[GE users] GAMESS cascading startup

reuti reuti at staff.uni-marburg.de
Thu Jul 30 13:24:37 BST 2009


users using the quantum chemistry package GAMESS might have noticed  
that a Tight Integration is working for a parallel startup only when  
you use up to two nodes, but not more (at least when your cluster is  
free of rsh and ssh). Reason is the cascading startup of the process,  
means that e.g. for a 4 node job the master node of the parallel job  
will issue only two qrsh's, and one of the two slaves will start the  
final fourth process. To get even with this setup a Tight  
Integration, the rsh-warpper must be accessible also on the slave  
nodes. As SGE isn't checking from which node the "qrsh -inherit" is  
originating, we need just a suitable link in the $TMPDIR of the  
slaves with a slight modification of 5 lines of the rsh-wrapper:

       echo $SGE_ROOT/bin/$ARC/qrsh -V -inherit -nostdin $rhost "ln - 
s $SGE_ROOT/cluster/gamess/rsh $TMPDIR ; exec $cmd"
       exec $SGE_ROOT/bin/$ARC/qrsh -V -inherit -nostdin $rhost "ln - 
s $SGE_ROOT/cluster/gamess/rsh $TMPDIR ; exec $cmd"
       echo $SGE_ROOT/bin/$ARC/qrsh -V -inherit $rhost "ln -s  
$SGE_ROOT/cluster/gamess/rsh $TMPDIR ; exec $cmd"
       exec $SGE_ROOT/bin/$ARC/qrsh -V -inherit $rhost "ln -s  
$SGE_ROOT/cluster/gamess/rsh $TMPDIR ; exec $cmd"

I attach the rsh-wrapper here, but remember you will hve to adjust  
the path to the installation of your parallel startup.

-- Reuti


To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].

    [ Part 2, "rsh"  Application/OCTET-STREAM (Name: "rsh") 3.7 KB. ]
    [ Unable to print this part. ]

More information about the gridengine-users mailing list