Author Topic: Problem when constructing input for parallel computing  (Read 8623 times)

hustzjy312

  • Jr. Member
  • **
  • Posts: 29
Problem when constructing input for parallel computing
« on: May 20, 2013, 08:15:18 AM »
Dear All,

It's my first time to use the parallel version of FEAP 83 that is compiled with the help of the research computing support group in my university. However, when I used Metis and ParMetis to generate the input file for parallel computing, both gave me wrong results. For the Metis, although new input files named Ifilename_0001 and so on are generated, the number of total nodes for each new input file is zero. For ParMetis, an error ,'tpwgts is null', appeared. Could you please tell me what are the problems? Are them related to my input files that I used to generate the new input files or the software of Metis and ParMetis?

Could you please give me a standard input file to test the compiling of the parallel version.
 
My serial input file and the input file for generating the input file for the parallel computing are attached.

Thanks!

Yang

FEAP_Admin

  • Administrator
  • FEAP Guru
  • *****
  • Posts: 993
Re: Problem when constructing input for parallel computing
« Reply #1 on: May 26, 2013, 12:43:52 PM »
Are you sure it did note work?  I ran your file and it partitioned just fine.  Perhaps you tried to partition the problem using $FEAPHOMEX_Y/main/feap instead of $FEAPHOMEX_Y/parfeap/feap?

I have attached the partitioned files to this message but I did these with the upcoming version 8.4; so they will not run with your version of the program but at least you can see what they should basically look like.

hustzjy312

  • Jr. Member
  • **
  • Posts: 29
Re: Problem when constructing input for parallel computing
« Reply #2 on: May 29, 2013, 01:41:31 PM »
Thanks very much for your kindly reply.

When I used METIS to partition the mesh, I did use the $FEAPHOMEX_Y/parfeap/feap. I used this command mpirun -np 2 $FEAPHOMEX_Y/parfeap/feap -iIfile to partition the mesh. Is this command correct? By the way, could you please tell me what version of METIS should be used when compiling with FEAP_83?

In addition, when I used ParMETIS to partition the mesh (the input file was changed to the command sets that ParMETIS requires), I got the file graph.filename. But this file is empty. No file named Ifilename_0001 or Ifilenaem_0001 was generated. I suspect that the version of ParMetis that I use may not be correct (version 4.0.2 and 4.0.3 are both tried).  Could you please tell me what version of ParMETIS should be used?

Moreover, I will appreciate if you can tell the version of Petsc I need to use and the compiling options I need to specify to build the binary file.

Yang

FEAP_Admin

  • Administrator
  • FEAP Guru
  • *****
  • Posts: 993
Re: Problem when constructing input for parallel computing
« Reply #3 on: May 29, 2013, 11:54:52 PM »
Unless your problem is huge there is no point in using parmetis to partition it.  Just use metis:

Macro> graph,node,xx
Macro> outd

With regard to petsc versions, with feap8.3, you will need to use petsc 3.1 or some of the early 3.2 releases.  If you use 3.2-p7 then you will need the patches found on the FEAP download site (see the discussion topic on the Installation board).

The next version of FEAP (version 8.4), which is very close to release, will be set up to run with the current petsc release.