|
|

Home > Computers and Information Technology > Computer Technology & Equipment > Hardware Components
Virginia Tech Installs New System X Supercomputer, Ranks Third on TOP500
 |
| Published Date:
January 2004
Published By:
IDC
Page Count:
9
Order Code:
R104-14513
|
|
|

| This IDC Insight covers the new System X cluster at Virginia Tech, number 3 on
the TOP500 supercomputer list. I wonder whether there was a campus debate
between the jocks and the geeks at Virginia Tech this past summer over which
would achieve the higher standing: the well-regarded Virginia Tech Hokies
football team or the new System X cluster in the previously unranked
Computational Sciences and Engineering (CSE) program. This much is clear about
System X: There?s nothing Hokie about it.
In many ways, the new System X installation at Virginia Polytechnic Institute
and State University is typical of the new breed of cluster installations,
albeit on a larger scale. For $7.2 million ? $5.2 million for computer
hardware, $1 million for facilities upgrades, and $1 million for a customized
cooling system ? VT purchased a 17.6-teraflop behemoth. One hundred and sixty
student volunteers booted, configured, and installed 1,100 dual-processor nodes
into 182 oversize racks, and after weeks of assembly, System X first booted up
on September 23.
Once System X became operational, it was immediately tested on LINPACK, the
benchmark used for the TOP500 supercomputer rankings. Because LINPACK favors
machines with high peak performance without relying heavily on memory, I/O, or
system bandwidth, cluster architectures tend to do very well on it, and System
X was no exception, debuting at number 3 on the list.
IDC categorizes this type of installation as a "special capability cluster," a
cluster that targets the capability segment but is not a standard vendor
product. These systems may be configured by users (as is the case with VT) or
configured by vendors as one-of-a-kind installations. Such clusters frequently
excel in measurements such as LINPACK but lag on actual capability workloads,
especially if peak megaflops are a major consideration in the cluster's
specifications.
Virginia Tech is attempting to reduce the gap between theoretical peak speed
and real application performance, and it has included the following four
features in System X:
y The processor, Apple?s new 64-bit, 2GHz G5 will provide a lot of performance
without stepping down to 32-bit addressing. The Apple G5 is based on IBM
PowerPC technology.
y The memory, 4GB per node, is already above average for a specialized
capability cluster. Because System X uses 64-bit processors, memory can be
expanded on some or all nodes if desired.
y The interconnect, Infiniband from Mellanox, provides substantially higher
system memory bandwidth than standard Ethernet. VT plans to implement
programming models that leverage that bandwidth to provide shared memory across
multiple nodes.
y A homegrown cluster management scheme called Déjà Vu is being designed to
address the traditional limitations in large cluster reliability, with features
such as application-transparent job migration and checkpoint restart.
Virginia Tech is relying heavily on in-house expertise to manage System X, and
the work on the programming model and cluster management tools is ongoing. For
now, System X is for research codes only. However, the fundamental technology
pieces are in place for Virginia Tech to advance the state of the art in
cluster computing.
|
Similar Products
• Web Smart Switches Gain Traction With SMBs Worldwide
Published Nov 2008 by In-Stat
• BRIC India Executive Market Watch 2Q 2008
Published Oct 2008 by IDC
• BRIC Brazil Executive Market Watch 2Q 2008
Published Oct 2008 by IDC
• 2009 U.S. Computer & Equipment Wholesale Industry Report
Published Oct 2008 by Barnes Reports
• 2009 U.S. Office Equipment Wholesales Industry Report
Published Oct 2008 by Barnes Reports
• Plimsoll Analysis- Computer Hardware Distributors (UK)
Published Oct 2008 by Plimsoll Publishing Ltd.
• Computer Hardware
Published Oct 2008 by Key Note Publications Ltd
• BC/DR Implementations and Demand for BC/DR Consulting Services Among US Businesses
Published Sep 2008 by In-Stat
• The 2009-2014 World Outlook for Computer Hardware
Published Sep 2008 by Icon Group International, Inc.
• The Taiwanese Server Motherboard Industry, 3Q 2008
Published Sep 2008 by Market Intelligence Center
|
|
|
|