I cant carry a penknife on a flight from Los Angeles to San Francisco, but I can sell a supercomputer to Pakistan or the Ukraine. Welcome to 2002.
In March 1999, lobbyists for industry trade groups were hoping to lift what was then the ceiling of 10 billion TOPs, or theoretical operations per second, for computer sales to Tier 2 countries (those posing a relatively low risk that the computers would be used to develop weapons—for example, Argentina).
As of this month, however, U.S. computer makers can sell much more powerful machines to even Tier 3 countries ("posing proliferation or other security risks," in the elegant phrase of the U.S. Bureau of Export Administration—including Israel, as well as my opening examples). The presidents order elevates the threshold of review to ... wait for it ... 190 billion TOPs.
A factor of 19 in 34 months: Thats doubling every eight months, more than twice the pace of Moores Law. And the rationale for this is ... what? If anything, the growing use of Beowulf clusters (www.beowulf.org) should slow the rate at which we raise the threshold: A given computer building block can now be combined much more readily to yield large multiples of that units power.
No, I dont really think that the government should limit U.S. IT exports to four-function calculators. Im challenging the idea that raw computational speed is a good measure of IT performance. If I have Mathematica and all you have is Excel, my oldest Pentium laptop and I will fly star ships before your Pentium 4 workstation can design a 747. Software matters. Development productivity matters.
Moores Law quantifies the power per dollar of input, not the ROI of the output. Imagine that next years hardware would cost at least as much as this years. How would it affect your thinking about application development, use of bandwidth and other resource trade-offs?
Anyone can do more with more. Is it possible that you can leap ahead of the pack by doing it with less?
Tell me whats too dangerous to sell at [email protected].