Presently the working directory for jobs where the sandbox
is fetched to is in /home/wpsix001
Since it is the pbs job that does this fetching this
can be done on the WNs local disk instead.
Obviously this makes a lot of sense if all your nodes are
identical so you have just as much space on the WNs to
share amongst 1 or 2 jobs rather than every user.
Also network trafic is reduced. Since doing this here the
CE's harddisk light is no longer on a constant red.
You problably do not want to do this if you are short
of space on the WN's disk of course.
The file to be replaced is
/opt/globus/libexec/globus-script-pbs-submit
on the CE.
This is the script that builds a wrapper script for the incoming
job and pipes that into PBS.
I have attached the replacement file.
The extra lines in there are.
#SMT next line
grami_directory=/tmp/$USER$grami_directory
and
#SMT next 3 lines
echo "if [ ! -d $grami_directory ]; then" >> $PBS_JOB_SCRIPT
echo " /bin/mkdir -p $grami_directory" >> $PBS_JOB_SCRIPT
echo "fi" >> $PBS_JOB_SCRIPT
The default tmpwatch is for /tmp is 240 hours and
that must not be shorter than your wall_max_times for
any queue.
Steve
Steve
--
Steve Traylen
[log in to unmask]
http://www.gridpp.ac.uk/
|