Computer Science – Performance
Scientific paper
Dec 1998
adsabs.harvard.edu/cgi-bin/nph-data_query?bibcode=1998aas...19312001w&link_type=abstract
American Astronomical Society, 193rd AAS Meeting, #120.01; Bulletin of the American Astronomical Society, Vol. 30, p.1427
Computer Science
Performance
Scientific paper
It is now possible to construct high-performance ``Beowulf'' computers entirely out of commodity components and freely available software, thus obtaining a significant price/performance advantage over typical parallel machines. Avalon is such a parallel supercomputer. It contains 140 Alpha microprocessors connected by a fast ethernet switch, as well as 35 Gigabytes of memory and 500 Gbytes of disk storage. The total cost of the machine was slightly over \$300k. Avalon has performed gravitational treecode simulations of galaxy formation at rates of nearly 20 Gigaflops using over 100 million particles. The machine's performance of 48.5 Gigaflops on the well-known Linpack benchmark ranks it near the 100th fastest computer in the world. The resulting price/performance demonstrates that for some applications Avalon can provide computing power equivalent to commercial machines which cost nearly ten times as much. Beyond the cost advantage there are also additional benefits such as flexibility, upgradability, and the ability to have control of your own supercomputing resources.
No associations
LandOfFree
Avalon: an Alpha/Linux Beowulf cluster achieves unprecedented price/performance for Astrophysical simulations and data analysis. does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.
If you have personal experience with Avalon: an Alpha/Linux Beowulf cluster achieves unprecedented price/performance for Astrophysical simulations and data analysis., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Avalon: an Alpha/Linux Beowulf cluster achieves unprecedented price/performance for Astrophysical simulations and data analysis. will most certainly appreciate the feedback.
Profile ID: LFWR-SCP-O-998391