Slashdot Mirror
ASCI White Detonates The First E-Bomb
totallygeek writes "Redefining the term vaporware, research scientists at Lost Alamos and Lawrence Livermore Labs detonated two computer simulations. ASCI White, the world's fastest supercomputer, ran the simulations of nuclear explosions. Scientists can now study nuclear weapon replacement components without violating the nuclear test ban, in effect since 1992.
Each simulation used more than 6.6 million CPU hours, which would take home machines 1000 years to complete. The data for each experiment was equivalent to 35 times the information available in the Library of Congress. ASCI White currently operates at 12 teraflops, but by early next year, Los Alamos expects to operate at 30 teraflops.
The seven month research project ended last Friday, and now the system is ready for use, after its sucessful testing."
Its primary purpose is to replace nuclear testing that has been banned for ten years -- ten years of having NO idea how the existing warhead supply is aging.
You may agree or disagree with their intended use, but right or wrong there are two critically important things that we have to know as long as a single warhead still exists.
1) As the parts age, will it work as designed, when it needs to go off
2) As the parts age, will it work as designed, when we sure as hell don't want it to.
In either case, failure carries terrifying outcomes. Think about it -- in one case, the warhead doesn't detonate completely, causing an incredible amount of fallout (Chernobyl-style), which is never the intent of a nuclear warhead. In the other case, people dye (very likely in a similarly polluting manner) when it goes off unexpectedly.
As long as nuclear warheads exist, this sort of research is absolutely critical, and its not anyones place to put down this research for ethical reasons related to the existance of the bomb. The two are related but totally separate, and you shouldn't cross those beams.
- analyze the human genome,
- fold cancer-curing proteins,
- locate possible sources of alien intelligence, or
- help the government explode a virtual nuke?
(No nastiness intended. My point is that it might be hard to get people to download a client with that particular goal in mind.)Running programs in parallel is pretty difficult; you have to figure out how to divide the problem amongst different processors. Some problems (which are said to be `embarrassingly parallel') are easy to do this -- every different processor just searches a different part of key-space for a key to decrypt a code, or a different part of frequency-space looking for a signal. There doesn't have to be any sort of inter-process communication to speak of in these problems.
A fluids or mechanical (or combined) simulation, however, requires lots of communication between computational elements. Each processor is simulating some region of space, and it constantly needs information about the fluid all around it to know what to do next. (Is a shock wave coming from the left?)
And even fluids/mechanics simulations are simpler than simulations involving long-range forces like gravity. In that case, every single computational element probably needs at least some information from every other computational element!
In cases like that, highly-distributed computing a la SETI@home won't work. Whereas for brute-force code-cracking, or searching for signals in reams of indepdendant data, it's perfect.