 |
Here are a few tips for getting started as a new user of the HPCC.
This list of tips and hints is being expanded and revised constantly,
so please check back often.
- how to log in —
Use ssh to log onto the HPCC headnode, deepthought.umd.edu.
To the user, deepthought appears much like any RE Linux platform
campus glue machine.
- qstat vs showq —
There are two qstat
programs on deepthought, one to view printer queues, and one to view torque queues.
Both of them are familiar to a lot of people.
To make sure you run the one that shows the job queues, either enter
torque_qstat
or put
/usr/local/torque/bin at the front of your PATH
in your .path file so you'll get the torque qstat first.
Maui is the scheduler, and sits on top of Torque (PBS).
We recommend that you use Maui's queue command, showq.
- mybalance —
Use mybalance to determine your remaining CPU allocation in seconds
- checkjob —
Use checkjob to find out why your job has been rejected/deferred
- submitting jobs —
Use qsub to submit your jobs.
The HPCC job queues are described
here.
- running MPI jobs —
Add one of the following to your .cshrc.mine file
depending on your MPI preference:
tap -q lam-gnu
tap -q lam-intel
tap -q mpich-gnu
tap -q mpich-intel
tap -q openmpi-intel
tap -q openmpi-gnu
tap -q openmpi-pgi
These add the intel-compiled or gnu-compiled versions of lam, mpich or openmpi to your path.
(Note that the "-q" option runs tap silently, a typical requirement for .cshrc startup situations
though unnecessary for interactive command line use.)
- Additional tools and libraries —
Various additional tools may require that you issue a tap -q command,
so add lines to your .cshrc.mine file for those you regularly use.
For example, if you want the Intel compilers in your path, add:
which also gives you the Intel Math Library, your best bet for highly optimized
versions of things like LAPACK/BLAS.
(We now have the full Intel Cluster Toolkit; click
here
for details.)
Or you can use the reference implementation LAPACK/BLAS libraries
(these are slower than those in our Intel Cluster Toolkit):
For netCDF libraries and tools use:
For Java 6 use:
- where to write your output —
Your HPCC home directory space is limited and on a relatively slow hard drive.
Your home directory space should never be used for large files or heavy I/O
activity by your cluster jobs, since constant/heavy activity or filling up
the available space there adversely affects all users of the cluster.
Please use other available disk space (i.e. /data/dt-raid5, /data/dt-raid10,
/data/dt-vol0, and /data/dt-vol1) for job I/O as described
here.
Please send your questions or comments to hpcc-help@umd.edu.
Coming soon ... a wiki for use by the HPCC user community.
|
|
|
