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

You are not logged in.

#1 2015-12-23 10:14:21

m_golbs
Member
From: Germany
Registered: 2009-11-09
Posts: 126

ASTK on cluster ASTK>>options>>run options>>....

Hello,

I like to configure a code-aster run by astk on a cluster. In ASTK are ASTK>>Options>>Run Options>>
# ncpus= ... ??? ncpus are the cores on one plattform, one host, one mashine?
# mpi_nbcpu =  ??? Whats behind this parameter?
# mpi_nbnoeud = ??? Whats behind this parameter?

About information I'm very happy.

Thanks Markus


The superfluous chase, is to miss the essential.
Jules Saliège

Offline

#2 2015-12-23 19:25:42

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

Re: ASTK on cluster ASTK>>options>>run options>>....

Hi,

m_golbs wrote:

I like to configure a code-aster run by astk on a cluster. In ASTK are ASTK>>Options>>Run Options>>

Note that for the options below to do something, the Code_Aster version must support them (e.g. the executable must be compatible with OpenMP/threads and/or MPI) and the server must have been correctly configured to handle them (in $ASTER_ROOT/etc/codeaster/asrun)



m_golbs wrote:

# ncpus= ... ??? ncpus are the cores on one plattform, one host, one mashine?

Number of OpenMP threads, sets OMP_NUM_THREAD environment variable. Shared-memory parallelism.

m_golbs wrote:

# mpi_nbcpu =  ??? Whats behind this parameter?

Number of MPI processes, used by the MPI launcher configured in the server to spawn mpi_nbcpu instances of Code_Aster.
Distributed memory parallelism.

m_golbs wrote:

# mpi_nbnoeud = ??? Whats behind this parameter?

Number of calculation nodes, may be used by the batch scheduler to spread MPI processes among several nodes to avoid system overhead.

See ASTK manual [U1.04.00] for more information.

TdS

Offline

#3 2016-01-09 03:06:52

P929
Member
From: Bogotá, CO
Registered: 2015-07-31
Posts: 111

Re: ASTK on cluster ASTK>>options>>run options>>....

Hi all,

after several pain I was able to install a MPI version of CA in a debian system. However, I have a doubt about what it's doing:

When I set ncpu=1 & mpi_nbcpu=24 and I run a self study I can see in the system options that all 24 cores are at 100% and the fans are running and sounding like the PC is under a very heavy load, and the RAM gets full all the time.

However, and thx to an advice from Reiteres I was trying to balance the load on the processors and try to 'play' with the settings of ncpu and mpi_nbcpu. The idea is not to use a single tread with 24 parallel processors, but having several 'threads' running in parallel. So I tried ncpu=4 & mpi_nbcpu=6, so all 24 cores will be used.

So far so good, it sound logic, but when I look at the system monitor not all the processors are shown to be in use, or at least during the first steps of the stat_non_line (pre-conditioner), when it starts iterating one can see that actually more processors get involved.

So the question is: is it normal that to happen? is there something one can do to use more juice from the machine? I mean to use all processors even during pre-conditioning..

Well, if you have any comment plese let me know,

cheers,

Ferney

Offline