open TELEMAC-MASCARET The mathematically superior suite of solvers

I am hoping that somebody might provide guidance on getting the telemac software to run in parallel mode for Windows 7 (64bit) standalone computer with intel fortran (ifort v11.1) and python installation.

I installed the scalar version and it runs correctly. I then reinstalled the system for wintelmpi

The parallel version completes splitting the mesh into submeshes with the partel log showing it to terminate normally but tails to run the cas file after that.

I include the systel cfg file and the screen output

I have used the compiled libmetis64b.lib and mpi is from the MPICH2 64bit version.


The compile of telemac indicated that it had comiled sucessfully.

I would welcome any guidance on this

kind regards

Tony Cawley

Hi,

You forgot to post the screen outputs.
For this specific topic, similar posts already exist.

You can refer to this post for debugging your configuration : www.opentelemac.org/index.php?option=com...mid=177&limitstart=0

Can you post your systel file and the output errors ?
Are the default parallel executables created ?
If so, you should try to run directly the mpiexec from the TMP directory?
Did you setup the passphrase which goes with smpd/mpi ?

Regards,

Fabien

Hi Fabien,


see attached documents (i thought that i had attached them)
I think its my mpi setup not sure how to fix it

in the temp directory it only creates one executable even though I specify 3 parallel processors.

Should I try and reinstall mpi ?






regards
Tony

sorry the systel file is attached


# _____ _______________________________
# ____/ TELEMAC Project Definitions /______________________________/
#
[Configurations]
configs: wintelmpi

# _____ ____________________________________
# ____/ windows intel parallel /___________________________________/
[wintelmpi]
#
root: C:\opentelemac\v6p2r1
version: v6p2
language: 2
modules: update system
#
options: parallel mpi
mpi_hosts: -localonly
mpi_cmdexec: C:\opentelemac\mpi\bin\mpiexec.exe <wdir> <ncsize> <hosts> <exename>
#
cmd_obj: ifort.exe /c /Og /QxHost /iface:cref /iface:nomixed_str_len_arg /nologo /names:uppercase /convert:big_endian /extend_source:132 <mods> <incs> <f95name>
cmd_lib: xilib.exe /nologo /out:<libname> <objs>
cmd_exe: xilink.exe /nologo /subsystem:console /stack:536870912 /nodefaultlib:libc.lib /out:<exename> <objs> <libs>
#
mods_all: /include:<config>
#
incs_parallel: /include:C:\opentelemac\mpi\include
libs_parallel: C:\opentelemac\lib\libmetis64b.lib
libs_all : C:\opentelemac\mpi\lib\fmpich2.lib
#
sfx_zip: .zip
sfx_lib: .lib
sfx_obj: .obj
sfx_mod: .mod
sfx_exe: .exe
#
val_root: <root>\..\validation
val_rank: all




regards
Tony

Hi,

Remove <hosts> from the mpiexec sequence in your config file.
This option usually goes with -machinefile. However, when running on a local PC, you don't need to provide any machinefile.

I will update the config files on SVN.

Parallelism within Telemac uses the very classical SPMD technique, so that's perfectly normal that there's only one executable in the TMP directory. Only data are distributed.

Regards,

Fabien

thanks Fabien, but unfortunately this did not work i had tried this earlier but tried again with complete clean and recompile of telemac. I attach the screen output which I think it might indicate that the error lies with mpiexec

I also disabled firewall with no change

in the mpi examples directory I tried the following command
mpiexec -n 3 cpi (where cpi.exe is the example fortran exec)
It also reported same error while connecting to host as the screen output I attach.

I have also tried this installation on other standalone W7 computers with similar outcome.

the MPICH2 was the compiled executable version for 86 and 64bit windows

Regards
Tony Cawley

Hi Tony

For me on a similar computer it works well. but I reminded that I had some problems at the beginning

For the configuration of telemac itself, on my side, I have a -n before <ncsize> ...

But the main problem should come from the mpi installation itself. Particularly if your intel compiler is also delivered with mpi...
Check in your computer haw many version of mpiexec you have.
I thought you will find 2 version, 1 in the mpi installation and 1 in the ..\Intel\ComposerXE-2011\redist\intel64\mpirt

So first remove the path to mpirt to be sure you execute the right mpiexec (both are not compatible...)

Maybe it's necessary also to install mpich2 manually from a DOS command windows with administrator rights.

Hope this helps

As Christophe has stated, you're missing a -n option before passing the number of required cores. The correct syntax is

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

However, the CPI program has to run in parallel. If not, there is also a problem with MPI. You can force the smpd to restart with "smpd -remove" and then "smpd -install". or you can play with wmpiconfig & wmpiregister.

Regards,

Fabien

Fabien, I included your -wdir and -n but same problem

Of interest if I run smpd -remove or -install it reports the follwong message

OpenSCmanager failed:
Access is denied <error 5>

This also occurs on other W7 computers

regards
Tony

Hi Christophe

I checked for a second version of mpiexec on the computer and within the fortran compiler directories and no such exec exists my version is 11.1 which is (2009-2010).

I reinstalled MPICH2 from the DOS prompt and inserted the -n into the config file but to no avail.

I feel the problem lies outside of telemac and with MPI and my windows 7 (even though the same problem exists on 2 other different W7 computers)
On the computer I am testing I have administrator rights and I am the only registered user of this compute

what tests could I carry out to identify if the mpiexec.exe is setup correctly.

Regards
Tony Cawley