The AMDeC cluster is set up as a ROCKS 3.2 Cluster - It is a Linux-based 88 CPU cluster for general purpose and parallel computing. It contains 44 nodes; each node contains two 2.4 GHz Intel Pentium 4 XeonCPUs plus 2 GB of memory and a 40 GB harddisk. The operating system is Red Hat Enterprise Linux 3 in a ROCKS configuration. The nodes are interconnected with 2 Gbit/sec low-latency Myrinet networking. GNU C, C++, and Fortran compilers are available. MPI programming is supported.

PLEASE NOTE:

All jobs should be submitted via Sun Grid Engine (SGE) to ensure proper load
balancing and minimize user load on the frontend. For further information,
please see the man pages for the following SGE commands:

qsub qdel qstat

If you are a cluster user, you now have a special home directory when you log
into the cluster. On other computers, this directory is located under
/fsnode1/users/USERNAME. On the cluster, it is /users/USERNAME. If you need
data or programs to run from your home directory on the cluster, you will need
to copy them to your cluster home directory on adgate or another amdec computer.
Since we are restricing the amount of space you can use in the cluster home
directory, you should only copy over essential data for your cluster projects.
These new cluster home directories will be backed up, so data will be secure,
unlike /scratch. These new home directories also have high speed I/O like
/scratch. You can now balance your I/O requirements between your home
directory, /scratch, and /tmp (note that /tmp is local to the processor box,
it cannot be seen by compute nodes outside the local processor pair). Please
note that /databases on the cluster nodes is also different than on other amdec
computers, If you need access to reference databases, you need to arrange with
the systems management team to keep a working copy of your databases available
on the cluster nodes. Note also that I/O to /databases on the cluster competes
with I/O with you home directory as they reside on the same fileserver.

PLEASE NOTE. Your cluster home directory is also mounted on all other amdec
computers, so there is no need to keep redundant copies of data and programes
in both locations. In addition, you will be charged for space used in each
location.

One important aspect of good cluster programming is to balance your I/O between
drive resources, reading and writing to different places when possible. Your
cluster home directory and /scratch reside on a filservers with much better I/O
performance than the fileserver where your regular home directory resides. To
use /scratch space create a directory named after your USERNAME under /scratch
using the command "mkdir /scratch/$USER", and work there. Please target
temporary data files to be written to /scratch or /tmp, and important results
to your cluster home directory. Also, please clean up /tmp on progarm exit, and
/scratch after your batch job is finished.

PLEASE NOTE, /scratch and /tmp is NOT BACKED UP and data is subject to total
loss. Please remember to move any critical files (code, scripts, results) to
your cluster home directory and to delete your files from /scratch once your
calculations are complete. /scratch will periodically be purged to ensure
that there is sufficient temporary scratch space available.