[molpro-user] MPPX Computations Won't Finish
Andy May
MayAJ1 at cardiff.ac.uk
Mon Jul 11 13:16:25 BST 2011
Harvey,
I don't see this problem, but I suspect it depends upon MPI flavour,
compilers etc., so could you send the details of the build, specifically:
- is this with or without GA? If using GA, can you send the details of
how you built GA.
- what flavour and version of MPI library are you using?
- Can you give details of how you configured Molpro and send the CONFIG
file and a full failing output file.
Best wishes,
Andy
On 08/07/11 16:18, Harvey Wasserman wrote:
> Synopsis: MPPX-compiled executable never finishes when run with more
> than one processor.
>
> I recently installed the MPPX-compiled binary from source code on a
> Nehalem Linux IB cluster on which
> the MPP version has been running happily for some time. I observed the
> following, using the simple
> h2.com sample input file:
>
> 1. Run molpro -n 1 h2.com: works
> 2. Run molpro -n 8 h2.com: works (other values of n work, too)
> 3. Run molprox -n 1 h2.com: works. This is the MPPX version.
> 4. Run molprox -n #, where # is any value greater than 1, and what
> happens is that the computation
> finishes but never returns from the program. In other words, it hangs
> indefinitely after printing the final
> result. I ran it with -v (verbose), and the last command executed is
> the mpirun command and the last info
> in the output file is
>
> HF-SCF
> -1.12870009
> **********************************************************************************************************************************
> Variable memory released
>
> Fortran compiler diagnostic
> ***************************
>
> integer size 8
> double precision size 8
> real size 8
> logical size 8
>
> diagnostic completed successfully
>
> but mpirun never returns, never comes back to the shell. Is this known?
>
> Thanks.
> _______________________________________________
> Molpro-user mailing list
> Molpro-user at molpro.net
> http://www.molpro.net/mailman/listinfo/molpro-user
More information about the Molpro-user
mailing list