TWIST Linux Cluster
Contact the system administrators:
Jump to...
- Cluster information
- OpenPBS information
- Acrobar Reader information
- OpenOffice information
- Information about printing
Cluster information:
- Nodes-
-
- OpenPBS-
-
- Users-
-
- Data disks-
-
- Misc-
-
Reserved Disks
Certain data disks are reserved for personnal work. Please DO NOT
use those disks without permission from the primary user. The
following list was updated March 15th, 2005 :
-
data13: Blair
-
data15: Jim
-
tw01y: Rob
OpenPBS Batch system
OpenPBS is a replacement for the obsolete DQS batch system.
It is under active development by the folks
at www.openpbs.org and
is supposed to be an overall improvement over DQS.
In TWIST setup, it works together with the Maui scheduler
(www.supercluster.org),
which actually decides when to run a job and what machine to use.
OpenPBS is installed on all nodes of the twist cluster and
you can start using it now:
-
use pbsrun -N jobname "your_command
your_args... >& logfile" to run jobs,
for example pbsrun -Ntest "echo Hello, world >&test.txt". See
"pbsrun -h" for usage information.
-
use the pbs qsub command to submit pbs jobs directly.
-
use xpbs and xpbsmon to see the status of the queues,
jobs and nodes. In "xpbs", go to the "preferences" window,
enter "e614db" in the "server hosts" box, select "ok" to close
the preferences, select "close" to exit xpbs and then "yes" to make
it save the new preferences. In "xpbsmon", go to
the "pref..." window,
enter "e614db" into the "server host" box, click "add", then
close "xpbsmon" to make it save the new preferences.
You may need to click on e614db... under Server, than
on workq under Queue to get jobs displayed.
-
Use /twist/local/maui/bin/checkjob JOBNUM to see what the
scheduler thinks about your job. You can also try other
commands from the /twist/local/maui/bin/
directory. A command line alternative to looking at jobs in
xpbs is qstat and qstat -f JOBNUM.
-
read man pbs, man qsub, etc.
Some man pages may be missing on certain computers. If you can't
find a man page, try again on e614db, which usually has
the full set.
Special TWIST-specific features:
-
to request running on the tw machines use qsub -l nodes=1:twcluster
or pbsrun -o "-l nodes=1:twcluster" ...
-
to request running on the machine with disk /twist/dataXXX use qsub -l nodes=1:dataXXX
or
pbsrun -o "-l nodes=1:dataXXX" .... To request the /twist/twXX disks, use "nodes=1:twXX".
-
to request running on faster machines, use qsub -l nodes=1:cpuXXX
or
pbsrun -o "-l nodes=1:cpuXXX" ...,
where the currently defined cpuXXX values are: cpu2200, cpu1600, cpu1000, and cpu900.
The use of the option prevents a job from going to any machine
having a smaller cpu value. Note that jobs go to the fastest available computer by
default. So the effect of using this option is that your job
will wait till a fast cpu becomes available instead of
immediately starting on a slower machine.
-
to run on a specific node XXX, use qsub -l host=XXX
or
pbsrun -o "-l host=XXX" ...
-
to wait till a raw data file becomes available on disk before
starting a job, use qsub -l datarun=XXX
or pbsrun -o "-l datarun=XXX" ...,
where XXX is the run number.
-
you can use datalocal=XXX instead of the datarun above.
This puts an additional restrictions: the job can run only
on a node having run XXX is on its local disk.
Be careful to not over-restrict your jobs. A use of the
datalocal attribute together with the nodes can
prevent your job from starting, if the requested run happens to
sit on a machine not belonging to the selected node set.
More information about OpenPBS is available from Konstantin's OpenPBS page, from www.openpbs.org and from NASA.
AcroRead:
Acrobat reader is available from TRIUMFCS as /triumfcs/linux/bin/acroread
OpenOffice:
To use OpenOffice. do this:
1) run the installation script "/triumfcs/linux/OpenOffice.org1.0/setup".
a) leave the registration page empty
b) select the "Workstation installation"
c) use the default installation path "$HOME/OpenOffice.org1.0"
2) to run open office use the command:
$HOME/OpenOffice.org1.0/soffice
3) to print, use "print to file"- the open office printer selection
does not like the triumf printing setup.
Printing:
You can print using pdq or lpr. pdq offers
better printer control (double vs single sided, paper vs transparency)
while lpr can print to all triumf printers via the triumfcs
print server (here is the list of printers).
How to use pdq
The PDQ printing system offers
simpler control of printer features (single or double sided, tray selection)
and does not depend on any daemons and queues. To print things,
it communicates with the printer directly.
To use pdq, add /home/olchansk/packages/pdq/bin into your path and use these pdq
commands:
- xpdq & - start the gui program to look at and control print job.
- pdq -Pqueuename filename - send a file to the printer.
- pdq -h - see the list of available print queues.
- pdq -Pqueuename -h - see the options for a given print queue.
Printers and queues (8-Aug-2001)
- las74 - Trailer U
- pdq -Plas74 - normal print queue
- pdq -Plas74t - use the manual feed tray (for transparencies)
- pdq -Plas74d - use duplex mode, from the bottom (paper) tray
- lpr -Plas74 - lpr queue via the triumfcs print server
- las75 - M13/TWIST/E614 Counting House
- lpr -Plas75local - *on midtwist only* normal print queue
- pdq -Plas75 - normal print queue
- pdq -Plas75t - use the manual feed tray (for transparencies)
- pdq -Plas75d - use duplex mode, from the bottom (paper) tray
- lpr -Plas75 - lpr queue via the triumfcs print server
Obsolete information---
DQS Batch system: (dismantled)
Misc...
Konstantin Olchanski
Last modified: Fri Dec 6 15:57:14 PST 2002