Supercomputing's next frontier strains software

Paul Messina, senior adviser for Energy’s Accelerated Strategic gic Computing
Initiative, said this month at a conference at the Johns Hopkins Applied Physics
Laboratory in Laurel, Md., that ASCI is spending $100 million a year to stretch existing,
sequentially programmed applications so that they can run over thousands of processors.


The 100-teraFLOPS machines that IBM Corp. and Silicon Graphics Inc. are scheduled to
make for Energy laboratories in 2004 will have more than 8,000 processors, Messina said.


ASCI officials want to encourage use of commercial processors from Compaq Computer
Corp., IBM and Intel Corp., but they are having a tough time finding ways to interconnect
thousands of nodes to share distributed memory.


Messina said the processors will pour out such a flood of 3-D computations that IBM is
developing new tape technology to store as much as 1 terabyte of data per tape. But that
is a drop in the bucket when compared with ASCI databases that will run as large as 100
petabytes, the equivalent of 100 million gigabytes.


The only way humans can comprehend such vast amounts of data is through use of what
Messina called data visualization corridors.


“We are supposed to have high-resolution wall displays,” he said. “In
2001, there will be high-res immersive displays and, by 2004, desktop displays.”


Because the ASCI program primarily focuses on weapons research, it must build on
existing software models developed over the last half-century by nuclear experts who
mostly have retired.


“No software exists,” Messina said. “We need it, so we are going to push
that development” through five universities that act as contractors.


“We don’t even have a full 3-D simulation that can take simultaneous account
of everything that happens in a shock wave—the chemistry, the aerodynamics, the
materials behavior,” he said by way of example.


The largest hydrodynamics problem ever solved at Energy’s Lawrence Livermore
National Laboratory in California updated 2.2 billion zones per second of an event in 3-D
and “lots of different scales from microscopic size to human size,” Messina
said.


ASCI also looks beyond nuclear weapons research to computational models for global
climate, combustion, effluent flows and biochemistry.


Supercomputer research is booming now, Messina said, although there was “a flat
spot” at the end of the High-Performance Computing and Communications Program when
development stagnated.


“ASCI shows what multiplying by a factor of 10 can do” in the hardware arena,
he said.


The government “is having to invest a lot in software algorithms and programming
methods,” Messina said. “Although it’s clunky, it’s a time machine
that will enable us to zoom forward.”  



inside gcn

  • machine learning

    Mitigating the risks of military AI

Reader Comments

Please post your comments here. Comments are moderated, so they may not appear immediately after submitting. We will not post comments that we consider abusive or off-topic.

Please type the letters/numbers you see above