Bush Lightens Supercomputer Export Restrictions
MrZeebo writes "According to a story on Yahoo! News, Bush has finally decided to lift the Cold War-era restrictions on how fast an exported computer can be. Now, computers as fast as 195,000 MTOPS (up from 85,000 MTOPS) can be exported to countries such as Russia, China, and Pakistan."
Wasn't it not all that long ago that Ashcroft (and the Bush administration) were beating their chests over the Clinton era "mistake" of lightening export restrictions on encryption software?
Anyone know what the impetus behind this move was?
"Moving through the masses like a fish through water." syrup
Why not just use a cluster of pc's/macs?
;-)
I think a big part of the answer might lie not in hardware, but in software.
As you know, the scientific and technical computing world still runs on Fortran. I know the SGI and Cray Fortran compilers are fantastic, especially the Cray vector-optimizing compiler; I would expect that the compilers NEC and Fujitsu use are similar. But as I understand it Absoft's Fortran compilers for Linux and Windows aren't up to those standards.
You might be able to run benchmarks or other C or assembly code as fast on a cluster as on a commercial supercomputer, but if the compilers aren't as good, your application will suffer.
It's important to note that this is just speculation on my part. I've only ever used SGI's and Cray's Fortran compilers, so everything I know about Absoft's comes to me second-hand. If Absoft rocks and I don't know it, it's not my fault.
Let's try this again.
Wouldn't clustering be a way to circumvent the law in the first place?
The problems that the law was intended to make difficult to solve (nuclear weapons simulation, aero flow analysis, cryptography, and so on) are, as far as I can tell, problems that can can be attacked in parallel, and so are good applications for clusters to tackle.
Well then, if the restriction prevented the export of any computer faster than x, couldn't a cluster of n export-legal computers of speed y (y less-than x ) produce a total throughput power Y (Y greater-than x)?
And for smaller values of y, substitute larger values of n to gain the same net power Y.
So really, I would think that clustering technology rendered (heh) the restriction moot a long time ago.
.
Want to learn about race cars? Read my Book