DEPARTMENT OF MECHANICAL ENGINEERING's
BEOWULF PARALLEL COMPUTER CLUSTERS:


The Department of Mechanical Engineering operates several parallel computing clusters for research purposes. Each of the clusters is available to all Mechanical Engineering faculty and graduate students for appropriate parallel based research computing (contact Dr. Richard Miller (rm@clemson.edu) for access).

The most recent addition is a rack based cluster added in August 2006:

This cluster consists of 16 dual processor nodes with each node containing two dual core 1.8 GHz AMD Opteron processors and 8 GB of memory. The system therefore has a total of 32 processors (64 processor cores), 128 GB of system memory, and more than 1 TB of hard drive storage. Internal communication is via a gigabit ethernet network. PathScale C and Fortran compilers optimized for the AMD Opteron architecture are installed. This cluster was purchased through TeamHPC .

The Department also operates a PC based parallel computing cluster originally constructed in 2000, with additions made each of the following two years:

This system is a distributed memory platform consisting of 48 processors contained in 24 dual processor PC boxes. The processors vary from 700 MHz to 1.28 GHz Pentium IIIs. Each slave node contains 1 GB of memory. Internal communication is via a high speed (1.28 gigabit/sec) low latency Myrinet network. The system is comprised of two sub-clusters: a primary 40 processor cluster, and a test bed 8 processor cluster.

The Beowulf Accounts and Operating Policy is posted on the Department of Mechanical Engineering's web site at: http://www.ces.clemson.edu/me/policy.html

A companion document to the parallel computing tutorial is: Dr. Miller's Beowulf Performance Testing Page.

An introductory tutorial on parallel programming can be found on:
Dr. Miller's Parallel Computing Tutorial Page


NEW ACCOUNTS:

Please contact the system adminstrator for new accounts: Dr. Richard Miller, rm@clemson.edu. Provide name and requested username. New accounts will be activated with a temporary password which must be changed by the user during first login. For security reasons, passwords should be at least eight characters and contain both letters and numbers.



ACKNOWLEDGEMENTS:

The original cluster was partially supported through the National Science Foundation through the Faculty Early Career Development (CAREER) program, Grant CTS-9983762, awarded to Dr. R.S. Miller. Thanks go to Dr. Ron Sass, Dr. Walt Ligon, and Dr. Dan Stanzione from the Electrical and Computer Engineering (ECE) Department for assistance in purchasing decisions. Special thanks also go to Phil Carns of ECE for hardware and software installation, and to Curt Moore of ECE for assembly and software assistance with the eight 933 MHz based nodes. The most recent rack based cluster was purchased from TeamHPC (teamhpc.com). Special thanks to Dr. Jim Qiao for providing financial support for the new cluster.