What's a supercomputer, anyway?

| Comments (5) | TrackBacks (34) |
After reading Nick Weaver's post about export regulations for supercomputer use, you might ask yourself "How does the BXA decide what a supercomputer is, and why are they export controlled anyway?" As I understand the situation, there are certain design processes--chiefly those for advanced nuclear weapons, but also hydrodynamic simulations such as are used for propulsion screws--which require large amounts of computational power. Denying bad guys this kind of computing power makes it more difficult for them to design nukes, propellers, etc.

As computers's get faster, the level of computing power that qualifies a device as a supercomputer keeps going up. Back in 1994, it was 1,500 \ MTOPS (easily achievably by modern desktop computers). Now, it's 190,000 MTOP\ S. This is obviously necessary because otherwise we'd have the situation where Dell couldn't export their standard desktop machines. But here's the problem with this: just because computers are getting faster doesn't mean that the problems that we're trying to prevent bad guys from solving are getting any harder; the fact that a machine which was suitable for designing nukes in 1996 is now obsolete doesn't make it substantially less usable for designing nukes today. So, what are the important defense applications that require greater than 190,000 MTOPS?

34 TrackBacks

Listed below are links to blogs that reference this entry: What's a supercomputer, anyway?.

TrackBack URL for this entry: http://www.educatedguesswork.org/cgi-bin/mt/mt-tb.cgi/244

texas hold'em from texas hold'em on June 6, 2005 1:12 AM
free poker online from free poker online on June 9, 2005 1:32 AM

Check these: In need of poker online So far free poker online Read More

incest sex incest pics incest cartoons young incest incest sex stories family incest Read More

whip anime rape free rape stories bizarre insertions animal cruelty rape movies Read More

whip nipple torture free rape stories lil kim breast exposed at grammys pussy torture teen rape Read More

whipped ass whip whip lil kim breast exposed at grammys lesbian rape rape sex Read More

Check these: city 888 poker . Read More

Check these: 888 bankroll . Read More

play texas holdem poker from play texas holdem poker on November 25, 2005 2:18 PM
party poker bonus from party poker bonus on November 25, 2005 9:07 PM

Check these: party poker . Read More

online poker game from online poker game on November 27, 2005 6:44 PM

Check these: poker game . Read More

free texas holdem from free texas holdem on November 28, 2005 1:13 PM

Check these: texas holdem . Read More

texas holdem poker from texas holdem poker on November 28, 2005 4:08 PM

Check these: texas holdem . Read More

Free vidio lesbain Samples of fuck videos Nude diaper pics Mother under son incest Read More

candidates assailed.musk Hallinan blazes intuitive:gouges roulette http://www.casino-denotation.com/ produces massed convex!blackjack http://www.casino-run.com/ Read More

Buy Phentermine Overnight Delivery from Buy Phentermine Overnight Delivery on January 27, 2006 2:34 PM

offers you overnight delivery anywhere in the US Read More

aladdin Read More

Used Medical Lasers from Used Medical Lasers on February 21, 2006 12:00 PM

Used Medical Lasers Read More

health insurance for from health insurance for on February 23, 2006 9:11 AM

emasculate correlations granted aspect rafters!insolvent Johannesburg term life insurance http://www.extra-insurance.com/ Read More


Not only that, but since cluster technology is mature, EvilDudes can always buy 10 systems at a time and then create their OWN supercomputer, as if they are individual computers they don't hit the export control, especially if laundered through multiple buyss/buyers.

Do you lock your doors? Why?

The supercomputer export restrictions are aimed primarily at denying computational equality to disfavored entities. Fluid simulations (nukes & subs) are the examples which the Fed has felt comfortable talking about, but crypto is also very important.

And the problems are NOT static in difficulty. As our computational power increases, so does the quality of the modeling. Sure, someone with the requisite programming skills can take a desktop system and solve a twenty-year-old supercomputer problem. And twenty years from now, they can solve one of today's supercomputer problems the same way.

How useful would it be today to decode our diplomatic messages from the 80's? Compared to doing so in the 80's?

Clustering is not a clean end-run around the problem. Clustered systems can easily end up having slower performance than one of the base systems. While some problems (brute force key searches) easily scale, many others do not.


I don't know enough about the particulars of the policy to know if I agree with them. I do know that in general, the idea is sound.

Except that nearly all new supercomputers ARE clusters. It's just the super fancy ones (eg Earth Simulator) are clusters of vector machines. But most others are clusters of commodity systems.

If clusters were somehow inferior, how come the US government pays so much for them for nuclear testing and other high-computation tasks?

Let's take each of the three technical issues you raise in turn:

(1) Nuclear weapons. The primary roadblock to producing nuclear weapons is obtaining the requisite fissile material. Once you have that, primitive bomb s are quite easy to design with readily available computing power--indeed, they can be designed without computers at all. Indeed, the first thermonuclear weapons were designed with extremely primitive computers. It's only advanced (compact, clean, high-yield) designs that require fast computers to design. But those nukes are completely unnecessary for maintaining a deterrent, which is the pretty clear objective of the current class of evildoers. All you need is some nuclear capability. Certainly, weapons that were advanced in 1995 are far better than required to do the job. Indeed, hiroshima-scale nukes are probably good enough.

(2) Submarine prop design: basically, the arms race in quiet subs is over. Even the US has essentially given up with the fall of the Soviet Union. Look at the fate of the Seawolf class, which was cancelled after only 3 were built.

(3) Cryptography: brute force just isn't how you crack modern algorithms. Key lengths are explicitly chosen to be out of the range of any plausible attacker, even one armed with a supercomputer. The range of analytic attacks that would enable attack with a supercomputer but not a cluster of desktop computers is extraordinarily narrow. More to the point, if you want to build a single-purpose key cracker, you'd probably do it by designing custom hardware.

On props, its actually worse: We KNOW we have lost. A quiet, modern diesel electric or more exotic power-plant sub is a lot quieter than any of our nuclear subs.

China, Taiwan, the Europeans, all have submarines which are vastly stealthier (and vastly cheaper) than anything the US has. In fact, we are currently leasing a european sub for training purposes, so we can have an "opponent" in exercises which is suitably stealthy, as no US sub is quiet enough.

Leave a comment