US has most powerful computer... uses it for nuclear weapons research by ANTHONYT

REAL WORLD EVENT DISCUSSIONS

US has most powerful computer... uses it for nuclear weapons research

POSTED BY: ANTHONYT
UPDATED: Wednesday, June 20, 2012 05:02
SHORT URL:
VIEWED: 1914
PAGE 1 of 1

Monday, June 18, 2012 12:35 PM

Freedom is Important because People are Important

http://money.cnn.com/2012/06/18/technology/sequoia-worlds-fastest-supe
rcomputer/index.htm?source=cnn_bin

Hello,

I was dismayed to learn that we have the most powerful supercomputer in the world and that we use it for nuclear weapons research. I was surprised to read about it, since my imagination has trouble encompassing what exactly a supercomputer brings to nuclear weapons. I had believed that the technology was already mastered. Certainly I never envisioned that we'd want to devote so many resources to the topic.

What are we doing with all those petaflops? How do they help nuclear weapons research?

And isn't there something more valuable we could be doing with all that computing capacity?

--Anthony

NOTIFY: Y | REPLY | REPLY WITH QUOTE | TOP | HOME

Monday, June 18, 2012 2:19 PM

MAL4PREZ

Quote:
Originally posted by ANTHONYT:
http://money.cnn.com/2012/06/18/technology/sequoia-worlds-fastest-supe
rcomputer/index.htm?source=cnn_bin

Hello,

I was dismayed to learn that we have the most powerful supercomputer in the world and that we use it for nuclear weapons research. I was surprised to read about it, since my imagination has trouble encompassing what exactly a supercomputer brings to nuclear weapons. I had believed that the technology was already mastered. Certainly I never envisioned that we'd want to devote so many resources to the topic.

What are we doing with all those petaflops? How do they help nuclear weapons research?

And isn't there something more valuable we could be doing with all that computing capacity?

--Anthony

I actually had a really entertaining job interview with the people who do this. Basically, I stopped in just because the guy was there and I thought it might be connected to seismology jobs. But no, he was hiring for exactly what you're talking about - computer tests of nukes. Not what I want to do with my life. It was an interesting conversation. :)

As the govt guy explained to me, the idea is that nuclear weapons are no good unless they work and unless we are quite confident that they work. But we can't test them. What we can do is build computer models of the explosives and test them in the virtual computer world. I would guess that much of this is about improving bomb designs (which seems a bit silly to me. Don't nukes do their job well enough as they are? What needs to be "better" about them?) but there is also a lot of concern about the results of aging. What happens as the metal breaks down? How often do we need to replace the various parts? To me, that seems a more reasonable thing that has to be done. If you're gong to have the bomb, you better know how to keep it up to date.

The reason it takes so damned much computing power is that computer models need to be cut up into teeny tiny pieces. Parameters (density, compressibility, conductivity, melting points, etc) are set within each little piece in the grid, and the laws of physics are applied at each boundary between the little pieces. The action is set in motion with some initial conditions, (ie explosion starts HERE) and the resulting state is computed at points going forward in time. To get a realistic solution, you need really small pieces and really small intervals in time. So - lots of computer space and speed.

I spent three years working on a code to do something like this, but in seismology. I used computers like this though I never ran on the really fast government systems. I used parallel computers in academia. Similar, but smaller and slower. It was actually really fun to work on parallel systems. You could have several hundred processors, each chewing on a different part of the grid, and running for days. Every time step they'd have to swap information - seismic wave is leaving one part of the grid and going to another, and that other processor needs to know. So part of the computer design is getting the wires between processors to be as short as possible to make that "swap time" really short.

I have visited several government computer labs over the years and seen a few that were at some time or another the world's fastest. They're pretty impressive. Keeping them cool is one of the big design challenges - they really pour out the heat. But, Anthony, you'll be happy to know that they indeed do lots of things. I'm guessing that time spent on weapons research is actually a small fraction. Just last year I visited the computer system at Oak Ridge where they do a lot of climate modeling, seismics, and engineering research. Examples: building tokamaks for fusion reactors (look it up on wiki if you don't know - I don't think they'll ever work, but they're pretty cool anyway). Heat flow models, like the flow of air above a candle. Fluid flow through porous rock. (Big in oil exploration). Astrophysics. Changes in world population. They have hundreds of applicants every year trying to get computer time for their projects.

Back to the petaflops: why so much needed? For better results we need to cut models into ever smaller pieces, which means we need more memory. Smaller pieces in space require smaller steps in time for computational stability, so we need faster computation to get all those time steps done. Well, that's my take on it since I did finite difference modeling. Finite element, which is probably what they use for weapons research, is a slightly different approach. Same general idea though.

