AENEAS Supercomputer

	The new generation of supercomputers is named for mythic heroes, a
testimony to the power of computers in our culture.  Initiated in the summer
of 1994 at NASA's Center of Excellence in Space Data and Information
Sciences, the first such cluster of 16 simultaneously operating PC's was
named "Beowulf" for the warrior hero of the Old English saga who battled
marvelous sea monsters to save his culture.  In other universities and
research labs around the country, the Beowulf model spawned copies,
including Los Alamos National Laboratory's Loki supercomputer, named for the
trickster god of Norse mythology.  Late last year UCI got its own champion,
the AENEAS (Array of Enhanced Nodes Supercomputer), named for the Trojan
warrior who founded Roman culture.  

	Despite its super powers, AENEAS came from humble stock. Designed by
Herbert Hamber, Professor of Physics, AENEAS was built from components that
can be found at Fry's Electronics.  Beowulf-class supercomputers are
composed of a series of PCs, or nodes, that work in parallel on a particular
research problem, dramatically increasing speed and the ability to handle
complex problems. Aeneas is the second-fastest Supercomputer of its kind in
the world.  AENEAS can perform 20 billion floating point operations per
second (20 GigaFlops) with all processors working simultaneously on the same
problem.  It has 8.5 GigaBytes of memory (8,500 MegaBytes) and can store 214
GigaBytes' worth of programs and data on its array of disk drives.

	"The idea started because UCI didn't have adequate resources to
handle large-scale problems," Hamber says.  "There are a number of problems
that you can address on an individual PC, but others are too big -- they
require either a lot of memory or a lot of computing time."  Such
large-scale problems range from Hamber's work in elementary particle theory,
which studies complex interactions among quantum mechanical particles, to
studies of DNA, which involve thousands or even millions of interactions
among atoms.

     The AENEAS began last November with a series of 16 nodes ganged around a
high-speed switch that enabled them to converse with each other.  The
computer quickly grew to its current size of 64 nodes and two switches.  The
nodes themselves are simple, cheap, and reliable, each using an unadorned,
state-of-the-art motherboard with a 300 MHz Pentium II processor, two or
three fast Ethernet adapters, and 128 megabytes of memory.  The system uses
the LINUX operating system, which, along with UNIX, is standard among
scientific researchers and costs nothing.  Installed on $100 chrome wire
kitchen racks from Target, the entire assemblage cost about $120,000 and
went together in a matter of days.   

	Until a few years ago, researchers could only work on large-scale
problems at the handful of DOE- and NSF-supported supercomputer centers
around the country that had multimillion-dollar supercomputers like Crays.
However, because less expensive desktop computers and increased Internet
access to supercomputers have cut down on the need for off-site
supercomputer access, many of those national or regional centers are being
scaled down, with money being redistributed to support campus-wide
resources.  The new generation of home-brewed supercomputers foregrounds not
the hardware, but the research that the hardware can support.  
For Hamber, having access to a supercomputer was a necessity.  "I have a
history of using a lot of computer time," he says.  "In 1989 I was the
largest user in the U.S. -- I think I was using about 10 percent of the
national Cray resources." 

   Although the AENEAS is relatively low in cost, funding it turned out to be
an unconventional challenge.  When the National Science Foundation rejected
an initial joint proposal in the summer of 1997, Hamber turned to his
colleagues for support.  For the initial rack of 16 nodes, he collected
money from colleagues in High Energy Physics, Condensed Matter Physics, and
Plasma Physics.  Later he turned for matching funds to Ralph Cicerone, now
Chancellor of UCI, and once the machine was running, to colleagues in
Mechanical Engineering, whose support allowed him to double the size of the
machine.  Last August, Hamber doubled the size of the machine again,
bringing it up to its current size of 64 nodes, by garnering support from
researchers in Astronomy and Biology and matching funds from Vice Chancellor
for Research Fred Wan.   

   From the beginning AENEAS has borne fruit.  At any given time it is able to
host a variety of projects, with several researchers using portions of the
available nodes.  Currently Professor of Mechanical Engineering Donald
Dabdub is working on an atmospheric study of pollution in the LA Basin with
collaborators in Earth System Science.  Associate Professor Feng Liu is
working on steam turbine blade design, and Hamber is using the remaining
nodes for his project in elementary particle theory.  

   The production of the AENEAS lives up to the promise of its eponymous
ancestor in several senses.  In building on the original Beowulf model,
AENEAS amasses an army of inexpensive PCs to do battle with both the high
cost and the complexity of research.  It allows UCI researchers active in a
variety of disciplines to pursue leading-edge science in a way that would
not have been possible before.  In addition, the project demonstrates the
power of collegial collaboration and has spawned a professional computer
culture, bringing researchers from disparate disciplines together to support
and use a common tool.