# Sample script to run a PBS job (referrer: pbs.html)
# This script shows several issues that need to be addressed when your
# job is more complicated.  You can use the "save as" feature of your
# browser to make a copy of this file, and then edit it as you need
# (removing most of the comments).

# Specify the shell for PBS (you could also use /bin/csh with different
# syntax).
#PBS -S /bin/sh

# Specify the queue explicitly.  The default isn't working yet.
#PBS -q sixpac@bud

# Specify the CPU time limit in seconds (required)
#PBS -l cput=3600

# It's a good idea to announce when and on which machine your job is
# running.
echo Job started on `uname -n` at `date`

# This script is going to work with two directories.  We'll make
# symbolic names for them.
# This is the name of the directory containing permanent data and
# programs. (Fill in the actual name here.  Use $HOME for your Mathnet
# home directory if that's where the permanent files are.)
h=$HOME/myproject

# This directory will contain large files to which high speed I/O is
# done. You want to put it on the local machine.  Local I/O is much
# faster than I/O via NFS to another machine.  Use $USER for your
# loginID to keep your files separate from other users.  You may make
# subdirectories if convenient.
s=/scr/$USER/flow_regime

# If some accident caused two jobs to run at the same time on the same
# machine, or if the previous job crashed and the results have not been
# saved properly, you want to exit.  The script creates this lock file.
# If it's still there, something went wrong.
if [ -f $s/itsrunning ] ; then
    echo Exiting because a previous job is using the directory
    cat $s/itsrunning
    exit 4
fi

# Since we have to rebuild the scratch directory anyway, let's just
# remove the whole thing first.  No problem if the directory never
# existed before.  See the end of the script for suggestions how to
# save important result files.
rm -rf $s

# Make the scratch directory (and containing directories), change to it,
# and populate it with needed files, especially the lock file.
mkdir -p $s
cd $s
echo The job is running, `date` > $s/itsrunning

# If the program needs to read a short file with a fixed name for its
# input parameters, an easy way to set that up is with a symbolic link
# to a pre-existing file in the permanent directory.  You could also
# just copy the file.  The most elegant solution is for the program to
# read its parameters from standard input, but let's assume it's not
# elegant.
ln -s $h/flow_regime_3 paramfile

# Here's an alternative way that keeps the parameters in this script.
cat - <<EOF > miscparams
xstart	1.000
xend	1.650
xdelta	5.0e-3
EOF

# Now the program runs.  PBS captures standard out and standard error.
# The program can rapidly write its large scratch files in the current
# directory, which is on the local machine.  It's generally reasonable
# to have the program file in the permanent directory.  If you run the
# same project on different architectures (i386-SunOS, sparc-SunOS,
# i386-Linux), you need to have separate executable files for each, with
# distinguishing names.
$h/program.i386-SunOS

# Different programs produce output in different ways and this section will
# have to be adapted to your particular needs.  Here's how to append a short
# summary file to a logfile in the permanent directory.
cat summaryout >> $h/logfile

# Files in your permanent directory have to have different names.  Here's a
# way to differentiate them using the hostname, date and time.  Remember to
# not exceed your disc quota.
mv flowmap $h/flowmap.`uname -n`.`date +%Y-%m-%d.%H:%M:%S`

# Be nice, get rid of your giant files that aren't needed any more.  (Fill in
# the names of the scratch files that the program writes.)
rm -f state_cache flow_phase_A flow_phase_B

# All done, remove the lock file allowing this directory to be used by the
# next instance of your job.
rm itsrunning