How long should the parallel ptraj test take? I'm at 30+ minutes on 4 cpu's (45+ by the time I finished this...). I'm running on the "bugfix 126" machine (head node for simplicity). Top says 100% CPU use and 0% memory. I just killed it and restarted it, and same behavior now, for about 2 minutes.
19674 installe 25 0 52688 7780 3248 R 100 0.0 33:24.91 ptraj.MPI
19675 installe 25 0 52688 7856 3324 R 100 0.0 33:24.45 ptraj.MPI
19673 installe 25 0 52752 7880 3312 R 100 0.0 33:24.08 ptraj.MPI
19676 installe 25 0 52688 7780 3248 R 100 0.0 33:24.03 ptraj.MPI
The last update to ptraj_comprehensive/ptraj.out was also 30+ minutes ago, and test.mdcrd is empty.
I ran like this:
First try:
export DO_PARALLEL='mpirun -np 4 '
sh test_parallel.sh # not according to instructions, but presumably ok
Cntrl-Z # these three commands are so that
bg # I don't get booted off the server while waiting...
top # hopefully, this isn't an issue
Second try:
./test_parallel.sh &
top
Third try:
./test_parallel.sh
(wait two minutes, check top from a different shell -- same behavior, cntrl-C out)
I used intel compilers that are probably good. I think. It's 10.0.026 for icc and 023 for ifort. Is 026 broken for C? Would be my luck... I'm about to try again with 11.0.083, which we thought might be bad, but isn't looking to be the issue.
$ mpicc -show
/programs/intel/cce/10.0.026/bin/icc -L/opt/hpmpi/lib/linux_amd64 -lhpmpio -lhpmpi -ldl -I/opt/hpmpi/include
$ mpif90 -show
/programs/intel/fce/10.0.023/bin/ifort -L/opt/hpmpi/lib/linux_amd64 -I/opt/hpmpi/include/64 -lhpmpio -lhpmpi -ldl
$ mpif77 -show
/programs/intel/fce/10.0.023/bin/ifort -L/opt/hpmpi/lib/linux_amd64 -I/opt/hpmpi/include/64 -lhpmpio -lhpmpi -ldl
$ echo $LD_LIBRARY_PATH
/opt/hpmpi/lib/linux_amd64:/programs/intel/cce/10.0.026/lib:/programs/intel/fce/10.0.023/lib
$ echo $AMBERHOME
/programs/amber11_AT_20100316_i10.0
$ echo $PATH
/programs/amber11_AT_20100316_i10.0/bin:/opt/pbs/10.0.0.82981/bin (etc.)
Aside -- I assume you are aware of the duplicate executables in bin/ and AmberTools/bin.
The bottom of ptraj_comprehensive says:
==========================================================
[installer.yeats test]$ tail ptraj_comprehensive/ptraj.out
The solvent atoms are :8-1584
The solvent atoms are :8-1584
ved from trajectory: :8-1584
STRIP: 4731 atoms will be removed from trajectory: :8-1584
STRIP: 4731 atoms will be removed from trajectory: :8-1584
STRIP: 4731 atoms will be removed from trajectory: :8-1584
Processing AMBER trajectory file trajectory.mdcrd
==========================================================
:-) Lachele
--
B. Lachele Foley, PhD '92,'02
Assistant Research Scientist
Complex Carbohydrate Research Center, UGA
706-542-0263
lfoley.ccrc.uga.edu
----- Original Message -----
From: case
[mailto:case.biomaps.rutgers.edu]
To: amber-developers.ambermd.org
Sent:
Tue, 16 Mar 2010 12:56:43 -0400
Subject: [AMBER-Developers] First AmberTools
release candidate
> Hi everyone:
>
> You can get the first AmberTools release candidate here:
>
> http://ambermd.org/downloads/AmberTools.16mar10.tar.bz2
>
> Notes:
>
> 1. This unpacks into a new directory scheme. You have to go to
> amber11/AmberTools/src to run "configure". Instructions are in the Users'
> Manual, so don't assume that everything is completely identical to what you
> are used to. (Note that you still set AMBERHOME to .../amber11, as before,
> and
> as the configure script instructs.)
>
> 2. Serial tests should be mostly working; there will be two nab-related
> bad file comparisons, but I am working on that, and they are not real
> errors.
> The only parallel test is for ptraj.MPI, and that should be working.
>
> 3. The AmberTools.pdf file is up-to-date as of 16 Mar. But it still needs
> lots of work.
>
> 4. It looks like everything a user should need is present, but let me know
> if that is not true. Also, be on the lookout for directories or files that
> should not be in the distribution...these are harder to catch, since they
> usually don't cause an errors.
>
> 5. This has been tested with gnu compilers on MacOS and intel 11.1 on
> linux.
> So there are lots of untested platforms out there....I am sure that Lachele
> can break this :-) Please report problems to amber-developers.
>
> ...thanks...dac
>
>
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers
>
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Tue Mar 16 2010 - 14:00:03 PDT
