COS/COE Computing Cluster

The College of Science (COS), and the College of Engineering (COE), are working together to provide High Performance Computing to researchers in both colleges.  The COS has purchased 4 dual opteron server class machines, each with 10Gig of RAM.  These machines are seen by the cluster as 8 64-bit processors, each with 5 Gig of RAM.  Other reasearchers have also purchased 4 more of these server class machines, giving the COS 8 64-bit dual-opteron cluster nodes.

In addition to the nodes which are owned by the COS, any idle nodes in the cluster which have been purchased by COE researchers are available to COS researchers (when not in use).  COS submitted jobs are given a lower priority on COE owned nodes, and COE submitted jobs are given a lower priority on COS owned nodes.  This relationship allows both colleges to get the most processing utilization from the money invested.

If you would like an account on this research cluster, send an e-mail to research@science.oregonstate.edu , include in the e-mail your ONID userid and OSU ID # .

The COE website provides documentation on how to submit jobs to the cluster.  This documentation should enable you to setup your submission scripts and submit both serial and parallel jobs.  The one difference to be noted though, is that the submit queue name for the 64-bit nodes is cos for the COS queue, instead of simply amd64, but we can submit jobs to the em64t, i386 and amd64-low queues as well.
NOTE: Using the cos queue restricts your job to only COS owned nodes, where as the amd64-low queue will run the job on the first available node.

The cluster has support for both mpich version 1 and version 2 libraries (make sure you setup your system environment for the proper version after you login to the submit host):
  for mpich v1 source /usr/local/apps/sge/settings.csh
  for mpich v2 source /usr/local/apps/sge/settings-mpich2.csh

To connect to the 64-bit submit machine, ssh to : submit-amd64-01.hpc.engr.oregonstate.edu

Disk Storage

All accounts on the cluster are given 1Gig of home directory space.  There also exists a 50Gig "scratch" partition for use to store data which are needed by the jobs which you submit to the cluster.  This storage area is not a permanent location for storing your files, and files stored there can be deleted by the SysAdmin, without warning.

Using Matlab

Some MPI functions have been written to allow the use of Matlab across multiple nodes within the cluster.  To setup your account/environment to take advantage of these MPI functions, please refer to the MPI Matlab documentation. 

Using R

Using R on the cluster for simulations, or other applications where there is much looping, can be handled very easily with the use of the Array options on the beowulf cluster. Please refer to the Beowulf Array documentation.  If you need to use R libraries which aren't currently installed on the cluster, please read the page How to use R Libraries which aren't installed on the cluster