Parallel Computing at JMU

	Dr. Sochacki sitting at the host of the 16 node parallel processing Beowulf system, Raptor housed at the Center for Computational Mathematics and Modeling (CCMM) in the Department of Mathematics and Statistics at JMU.
FIGURE CAPTION: Jim Sochacki slays computational dragons with 16- node processor Beowulf.

Raptor is a 16 node linux cluster with a host using the Beowulf system developed at NASA Goddard. The host has an AMD 600 MHZ processor, 798 MB of RAM, an Intel motherboard and two Intel ethernet cards. Each of the 16 nodes has a 450 MHz Pentium processor, 1 GB of RAM, an Intel motherboard and two Intel ethernet cards. The host and nodes communicate to each other through a 100 Mbit Cisco switch. We use Portland Groups High Performance Fortran (HPF) and Message Passing Interface (MPI) for our parallel software. We use both Fortran 90 and C. Students in the Computational Science program write programs using this software and submit jobs to Raptor through the Profile Batch Scheduler (PBS). We have found that communication is the bottle neck in parallel computing using linux cluster Beowulf systems. We have also had more success with MPI and are using that as our software for communication among the nodes. See the sites below for codes that run on the beowulf system and speed up results from using linux cluster Beowulf systems.

Dave Pruett

MPI Fortran Codes