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

You are not logged in.

#1 2018-02-15 12:58:27

DURAI04
Member
Registered: 2016-04-15
Posts: 52

Code Aster Memory and CPU Sharing

Dear Team,

I have a 3 system with different configurations, I set a same case for all three systems with 5 lakh elements.

All three taking the same time ( 2 hours ) to finish the job, But i set a CPU processor and memory's are differently.

I am using ASTK Only.

For System 1 : Memory = 4096 MB with 2 Processor

For System 2 : Memory = 3096 MB with 2 Processor

For System 3 : Memory = 4096 MB with 4 Processor

Any specification to set a Memory and Processor in aster ?

SYSTEM NO 1:

Memory     :  7.7 GB

Processor :  Intel Core i5-4460 CPU @ 3.20 GHz * 4

Graphics   : NVD9


SYSTEM NO 2:

Memory     :  7.7 GB

Processor :  Intel Core i5-2320 CPU @ 3.00 GHz * 4

Graphics   : Intel Sandybridge


SYSTEM NO 3:


Memory     :  15.6 GB

Processor :  Intel Core i7-7700 CPU @ 3.60 GHz * 8

Graphics   : HD Graphics 630


How to make a run time lesser with higher configuration workstations.

My question is when i am puting a run in core i5 its taking 2 hours to finish the job. When i am using core i7 with more memory and cpus compare to core i5 its taking same time (2 hours) to finish the job. How to reduce a solver time is less.

Last edited by DURAI04 (2018-02-17 05:26:53)

Offline

#2 2018-02-20 10:20:29

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

Re: Code Aster Memory and CPU Sharing

Hello,

Here is some information regarding parallel performance:

  • code_aster supports both shared-memory parallelism (through OpenMP) or distributed-memory parallelism (through MPI)

  • shared-memory parallelism requires the version to be compiled with OpenMP support. The code_aster versions included in Salome-Meca 2017 are correctly configured for that. Using ASTK ensures that, provided you don't specify anything for the number of threads, code_aster will automatically use the relevant number of cores (up to 6). Note that this kind of parallelism is only efficient if you have big linear systems to solve and you're using MULT_FRONT or MUMPS to do that. Otherwise it is useless (PETSC iterative solver won't take advantage of OpenMP for example)

  • MPI parallelism requires a MPI implementation as well as a correctly configured version. You need to compile such a version yourself as there are no binaries for that. It may be cumbersome but it will provide the most efficient parallelism in code_aster, accelerating also models which don't necessarily have big linear solves but also heavy elementary computations.

TdS

Offline

#3 2018-03-05 04:44:38

DURAI04
Member
Registered: 2016-04-15
Posts: 52

Re: Code Aster Memory and CPU Sharing

Hi,

   Thanks for your reply. Now i understood very well about that.

Thank you!

Last edited by DURAI04 (2018-03-05 04:45:03)

Offline

#4 2019-05-23 10:18:09

nirmaljoshi
Member
From: Japan
Registered: 2018-10-12
Posts: 161

Re: Code Aster Memory and CPU Sharing

I am experiencing similar situation.
I have system with 32Gb ram and 12 processors. But CA uses only 2gb ram at max and 2 CPUs during computation.

For the analysis i am running, it takes 2-3 hours of calculation with MULTI_FRONT. There are about 2000 8-noded quads with 5000 steps. I think this is fairly large problem to solve.

..only efficient if you have big linear systems to solve and you're using MULT_FRONT or MUMPS to do that..

So what does it require to be a "big linear system".

How do we set the CA to use most of the resource and compute the solution faster? And more thing i noted was, it takes significant time to write the result to the output files (e.g. med file or txt file).

Last edited by nirmaljoshi (2019-05-23 11:32:01)

Offline