows CPUs.
The administrator has disabled public write access.

V7p0 run in papallel mode using Windows Cluster 9 years 5 months ago #17351

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

Are you using MS-MPI (HPC Pack) or MPICH2? Also, can you post your systel.cfg file?

Regards,
Costas
The administrator has disabled public write access.

V7p0 run in papallel mode using Windows Cluster 9 years 5 months ago #17352

  • konsonaut
  • konsonaut's Avatar
  • OFFLINE
  • openTELEMAC Guru
  • Posts: 413
  • Thank you received: 144
Hi,


as a first advice, if you use the configuration file which I submitted yesterday, for mpiexec you should remove the flag -localonly.

Best regards,
Clemens
The administrator has disabled public write access.

V7p0 run in papallel mode using Windows Cluster 9 years 5 months ago #17353

  • joysanyal21
  • joysanyal21's Avatar
I am using MPICH2.

Please find my systel.cfg file

File Attachment:

File Name: systel_parallel_v7p0.cfg
File Size: 2 KB
The administrator has disabled public write access.

V7p0 run in papallel mode using Windows Cluster 9 years 5 months ago #17354

  • cyamin
  • cyamin's Avatar
  • OFFLINE
  • openTELEMAC Guru
  • Posts: 997
  • Thank you received: 234
I have stopped using mpich2 for some time but I have some suggestions:
  1. As Clemens said, drop the -localonly flag
  2. Give the hosts names without spaces (CL06 CL 40 is 3 hosts, not two). Also check that space is the delimiter for hosts (and not e.g. a comma)
  3. The part that I am not sure of: Add '/host <hosts>' in your mpiexec line. <hosts> should be replaced by the entry in mpi_hosts automatically.

Regards,
Costas
The administrator has disabled public write access.

V7p0 run in papallel mode using Windows Cluster 9 years 5 months ago #17356

  • joysanyal21
  • joysanyal21's Avatar
Thanks cyamin for your quick reply.

I am a bit confused about your suggestion about "Add '/host <hosts>' in your mpiexec line"

the mpiexec portion of my config file looks like (deleted the localonly flag):

"mpi_cmdexec: C:\opentelemac-mascaret\mpich2\bin\mpiexec.exe -wdir <wdir> -n <ncsize> <exename>"

Am I supposed to add "/host <hosts>" somewhere in the above line in the config file? If so, where?

It will helpful if you let me know the launch commannd.

Normally I cd to the dir where my files are and type: telemac2s.pt something.cas

What am I supposed to type in this case?
The administrator has disabled public write access.

V7p0 run in papallel mode using Windows Cluster 9 years 5 months ago #17357

  • cyamin
  • cyamin's Avatar
  • OFFLINE
  • openTELEMAC Guru
  • Posts: 997
  • Thank you received: 234
The "-host <hosts>" flag should go where the "-localonly" flag was. Essentially it defines the hosts where the computation should run (instead of just the local host). If I am correct, the string <hosts> should be replaced by the mpi_hosts line in your systel.cfg.

Costas
The administrator has disabled public write access.

V7p0 run in papallel mode using Windows Cluster 9 years 5 months ago #17368

  • joysanyal21
  • joysanyal21's Avatar
I noticed that in some of the config files there are some additional features for running parallel computations in multiple CPUs

For example:

#- reseau de PCs
#RUN_MPI="mpirun mpirun.txt"
#RUN_MPI="mpirun -logon mpirun.txt"
#- PCs en reseau avec solution de connection simplifiee
#RUN_MPI="mpirun_Simple mpirun.txt"

Also there is a mention of mpi_telemac.conf file where there are options of name of the hosts and number of hosts in each host. If I make one where do I place it?

Are all these necessary for the python version? or they are only required for the perl version?

Any advice will be greatly appreciated.
The administrator has disabled public write access.

V7p0 run in papallel mode using Windows Cluster 9 years 5 months ago #17375

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

The example configuration files are particular for the installation they refer to and are meant to be a starting point if your setup is similar. They are meant for python, but many of them are for linux installations so don't get confused.

I would recommend to follow these steps:
  1. Do a successful parallel computation in your local PC
  2. Do the same computation in single a remote PC
  3. Split a parallel computation between 2 machines

I wouldn't bother with the 'machinefile' (the file with the list of your hosts) at this stage. Complete the above steps as simple as possible and then you can add complexity.

Regards,
Costas
The administrator has disabled public write access.
Moderators: borisb

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