Solutions for Linux
AHPCC expands computing frontier with IBM Linux cluster.
Overview
Application Supercomputing Linux ® cluster for government, academic and private-sector research Business Benefits Ability to process 375 billion computations per second, at 10% of the cost of supercomputers built with special-purpose hardware Software Red Hat Linux Servers IBM ^ xSeries ™, IBM RS/6000 ® SP ™ Services IBM Advanced Technology Group
With its 256-node supercluster of IBM ^ xSeries servers running Linux, LosLobos, the newest AHPCC Linux cluster, ranked as the largest cluster in academia when it first went live.1
In the brave new world of Linux
deploying them on vast national
clusters, supercomputing is not only
infrastructures or grids consisting of
for richly endowed research institutions
many Linux clusters joined together.
and wealthy corporations. It’s an affordable and feasible option for midrange companies as well, since Linux clusters have cut the purchase price of supercomputers by about 90 percent, according to users. And if research currently underway at the Albuquerque High Performance Computing Center (AHPCC) is any indication, organizations in the future will be able to leverage their investments in Linux clusters by
“Linux clusters are definitely the future of supercomputing, and companies need to be able to see how this technology can help them.” –Frank Gilfeather, Executive Director, High Performance Education and Research Center, University of New Mexico
Linux — freedom to innovate
The AHPCC has been a leading-
through a partnership with the
edge parallel computing center using
Alliance, and is supported through
cluster technology since 1994, when
the Alliance by the National
the center was founded. Managed
Science Foundation.
by the High Performance Computing Education and Research Center (HPCERC) of the University of New Mexico (UNM), the AHPCC serves the National Science Foundation and other research members of the National Computational Science Alliance (Alliance). The AHPCC maintains an extensive array of supercomputing resources, including four Linux superclusters, a 32-node IBM SP3 machine and an advanced
Computing time on LosLobos, the AHPCC supercluster, is available to researchers through a competitive grant application process.
When considering vendors for its latest acquisition, the university conducted an open-bid competition. According to Frank Gilfeather, executive director of HPCERC, IBM offered the best quality and value for hardware, installation services and maintenance. “IBM demonstrated the quality of the IBM ^ xSeries hardware and the ability to deliver a stable, reliable system. And the company has
visualization laboratory.
delivered exactly what it promised.”
LosLobos, the newest and the
LosLobos speeds up
largest of the center’s superclusters, is a 256-node, dual-processor IBM ^ xSeries 733 MHz machine, with a high-bandwidth, low-latency Myrinet network for fast interprocessor communication. LosLobos is managed by AHPCC
computational sciences Most of LosLobos’ 300 users are National Science Foundation researchers exploring problems in computational science, physics, chemistry, biology and genomics, fluid dynamics, scientific visualization and other computation-intensive fields of research. The AHPCC invites researchers to submit proposals for research and awards processing time after a panel review.
“ Superclusters made from high-quality, off-the-shelf Intel-based components, such as the IBM xSeries servers, provide a cost-effective way to offer supercomputer performance to the user community.” – Frank Gilfeather
“The number of scientific problems that require supercomputing continues to grow,” says Gilfeather. “Linux open-source software decreases the cost-of supercomputing dramatically, and superclusters made from high-quality, off-the-shelf Intel-based components, such as the IBM xSeries
servers, provide a cost-effective way to offer supercomputer performance to the user community.” Apportioning processing power To balance processing demands and make sure that LosLobos’ 512 processors are efficiently occupied at all times, scientists at the AHPCC are using the Maui Scheduler, a widely used batch scheduler that was developed at the Maui High Performance Computer Center, which is also managed by the HPCERC. In addition to 256 IBM xSeries servers and one IBM RS/6000 H70 server, the IBM hardware components include
Students at the University of Albuquerque can learn about supercomputing on one of the most powerful machines ever built, capable of processing 375 billion operations per second.
one terabyte of IBM Serial Storage Architecture (SSA) disk, the highestperforming IBM disk subsystem for the RS/6000. The RS/6000 functions
“Superclusters are becoming
Grid computing: moving beyond
as the processor for the SSA, which
supercomputers in their own right,”
supercomputing clusters
stores file server information as well
notes Gilfeather. “With the Linux
As powerful as they are, the Linux
as data used for research in fluid
cluster built by IBM and the xSeries
clusters potentially can enable much
dynamics and other scientific fields.
servers, we have a solution that
greater computer processing by
scales from the smallest desktop
working with existing IBM SP2® and
system to the largest cluster.
