So, the idea is to prepare a list of hosts in the start_proc_args, which can 
directly be used by ddikick.x.

Please find attached the both start/stop_proc_args. They are meant to be put in 
/usr/sge/cluster - otherwise please check all the directories inside and adjust 
them. All the stuff which I don't want to be destroyed by an update of SGE I 
put in /usr/sge/cluster (like some prolog and epilog scripts). Important is, 
that you compiled GAMESS to use just `rsh`for remote calls (i.e. left it 
untouched), so that SGE's wrapper can catch it and route it to any startup 
method you like. But it's also possible to define in the job script "export 
DDI_RSH=rsh" in case you have something else compiled in.

====
$ qconf -sp gamess
pe_name            gamess
slots              128
user_lists         NONE
xuser_lists        NONE
start_proc_args    /usr/sge/cluster/gamess/startgamess.sh -catch_rsh \
                   $pe_hostfile
stop_proc_args     /usr/sge/cluster/gamess/stopgamess.sh
allocation_rule    $round_robin
control_slaves     TRUE
job_is_first_task  TRUE
urgency_slots      min
accounting_summary FALSE
====

Then you have a jobscript:

====
#!/bin/sh 
#$ -N foobar
#$ -o /home/reuti/err/foobar.stdout_$JOB_ID -e 
/home/reuti/err/foobar.stderr_$JOB_ID
#$ -m ea
#$ -A gamess_serial
export LD_LIBRARY_PATH=/opt/chemsoft/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}
cd $TMPDIR
cp /home/reuti/foobar.in foobar.F05
ln -s /home/reuti/foobar.orb
JOB=foobar
CUSTOM_TMPDIR=/home/reuti/scr
BINARY_LOCATION=/opt/chemsoft/$ARC/GAMESS_OCT012010R3
EXTERNAL_BASISSET=/dev/null
. /home/soft/scripts/subgms_export
unset JOB
unset CUSTOM_TMPDIR
unset BINARY_LOCATION
unset EXTERNAL_BASISSET
rm -f $IRCDATA
rm -f $PUNCH
rm -f $SIMEN
rm -f $SIMCOR
HOSTFILE=`hostname`
/opt/chemsoft/$ARC/GAMESS_OCT012010R3/ddikick.x 
/opt/chemsoft/$ARC/GAMESS_OCT012010R3/gamess.00.x foobar -ddi 1 1 $HOSTFILE 
-scr $TMPDIR < /dev/null > /home/reuti/foobar.out
====

And for parallel:

====
#!/bin/sh 
#$ -N foobar
#$ -o /home/reuti/err/foobar.stdout_$JOB_ID -e 
/home/reuti/err/foobar.stderr_$JOB_ID
#$ -m ea
#$ -A gamess_parallel
#$ -R y
#$ -pe gamess 4
export LD_LIBRARY_PATH=/opt/chemsoft/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}
JOB=foobar
CUSTOM_TMPDIR=/home/reuti/scr
BINARY_LOCATION=/opt/chemsoft/$ARC/GAMESS_OCT012010R3
cd $TMPDIR
cp /home/reuti/foobar.in foobar.F05
. /home/soft/scripts/subgms_export
unset JOB
unset CUSTOM_TMPDIR
unset BINARY_LOCATION
rm -f $IRCDATA
rm -f $PUNCH
rm -f $SIMEN
rm -f $SIMCOR
HOSTFILE=`cat $TMPDIR/machines`
/opt/chemsoft/$ARC/GAMESS_OCT012010R3/ddikick.x 
/opt/chemsoft/$ARC/GAMESS_OCT012010R3/gamess.00.x foobar -ddi $NHOSTS 4 
$HOSTFILE -scr $TMPDIR < /dev/null > /home/reuti/foobar.out
====

The installation of GAMESS then need only:

reuti@foobar:/opt/chemsoft/lx24-amd64> ls GAMESS_OCT012010R3
auxdata  ddikick.x  gamess.00.x
reuti@foobar:/opt/chemsoft/lx24-amd64> ls GAMESS_OCT012010R3/auxdata/
BASES  MCP  QUANPOL  ericfmt.dat


I'll prepare the "subgms_export" now, it will just set all variables which are 
normally set in rungms and post it ASAP.

