Navy stares down big problems

Researchers who use the
new supercomputer installed at the Naval Research Laboratory all have very large problems,
computationally speaking.

The mission of the laboratory is somewhat akin to skating near the edge of the ice, and
it explains why the Navy lab in Washington owns one of the few Cray Origin2000
128-processor computers in existence.

Silicon Graphics Inc. built the NRL’s Origin2000, which is basically a
nonpartitioned processing dynamo, said Henry Dardy, chief scientist for NRL’s Center
for Computational Science.

The system holds 32G of memory, a half terabyte of internal disk storage and 1 terabyte
of external Fibre Channel disk storage.

Nothwithstanding its 128 MIPS R10000 processors, the NRL supercomputer has one parallel
kernel, one parallel file system and one set of parallel libraries, Dardy said.

Researchers advance science more swiftly if they can lay their data out across
massively parallel machines, he said. "That’s how they make their
breakthroughs—by attempting to scale up to the next level of understanding—and
they need a large machine to do that.’’

To get time on the machine, Dardy said, researchers must report their research results
to the Defense Department’s High-Performance Computing Modernization Office, which
provided funds for the lab to buy the 128-processor system.

The Origin2000, with its 50 billion floating-point operations per second, is not the
biggest computer in existence, Dardy said.

But then his frame of reference for scalability is the Energy Department’s
Accelerated Strategic Computing Initiative, for which the government will consume
teraFLOPS of computational power.

"A research lab can’t afford to have a major large machine because by the
time you get it, it’s obsolete,’’ Dardy said. But the laboratory needs
machines big enough to prove scalability.

"We’ll test it out,’’ Dardy said, then share the results with DOD
Major Shared Resource Centers, which need large computers to support day-to-day Defense
operations and warfighting.

The relationship with Silicon Graphics is very much a two-way street, Dardy said. Take
debugging the kernel—no vendor test suites can duplicate the research applications in
computational fluid dynamics, electromagnetic modeling, simulation and crystallography, he

Dardy said his biggest technical challenge is creating a load-balanced architecture
that gives all researchers the same level of computing performance, with no penalty for
long-distance access. More than half of NRL researchers work off-site in different
locations around the country, he said.

Balancing memory, disk and network bandwith to get the best possible performance is a
challenge. "Networking, storage and memory are all challenging when you get to these
rates,’’ Dardy said.

The NRL has been using OC-3 and OC-12 network lines directly into the Origin2000 and
directly to the researchers’ desktops, really pushing the limits of asynchronous
transfer mode technology, Dardy said.

Without a key piece of infrastructure software known as the Andrew File System, the lab
would be hard pressed to offer researchers the support it does, Dardy said. "We
couldn’t survive without it,’’ he said.

NRL uses a version of the Andrew File System that it licensed from Transarc Corp. of
Pittsburgh, a subsidiary of IBM Corp. The multiresident AFS, as Dardy refers to the
optimized version running on the researchers’ workstations and NRL’s computers,
provides a uniform file-naming scheme.

Dardy said he won’t worry about the possible discontinuation of the MIPS processor
after two more generations of the chip—that is, depending on how Silicon Graphics
makes the move to Intel IA-64 Merced processors. "What poses a problem is if the Irix
software were to change,’’ he said.


  • senior center (vuqarali/

    Bmore Responsive: Home-grown emergency response coordination

    Working with the local Code for America brigade, Baltimore’s Health Department built a new contact management system that saves hundreds of hours when checking in on senior care centers during emergencies.

  • man checking phone in the dark (Maridav/

    AI-based ‘listening’ helps VA monitor vets’ mental health

    To better monitor veterans’ mental health, especially during the pandemic, the Department of Veterans Affairs is relying on data and artificial intelligence-based analytics.

Stay Connected