SP3 machines, which are clusters of
Researchers can use smaller clusters
RS/6000 servers. A research project at
for research development and then
AHPCC, dubbed Vista Azul, combines
turn to superclusters for large-scale
an SP2 cluster named Vista, a Linux
computations. This is an exciting
IBM cluster called Azul, additional
new approach—develop research
storage and two IBM scalable graphics
locally, then run it globally using
engines to do realtime visualizations
open-source software created to
using the clusters.
LosLobos’ twin 733 MHz Intel IA-32 processors, providing a peak theoretical performance of approximately 375 billion operations per second. Myricom’s Myrinet—a powerful, internal network communication system that enables all the computer nodes of the cluster to behave as a single supercomputer— provides the high-performance interconnect network between the cluster nodes. The components were integrated and installed by IBM Advanced Technology Services.
support specific research disciplines. It’s a model that will bring highperformance computing to a broader base of researchers.”
Scientists at the AHPCC are
Testing center for commercial customers
For more information
attempting to run applications across
In its mission to expand the frontier
please contact your
both architectures to use their
of supercomputing, it seems almost
IBM marketing representative or
combined power. The combination
inevitable that the AHPCC should
IBM Business Partner
could be useful for solving molecular
be interested in helping companies
dynamics simulations, for example,
understand how Linux superclusters
which require different computer
are a feasible option for boosting
architectures to solve different types
their computing power exponentially.
of problems. “Eventually, we hope
IBM and the AHPCC have joined to
to even provide a gigabit switch, or
help companies make the transition
a fast connection between the two
to Linux clusters by providing a
switch architectures, so that we can
briefing center for IBM Linux clusters.
run applications in both environments
Companies are welcome to visit
simultaneously,” says Gilfeather.
the center, run their code on
In the future, Gilfeather hopes to be able to demonstrate that the SP environment, which many companies currently have, can operate side-byside with the Linux cluster environment,
centers by creating large national infrastructures or grids, in which supercomputing clusters like those
Computing Center, visit: www.ahpcc.unm.edu
visiting the center should contact their IBM representative.
with this technology can be an
expanding individual computing
Albuquerque High Performance
Gilfeather, companies interested in
and run applications across the two.
clusters, AHPCC is looking into
For more information about
support personnel. According to
“Getting some hands-on experience
of combining different types of
ibm.com/linux
LosLobos and talk to AHPCC user
and that companies can actually port
In addition to exploring the feasibility
Visit us at:
© Copyright IBM Corporation 2001 IBM Corporation Worldwide Linux Marketing Group Route 100 Somers, New York 10589 U.S.A.
eye-opening experience, and
Printed in the United States of America 08-01 All Rights Reserved
we’re glad to be able to help,” comments Gilfeather. “Linux clusters are definitely the future
The e-business logo, the ^ logo, IBM, the IBM logo, RS/6000, SP, SP2 and xSeries are trademarks or registered trademarks of International Business Machines Corporation in the United States, other countries or both.
of supercomputing, and companies need to be able to see how this technology can help them.”
Linux is a registered trademark of Linus Torvalds.
at AHPCC are joined together
Other company, product or service names may be trademarks or service marks of others.
seamlessly over the Internet. “The
This case study illustrates how one IBM customer uses IBM products. Many factors have contributed to the results and benefits described. IBM does not guarantee comparable results. All information contained herein was provided by the featured IBM customer. IBM does not attest to its accuracy.
question we’re working on now,” says Gilfeather, “is, ‘Can you compute across this grid at a number of locations on a single application and on a single problem?’”
1
Top500 Supercomputer Sites. www.top500.org References in this publication to IBM products or services do not imply that IBM intends to make them available in all countries in which IBM operates.
Printed in the United States on recycled paper containing 10% recovered post-consumer fiber.
GK20-5864-00