Re: NAMD 2.9 install on CLuster mpirun/charmrun with ibverb version ?

From: Axel Kohlmeyer (akohlmey_at_gmail.com)
Date: Thu Nov 14 2013 - 06:52:52 CST

On Thu, Nov 14, 2013 at 1:43 PM, Norman Geist
<norman.geist_at_uni-greifswald.de> wrote:
> If you need mpirun or charmrun depends on if you compiled namd against mpi
> or charm++. If you just downloaded a precompiled build, itís likely without
> mpi and charmrun needs to be used. If you need to add one of them in general

*but* ... charmrun can use mpirun in turn to launch the job. on our
cluster, mpirun is compiled with torque support, so no hostfile needs
to be provided at all and something like this does the trick.

        ${NAMD_HOME}/charmrun-ibverbs ++p ${NUM_PROCS} ++mpiexec
++remote-shell ${MPI_BIN}/mpirun \
                ${NAMD_HOME}/namd2-ibverbs "$@"

this is using the pre-compiled ibverbs executable, of course.

axel.

> depends on if you use more than one physical node. For multiple nodes you
> will of course need mpirun/charmrun. But, in case of charmrun the
> machinefile (nodelist) has a different format:
>
>
>
> group main
>
> host node1
>
> host node1
>
> [Ö]
>
>
>
> For mpirun simply:
>
>
>
> node1
>
> node2
>
> [Ö]
>
>
>
> For local parallel jobs you need at least a multicore build of namd which
> has shared memory parallelism.
>
>
>
> Norman Geist.
>
>
>
> Von: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] Im Auftrag
> von Nicolas Floquet
> Gesendet: Donnerstag, 14. November 2013 12:18
> An: namd-l_at_ks.uiuc.edu
> Betreff: namd-l: NAMD 2.9 install on CLuster mpirun/charmrun with ibverb
> version ?
>
>
>
> Dear all ,
>
>
>
> I try to install NAMD on a cluster with infiniband and tested different
> versions of NAMD
>
>
>
> My submission script is as follows (in which I tested different solutions):
>
>
>
> #!/bin/sh
>
>
>
> # @ job_name = TMD1
>
> # @ output = $(job_name).out
>
> # @ error = $(job_name).err
>
> # @ environment = COPY_ALL
>
> # @ class = specialIntel
>
> # @ account_no = RCPGs2
>
> # @ job_type = mpich
>
> # @ notify_user = nicolas.floquet_at_univ-montp1.fr
>
> # @ node = 10
>
> # @ total_tasks = 120
>
> # @ environment = COPY_ALL
>
> # @ wall_clock_limit = 36:00:10,36:00:01
>
> # @ queue
>
>
>
> source
> /opt/cluster/softs/gcc-4.6.x-soft/system/module/3.2.10/Modules/3.2.10/init/sh
>
> module load hpclr-wrapper gcc-4.6.4 openmpi-1.6.5-gcc
>
> export CPPFLAGS=-I/opt/cluster/gcc-soft/fftw/3.3.3-shared-float/include
>
> export LDFLAGS=-L/opt/cluster/gcc-soft/fftw/3.3.3-shared-float/lib
>
>
>
> /opt/cluster/softs/gcc-4.6.x-soft/system/openmpi/1.6.5/bin/mpirun -x
> LD_LIBRARY_PATH -np $LOADL_TOTAL_TASKS -machinefile $LOADL_HOSTFILE
>
> /opt/cluster/gcc-soft/namd/2.9/NAMD_2.9_Linux-x86_64-ibverbs/namd2
> /scratch/floquetn/NEURO/TMD2/POINT3/tmd1.namd
>
>
>
> #/opt/cluster/softs/gcc-4.6.x-soft/system/openmpi/1.6.5/bin/mpirun -x
> LD_LIBRARY_PATH -np $LOADL_TOTAL_TASKS -machinefile $LOADL_HOSTFILE
>
> /opt/cluster/gcc-soft/namd/2.9/NAMD_2.9_Linux-x86_64/namd2
> /scratch/floquetn/NEURO/TMD2/POINT3/tmd1.namd
>
>
>
> #/opt/cluster/gcc-soft/namd/2.9/NAMD_2.9_Linux-x86_64/charmrun +p
> $LOADL_TOTAL_TASKS ++nodelist $LOADL_HOSTFILE /opt/cluster/gcc-soft/namd/
>
> 2.9/NAMD_2.9_Linux-x86_64/namd2
> /scratch/floquetn/NEURO/TMD2/POINT3/tmd1.namd
>
>
>
>
>
> Which solution is the best one ?
>
> The one selected seems to launch but not in parallel ? do I need to use
> charmrun with the namd2 executable for iverbs ? do I need to include mpirun
> as I did?
>
>
>
> Thank you in advance for your help.
>
>
>
> Nicolas Floquet

-- 
Dr. Axel Kohlmeyer  akohlmey_at_gmail.com  http://goo.gl/1wk0
International Centre for Theoretical Physics, Trieste. Italy.

This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:21:54 CST