Cbench - Scalable Cluster Benchmarking Code

add undocumented --pokejobstatus option. not really a general use
option at this point. i just needed it for swl post-processing.

# if --pokejobstatus <file> was given, we need to read the file into
# the %pokestatus hash. the purpose of this option is to allow a way
# to override the status (e.g. PASSED, ERROR(FOO), ERROR(BAR)) of jobs
# that are specified in the simple text file. this is mostly useful
# for massive use of Cbench, like in a synthetic workload sustained
# test which will run thousands of jobs, where some jobs need to be
# post processed with extra information correlation . for instance,
# many real world applications run as many simulation timesteps as
# they can fit in before a batch scheduler kills them. then, they
# continue from the last set of restart files. Cbench includes jobs like
# this in testsets from the Openapps tree, i.e. the LAMMPS rhodolong.scaled
# job. the cbench output parser sees these jobs killed by walltime
# as errors since the application does not shutdown normally.
# --pokejobstatus gives the ability to tell Cbench to assign any
# status to any jobid via a text file formatted like:
#
# job1 status1 comment
# job2 status2 comment
#
# each line in the file is a single jobid

this is so olllllllddddd tooo!

updated this Tbird era patch since it is useful
in the sense that it reminds me that HPL was
compiled to use MPI Datatypes on tbird to use
OpenMpi's superiour Datatype engine.

cd opensource/HPLinpack
patch -p2 < stuntmode-debug-tweaks.patch
make

This will create xhpl binary using MPI Datatypes.

this is so ollllllllddd

for longstress_stress.in job tell cbench to use
8.5 hour walltime

updated mpigraph_mpigraph.in job with settings
found useful by Marcuse/John on sky with the
addition of 10x more iterations (so it will run
longer as a sustained test in the default Cbench
setup)

remove the (mostly) pointless usleep, and make the output formatting a little nicer

redoing makefile to be based on downloading the latest
IOR src tarball from sourceforge. get rid of outdated
lustre compiles. use latest IOR version which is IOR-2.10.2

updating ior make setup to be tarball based

Apply the fix of revision 673 to nodehwtest_output_parse.pl, too :).

Ignore bogus 0.000 min values being reported for tests in both statsmode and dplot.