Author Topic: compiling PETSc in HPC  (Read 7014 times)

shenrilin

  • Full Member
  • ***
  • Posts: 67
compiling PETSc in HPC
« on: August 04, 2017, 07:53:24 AM »
Dear All,

I want to run feap in HPC and try to configure the petsc firstly with openmpi.

I met the following errors:
Code: [Select]
Running test examples to verify correct installation
Using PETSC_DIR=/rigel/free/users/rs3741/SourceCode/petsc-3.5.4/ and PETSC_ARCH=gnu-dbg
[size=12pt][b]Possible error running C/C++ src/snes/examples/tutorials/ex19 with 1 MPI process[/b][/size]
See http://www.mcs.anl.gov/petsc/documentation/faq.html
[warn] Epoll ADD(4) on fd 1 failed.  Old events were 0; read change was 0 (none); write change was 1 (add): Operation not permitted
[1501856734.216578] [holmes:48951:0]         sys.c:744  MXM  WARN  Conflicting CPU frequencies detected, using: 2200.00
[1501856734.224692] [holmes:48951:0]         mxm.c:196  MXM  WARN  The 'ulimit -s' on the system is set to 'unlimited'. This may have negative performance implications. Please set the stack size to the default value (10240)
[1501856734.229386] [holmes:48951:0]         mxm.c:196  MXM  WARN  The 'ulimit -s' on the system is set to 'unlimited'. This may have negative performance implications. Please set the stack size to the default value (10240)
lid velocity = 0.0016, prandtl # = 1, grashof # = 1
Number of SNES iterations = 2
[size=12pt][b]Possible error running C/C++ src/snes/examples/tutorials/ex19 with 2 MPI processes[/b][/size]
See http://www.mcs.anl.gov/petsc/documentation/faq.html
[warn] Epoll ADD(4) on fd 1 failed.  Old events were 0; read change was 0 (none); write change was 1 (add): Operation not permitted
[1501856735.049456] [holmes:48962:0]         sys.c:744  MXM  WARN  Conflicting CPU frequencies detected, using: 2200.00
[1501856735.055401] [holmes:48963:0]         sys.c:744  MXM  WARN  Conflicting CPU frequencies detected, using: 2200.00
[1501856735.057904] [holmes:48962:0]         mxm.c:196  MXM  WARN  The 'ulimit -s' on the system is set to 'unlimited'. This may have negative performance implications. Please set the stack size to the default value (10240)
[1501856735.061983] [holmes:48963:0]         mxm.c:196  MXM  WARN  The 'ulimit -s' on the system is set to 'unlimited'. This may have negative performance implications. Please set the stack size to the default value (10240)
[1501856735.062214] [holmes:48962:0]         mxm.c:196  MXM  WARN  The 'ulimit -s' on the system is set to 'unlimited'. This may have negative performance implications. Please set the stack size to the default value (10240)
[1501856735.066630] [holmes:48963:0]         mxm.c:196  MXM  WARN  The 'ulimit -s' on the system is set to 'unlimited'. This may have negative performance implications. Please set the stack size to the default value (10240)
[1501856735.610307] [holmes:48962:0]         mxm.c:196  MXM  WARN  The 'ulimit -s' on the system is set to 'unlimited'. This may have negative performance implications. Please set the stack size to the default value (10240)
[1501856735.610325] [holmes:48963:0]         mxm.c:196  MXM  WARN  The 'ulimit -s' on the system is set to 'unlimited'. This may have negative performance implications. Please set the stack size to the default value (10240)
lid velocity = 0.0016, prandtl # = 1, grashof # = 1
Number of SNES iterations = 2
[size=12pt][b]Possible error running Fortran example src/snes/examples/tutorials/ex5f with 1 MPI process[/b][/size]
See http://www.mcs.anl.gov/petsc/documentation/faq.html
[warn] Epoll ADD(4) on fd 1 failed.  Old events were 0; read change was 0 (none); write change was 1 (add): Operation not permitted
[1501856742.343899] [holmes:49045:0]         sys.c:744  MXM  WARN  Conflicting CPU frequencies detected, using: 2200.00
[1501856742.350678] [holmes:49045:0]         mxm.c:196  MXM  WARN  The 'ulimit -s' on the system is set to 'unlimited'. This may have negative performance implications. Please set the stack size to the default value (10240)
[1501856742.354726] [holmes:49045:0]         mxm.c:196  MXM  WARN  The 'ulimit -s' on the system is set to 'unlimited'. This may have negative performance implications. Please set the stack size to the default value (10240)
Number of SNES iterations =     4
Completed test examples


Does anybody have experiences on this stuff?

Best,
« Last Edit: August 04, 2017, 07:55:42 AM by shenrilin »

FEAP_Admin

  • Administrator
  • FEAP Guru
  • *****
  • Posts: 993
Re: compiling PETSc in HPC
« Reply #1 on: August 04, 2017, 08:23:11 AM »
1) You should not use old versions of PETSc unless you have to.

2) Problems with PETSc installation/configuration are best posted to the PETSc mailing list.

shenrilin

  • Full Member
  • ***
  • Posts: 67
Re: compiling PETSc in HPC
« Reply #2 on: August 05, 2017, 09:11:46 AM »
Dear Feap_Admin,

Thank you for your advice.

I'm now using Feap8.3, I think Petsc3.5.4 is compatible with feap8.3. If I replaced the petsc3.5.4 with the latest version, I'm not sure which part of feap8.3 should be changed. Could you give me some instructions on this issue?

Best,

Prof. S. Govindjee

  • Administrator
  • FEAP Guru
  • *****
  • Posts: 1160
Re: compiling PETSc in HPC
« Reply #3 on: August 05, 2017, 09:37:17 AM »
The standard distribution of ver83 is only compatible with much earlier versions of PETSc, though there is a posted update to 3.2p7 on the download site.

FEAP 8.3 is rather old.  I suggest upgrading to version 8.5; that will let you use the current release of PETSc.  If you can not do that, you can follow the PETSc release notes and make the necessary changes to FEAP.  FEAP only uses a limited selection of PETSc commands so it is not too hard to make the updates.

shenrilin

  • Full Member
  • ***
  • Posts: 67
Re: compiling PETSc in HPC
« Reply #4 on: August 05, 2017, 10:07:35 AM »
Dear Prof. Govidjee,

Thanks for your reply.

I'm not sure which subroutines of feap8.3 cite the Petsc, could you give me some more specific instructions such as subroutines list?

Best,

Prof. S. Govindjee

  • Administrator
  • FEAP Guru
  • *****
  • Posts: 1160
Re: compiling PETSc in HPC
« Reply #5 on: August 05, 2017, 02:32:36 PM »
I think only the routines in the directory parfeap (and below) that have the .F extender will have PETSc commands.  If you search for the word "Petsc", "Mat", and "KSP" you should get everything (I think).  There really are not too many.
« Last Edit: August 05, 2017, 02:56:04 PM by Prof. S. Govindjee »