[gmx-users] Low cpu utilization
Mark Abraham
mark.j.abraham at gmail.com
Sun Oct 16 22:48:17 CEST 2016
Hi,
GROMACS is compute bound when it is not network bound, but the output of ps
is barely informative. Looking inside md.log file for the helpful
diagnostics mdrun prints is a fine start. Also, do check out
http://manual.gromacs.org/documentation/2016/user-guide/mdrun-performance.html
for
basics. To know how much hardware makes sense to use for a given
simulation, you need to look at performance on a single core and on all
cores of a node before worrying about adding a second node.
Mark
On Sun, Oct 16, 2016 at 8:24 PM Parvez Mh <parvezmh89 at gmail.com> wrote:
> Not sure, I do not use OpenMPI, you could try compile following simple mpi
> program and run, see if you get proper node allocation .
>
> --Masrul
>
> On Sun, Oct 16, 2016 at 1:15 PM, Mahmood Naderan <mahmood.nt at gmail.com>
> wrote:
>
> > Well that is provided by nodes=2:ppn=10 in the PBS script.
> >
> > Regards,
> > Mahmood
> >
> >
> >
> > On Sun, Oct 16, 2016 at 9:26 PM, Parvez Mh <parvezmh89 at gmail.com> wrote:
> >
> > > Hi,
> > >
> > > Where is -np option in mpirun ?
> > >
> > > --Masrul
> > >
> > > On Sun, Oct 16, 2016 at 12:45 PM, Mahmood Naderan <
> mahmood.nt at gmail.com>
> > > wrote:
> > >
> > > > Hi,
> > > > A PBS script for a gromacs job has been submitted with the following
> > > > content:
> > > >
> > > > #!/bin/bash
> > > > #PBS -V
> > > > #PBS -q default
> > > > #PBS -j oe
> > > > #PBS -l nodes=2:ppn=10
> > > > #PBS -N LPN
> > > > #PBS -o /home/dayer/LPN/mdout.out
> > > > cd $PBS_O_WORKDIR
> > > > mpirun gromacs-5.1/bin/mdrun_mpi -v
> > > >
> > > >
> > > > As I ssh'ed to the nodes and saw mdrun_mpi process, I noticed that
> the
> > > cpu
> > > > utilization is not good enough!
> > > >
> > > >
> > > > [root at compute-0-1 ~]# ps aux | grep mdrun_mpi
> > > > dayer 7552 64.1 0.0 199224 21300 ? RNl Oct15 1213:39
> > > > gromacs-5.1/bin/mdrun_mpi -v
> > > > dayer 7553 56.8 0.0 201524 23044 ? RNl Oct15 1074:47
> > > > gromacs-5.1/bin/mdrun_mpi -v
> > > > dayer 7554 64.1 0.0 201112 22364 ? RNl Oct15 1213:25
> > > > gromacs-5.1/bin/mdrun_mpi -v
> > > > dayer 7555 56.5 0.0 198336 20408 ? RNl Oct15 1070:17
> > > > gromacs-5.1/bin/mdrun_mpi -v
> > > > dayer 7556 64.3 0.0 225796 48436 ? RNl Oct15 1217:35
> > > > gromacs-5.1/bin/mdrun_mpi -v
> > > > dayer 7557 56.1 0.0 198444 20404 ? RNl Oct15 1062:26
> > > > gromacs-5.1/bin/mdrun_mpi -v
> > > > dayer 7558 63.4 0.0 198996 20848 ? RNl Oct15 1199:05
> > > > gromacs-5.1/bin/mdrun_mpi -v
> > > > dayer 7562 56.2 0.0 197912 19736 ? RNl Oct15 1062:57
> > > > gromacs-5.1/bin/mdrun_mpi -v
> > > > dayer 7565 63.1 0.0 197008 19208 ? RNl Oct15 1194:51
> > > > gromacs-5.1/bin/mdrun_mpi -v
> > > > dayer 7569 56.7 0.0 227904 50584 ? RNl Oct15 1072:33
> > > > gromacs-5.1/bin/mdrun_mpi -v
> > > >
> > > >
> > > >
> > > > [root at compute-0-3 ~]# ps aux | grep mdrun_mpi
> > > > dayer 1735 0.0 0.0 299192 4692 ? Sl Oct15 0:03
> mpirun
> > > > gromacs-5.1/bin/mdrun_mpi -v
> > > > dayer 1740 9.5 0.0 209692 29224 ? RNl Oct15 180:09
> > > > gromacs-5.1/bin/mdrun_mpi -v
> > > > dayer 1741 9.6 0.0 200948 22784 ? RNl Oct15 183:21
> > > > gromacs-5.1/bin/mdrun_mpi -v
> > > > dayer 1742 9.3 0.0 200256 21980 ? RNl Oct15 177:28
> > > > gromacs-5.1/bin/mdrun_mpi -v
> > > > dayer 1743 9.5 0.0 197672 19100 ? RNl Oct15 180:01
> > > > gromacs-5.1/bin/mdrun_mpi -v
> > > > dayer 1744 9.6 0.0 228208 50920 ? RNl Oct15 183:07
> > > > gromacs-5.1/bin/mdrun_mpi -v
> > > > dayer 1746 9.3 0.0 199144 20588 ? RNl Oct15 176:24
> > > > gromacs-5.1/bin/mdrun_mpi -v
> > > > dayer 1749 9.5 0.0 201496 23156 ? RNl Oct15 180:25
> > > > gromacs-5.1/bin/mdrun_mpi -v
> > > > dayer 1751 9.1 0.0 200916 22884 ? RNl Oct15 173:13
> > > > gromacs-5.1/bin/mdrun_mpi -v
> > > > dayer 1755 9.3 0.0 198744 20616 ? RNl Oct15 176:49
> > > > gromacs-5.1/bin/mdrun_mpi -v
> > > > dayer 1758 9.2 0.0 226792 49460 ? RNl Oct15 174:12
> > > > gromacs-5.1/bin/mdrun_mpi -v
> > > >
> > > >
> > > >
> > > > Please note that the third column is the cpu utilization.
> > > > Gromacs is a compute intensive application, so there is little IO or
> > > > something else for that.
> > > >
> > > >
> > > > Please also note that in compute-0-3 the first process is "mpirun
> > > > gromacs-5.1...." while the others are only "gromacs-5.1...."
> > > >
> > > >
> > > > Any idea is welcomed.
> > > >
> > > > Regards,
> > > > Mahmood
> > > > --
> > > > Gromacs Users mailing list
> > > >
> > > > * Please search the archive at http://www.gromacs.org/
> > > > Support/Mailing_Lists/GMX-Users_List before posting!
> > > >
> > > > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
> > > >
> > > > * For (un)subscribe requests visit
> > > > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users
> or
> > > > send a mail to gmx-users-request at gromacs.org.
> > > >
> > > --
> > > Gromacs Users mailing list
> > >
> > > * Please search the archive at http://www.gromacs.org/
> > > Support/Mailing_Lists/GMX-Users_List before posting!
> > >
> > > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
> > >
> > > * For (un)subscribe requests visit
> > > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
> > > send a mail to gmx-users-request at gromacs.org.
> > >
> > --
> > Gromacs Users mailing list
> >
> > * Please search the archive at http://www.gromacs.org/
> > Support/Mailing_Lists/GMX-Users_List before posting!
> >
> > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
> >
> > * For (un)subscribe requests visit
> > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
> > send a mail to gmx-users-request at gromacs.org.
> >
> --
> Gromacs Users mailing list
>
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before
> posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
> send a mail to gmx-users-request at gromacs.org.
More information about the gromacs.org_gmx-users
mailing list