-- Reuti

PS: Both job scripts are in fact created by a wrapper for the users, but I 
think for a first test it's better to start with a plain job script to avoid an 
additonal layer of trouble. The wrapper I could send you later, when it's 
working in principle.


Am 19.01.2012 um 20:33 schrieb Semion Chernin:

> Please send me qconf -sp gamess and start & stop scripts.
> 
> ----- Original Message -----
> From: Mazouzi <mazo...@gmail.com>
> Date: Thursday, January 19, 2012 17:32
> Subject: Re: [gridengine users] How I can integrate GAMESS under SGE?
> To: Reuti <re...@staff.uni-marburg.de>
> Cc: Semi <s...@bgu.ac.il>, "users@gridengine.org" <users@gridengine.org>
> 
> > Hi,
> > 
> > In our cluster, we use GAMESS with a special PE (gamess) using
> > allocation_rule    2.
> > 
> > The user MUST request an even number of slots ex ( 2*n).
> > 
> > the command line is :
> > 
> > gamess -n $NSLOTS my-job.inp >& my-job.out
> > 
> > 
> > Hope this will help.
> > 
> > On Thu, Jan 19, 2012 at 4:01 PM, Reuti <re...@staff.uni-
> > marburg.de> wrote:
> > 
> > > Hi,
> > >
> > > Am 19.01.2012 um 14:22 schrieb Semi:
> > >
> > > > How I can integrate GAMESS under SGE?
> > > >
> > > > I found some info about this in file called gms, but without pe,
> > > start&stop scripts, example how to use this wrapper.
> > > >
> > > > #    SGE job submission:
> > > > #    A 'parallel environment' named 'ddi' was 
> > set up on ISU's cluster,
> > > > #    this SGE prolog file creates the SGE 
> > directory $TMPDIR on every
> > > node,
> > > > #    and this epilog script erases $TMPDIR, 
> > to be sure the scratch disk
> > > is
> > > > #    always cleaned up, and to remove dead 
> > semaphores.> > #
> > > > #    SGE command 'qconf -sp ddi' shows the 
> > details of this environment,
> > > > #    including pathnames to prolog/epilog 
> > scripts.  Also, 'qconf -spl'.
> > > > #    Other useful SGE commands: 'qconf -sc' 
> > shows config for resources.
> > > > #
> > > > #    Mirabile dictu!  SGE allows you to 
> > pass args to a job script by
> > > > #    just placing them behind the script 
> > name.  In all my living days,
> > > > #    I've never seen a batch program that 
> > permitted this.  Glory be!
> > > > #
> > > > if ($SCHED == SGE) then
> > > >   qsub -cwd -o $LOGFILE -j yes -pe ddi $NNODES -N 
> > $JOBNAME> $SGE_RESOURCES \
> > > 
> > >              ~/scr/$JOB.script $JOB $VERNO $NCPUS $PPN
> > > > endif
> > >
> > > I think this is something set up by your site already. I've 
> > never seen it
> > > before.
> > >
> > > The rungms they provide with GAMESS seems targeting in the 
> > first place the
> > > clusters of the author, what is understandable. But I would 
> > have really
> > > liked that they provide a file which you have to source to set 
> > all their
> > > environment variables and then just just issue ddikik.x with 
> > the correct
> > > parameters and you are done.
> > >
> > > I could only send you my scripts to set it up from scratch. 
> > The export
> > > list of variables I would have to refresh, as I have it for 
> > the October 01,
> > > 2010 R3 version only right now.
> > >
> > > -- Reuti
> > > _______________________________________________
> > > users mailing list
> > > users@gridengine.org
> > > https://gridengine.org/mailman/listinfo/users
> > >
> >
> ‎

Attachment: gamess.tgz
Description: Binary data

_______________________________________________
users mailing list
users@gridengine.org
https://gridengine.org/mailman/listinfo/users

Reply via email to