I tried openMPI 1.6.5 and ran the tests, again. They ran but with one error
that I do not understand.
cd nfe && ./run-pmemd-parallel.sh
/bin/sh: ./run-pmemd-parallel.sh: Permission denied
make[2]: [test.parallel.pmemd.nfe] Error 126 (ignored)
export TESTsander='../../../bin/pmemd.MPI'; cd emil/emil_pmemd_gbsa &&
./Run.emil EMIL
DO_PARALLEL is set: mpirun -np 2
diffing mdout.emil.0.0.save with mdout.emil.0.0
PASSED
I've attached the test_amber_parallel log. Is the error a problem or can it
be ignored?
I'll try some other variations on openMPI 1.10 and 2.0 (for example, the
1.10.5 I tried had the cuda capability enabled) and see if they work.
Thanks for the help.
Jim
-----Original Message-----
From: David Case [mailto:david.case.rutgers.edu]
Sent: Saturday, January 28, 2017 10:01 AM
To: jimkress_58.kressworks.org; AMBER Mailing List <amber.ambermd.org>
Subject: Re: [AMBER] amber installation test question
On Fri, Jan 27, 2017, James Kress wrote:
> I installed Amber16 using OpenMPI 1.10.5 on RHEL 7.3 Workstation with
> no errors. I then ran the MPI tests with np=2 ( using export >
> DO_PARALLEL="mpirun -np 2").
>
> All went well until the Run.dhfr.noboxinfo test which hung.
>
> Is this expected?
No. Is the error with sander or with pmemd (or both)? Do you see the
"peek_ewald" error message in the mdout.dhfr.noboxinfo file?
This test exits with an error message, and it may be that the master process
is exiting but the second thread is not seeing the signal, and is therefore
hanging (even though mpi_abort is being called with MPI_COMM_WORLD). This
sort of behavior could be a serious problem in a production code.
I think that most/all Amber testing with openmpi has used releases prior to
1.10, so it's possible that there is something funny going on. We'll have
to see if someone here can reproduce the problem.
...thanks for the report....dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sat Jan 28 2017 - 12:30:02 PST
