PRWeb The Leader Press Release Distribution
See How PRWeb Works

We're here to help 1-866-640-6397

Login Create Free Account


All Press Releases for January 11, 2005 Subscribe to this News Feed    
 

128-Xserve G5 Cluster Achieves 1.21 TeraFlop - "Dawson" Cluster Exceeds a TeraFlop Using 256 2-GHz G5's

Dauger Research, Inc., announces today that a new milestone for cluster computing and the Macintosh platform has been accomplished. Software written by researchers at Dauger Research and the University of California, Los Angeles, (UCLA) achieved over 1.21 trillion floating-point operations per second on a cluster of 128 Dual-Processor Xserve G5s.

Huntington Beach, CA (PRWEB) January 11, 2005 -- The Plasma Physics Group at UCLA acquired 128 Xserve G5s to create the new "Dawson" high-performance parallel computing cluster to advance plasma physics research. Among the codes executed on the Dawson cluster, the AltiVec Fractal benchmark was one that achieved over a TeraFlop (1 TF = 1000 GF = one trillion floating-point operations per second). This result is the largest calculation yet performed using Pooch Pro.

With the advent of the Dawson cluster, the Xserve has finally been successfully applied to large numerically-intensive physics calculations. Using Pooch Pro, provided by Dauger Research, this new Xserve cluster ran parallel computing jobs that achieved over a TeraFlop. A similar result using Linpack was accomplished on the Dawson cluster, placing it in the most recent Top 500 Supercomputer list.

Most recently, the Pooch family introduced new technologies to access and run clusters behind firewalls and automatically access job output. Adding support for the use of Myrinet hardware from Myricom and the open-source LAM/MPI implementation, Pooch today supports five distinct MPI implementations, making good on its "MPI-agnostic" support for clustering.

The software that made the 1.21 TeraFlop result obtainable is Pooch Pro and MacMPI, the latest incarnation of AppleSeed, a project begun by physics researchers at UCLA in 1998. For seven years and counting, their software is being used world-wide to transform Macintoshes into easy-to-use, numerically-intensive parallel computers. Complementing milestone benchmarks performed using 76 Power Macintoshes at the University of Southern California (USC) in 2001 and another using 33 Xserve G4s at NASA's Jet Propulsion Laboratory (JPL) in 2002, this most recent calculation demonstrates the scalability and computational power of clusters based on the Xserve.

The hardware included 128 Dual-Processor G5/2GHz Xserves and a Cisco 6500 Gigabit switch. The application code was the AltiVec Fractal Carbon benchmark. The operating system was Mac OS X Server 10.3, and the parallel computing code was launched and managed using Pooch Pro. For full details on the benchmark, readers may visit the Dauger Research web site. To learn more about parallel computing, readers may visit the Pooch web site, where users can download Pooch, a Software Development Kit including example source code, and tutorials and updated information about designing, compiling, and running code in parallel. The Pooch site makes it easier than ever to write, develop, and run your own parallel code today.

Dauger Research, Inc., was incorporated and founded by Dr. Dean E. Dauger, the award-winning author of Atom in a Box and Fresnel Diffraction Explorer who co-authored the original, award-winning Kai's Power Tools software. After completing his Ph. D. in physics, he founded Dauger Research, Inc., to bridge the divides between the scientifically and technically complex and the mainstream by making high-performance computation and visualization easy to use and accessible to users.

# # #

OPTIONS
Printer Friendly Version
Email this story to a colleague
CONTACT INFORMATION
Dr. Dean Dauger
Dauger Research, Inc.
714 840-0013
Email us Here
ATTACHED FILES

Dawson Cluster Clip
UCLA's Plasma Physics Group uses the new Dawson Cluster for state-of-the-art physics research. 128 Dual-Processor Xserve G5s drive this new high-performance system.

Dawson Cluster Performance Scalability Chart
This chart shows the performance of the 128 Xserve DP G5 cluster as a function of node count and problem size. As expected, larger problem sizes scale better, achieving over a TeraFlop in performance.

ABOUT PRESS RELEASES
If you have any questions regarding information in these press releases please contact the company listed in the press release. Please do not contact PRWeb. We will be unable to assist you with your inquiry. PRWeb disclaims any content contained in these release. Our complete disclaimer appears here.