Re: No Singularity?

J. R. Molloy (
Thu, 18 Nov 1999 09:06:32 -0800

Another obstacle to the Singularity: Bureaucracy.

--<>-- --<<<+>>>-- --<>--

Government scales back its lead role in supercomputing By Mark Hall

PORTLAND, Ore. -- The U.S. government once dominated supercomputer development but has surrendered that leadership to industry. That retreat has put supercomputing technology at risk, according to analysts at International Data Corp. (IDC).

The analysts spoke here at SC99, formerly known as the Supercomputing Conference.

During a roundtable yesterday, IDC analyst Debra Goldfarb said the government lacks a single vision to give supercomputing vendors direction. This has put commercial users of high-performance computers in the role of providing leadership to research and development for the highest end of the computer industry.

"The government has shrunk in its influence at the high end," Goldfarb said.

Analysts said political and economic pressures have contributed to this decline, in addition to the government's and vendors' insistence on building systems that serve commercial interests as well.

"The government recognizes that commercial applications need to be taken into account," said Earl Joseph, also an analyst at Framingham, Mass.-based IDC.

But both analysts agreed that the strategic nature of supercomputers makes it imperative for the government to step in and work with vendors to continue to advance the state of the art. The problem is that inside the government, "religious wars" exist about which direction supercomputer development needs to take, they said.

For example, at the three major U.S. weapons laboratories -- Lawrence Livermore, Los Alamos and Sandia -- budget and political considerations forced researchers to build only a single, shared 100-TFLOP computer destined for delivery in 2004 instead of one for each lab, as some had hoped for.

Dona Crawford, a director at Sandia National Laboratory, told an audience here yesterday that her research team is trying to construct this system -- and the network that will connect it to the three labs -- with as many industry-standard components as possible to keep costs down. She said working with "lowest common denominator" technologies, such as TCP/IP, are "barriers for us to tackle."

Joseph said, "Commoditization of technology, whether processors or operating systems, is distracting us from doing real science."