*sigh* I kind of miss working of that stuff. Not bombs, of course. But the codes were cool, and visualizing the output.

Oh - you could look at some of the Oak Ridge apps here: http://www.ornl.gov/ornlhome/high_performance_computing.shtml

NOTIFY: Y | REPLY | REPLY WITH QUOTE | TOP | HOME

Monday, June 18, 2012 2:39 PM

KWICKO

"We'll know our disinformation program is complete when everything the American public believes is false." -- William Casey, Reagan's presidential campaign manager & CIA Director (from first staff meeting in 1981)

Also, that kind of computational power would he helpful in making the bombs more efficient. That sounds sick and macabre, but I mean it in a very real sense - the idea that you can make a nuclear warhead that gives you as big a bang or even a bigger one than the stuff you currently have on the shelf, and does so with less fissionable material and less radioactive fallout produced.

In short, we're still trying to produce nuclear bombs that are "acceptable" for daily use on battlefields, as if such a thing exists, or could exist.

"I supported Bush in 2000 and 2004 and intellegence [sic] had very little to do with that decision." - Hero

"I've not watched the video either, or am incapable of intellectually dealing with the substance of this thread, so I'll instead act like a juvenile and claim victory..." - Rappy

NOTIFY: Y | REPLY | REPLY WITH QUOTE | TOP | HOME

Monday, June 18, 2012 2:45 PM

NEWOLDBROWNCOAT

I'm a little surprised that the fastest go to Nuke research-- I do agree that what Mal4Prez describes is what they do, and why they need really big, really fast machines to do it.

But I'da thought that what the NSA does would rate even higher, snooping on all that communications traffic, and banging on really complicated codes and ciphers, using really complex math functions with really big numbers. I would think that the NSA has a better tactical payoff, too-- real, short term results that might actually catch or kill a bad guy, or actually stop a real world attack.

Oh, well, I'll bet they're in Second place.

NOTIFY: Y | REPLY | REPLY WITH QUOTE | TOP | HOME

Monday, June 18, 2012 2:52 PM

ANTHONYT

Freedom is Important because People are Important

Hello,

Thanks for the education, guys. :-)

--Anthony

Note to Self:
Raptor - women who want to control their reproductive processes are sluts.
Wulf - Niki is a stupid fucking bitch who should hurry up and die.
Never forget what these men are.
“The stupid neither forgive nor forget; the naive forgive and forget; the wise forgive but do not forget.” -Thomas Szasz

NOTIFY: Y | REPLY | REPLY WITH QUOTE | TOP | HOME

Monday, June 18, 2012 3:00 PM

1KIKI

Goodbye, kind world (George Monbiot) - In common with all those generations which have contemplated catastrophe, we appear to be incapable of understanding what confronts us.

http://www.wired.com/threatlevel/2012/03/ff_nsadatacenter/

The NSA Is Building the Country’s Biggest Spy Center (Watch What You Say)

It has created a supercomputer of almost unimaginable speed to look for patterns and unscramble codes. Finally, the agency has begun building a place to store all the trillions of words and thoughts and whispers captured in its electronic net. And, of course, it’s all being done in secret. To those on the inside, the old adage that NSA stands for Never Say Anything applies more than ever.

Now, this computer is only ever going to be used for one thing.

NOTIFY: N | REPLY | REPLY WITH QUOTE | TOP | HOME

Monday, June 18, 2012 3:07 PM

1KIKI

Goodbye, kind world (George Monbiot) - In common with all those generations which have contemplated catastrophe, we appear to be incapable of understanding what confronts us.

Mal4Prez

|| computing is something I've not really thought about or even looked into. When I have briefly run into it, what I wonder is, how do you synchronize the || processes so that they're working on the same particular part of the process at the same time, so that the outputs/ inputs of each computation are lined up. Or is that not a problem?

NOTIFY: N | REPLY | REPLY WITH QUOTE | TOP | HOME

Monday, June 18, 2012 3:51 PM

MAL4PREZ

Quote:
Originally posted by 1kiki:
Mal4Prez

|| computing is something I've not really thought about or even looked into. When I have briefly run into it, what I wonder is, how do you synchronize the || processes so that they're working on the same particular part of the process at the same time, so that the outputs/ inputs of each computation are lined up. Or is that not a problem?

The language I used was MPI-C, (MPI = Message passing interface) which has probably been replaced by something more sophisticated by now. But it handled all of this quite gracefully. I only needed to write one script, and it ran on all processors at the same time. The first steps of the code would assign each processor a number (n) and use that as a guide throughout.

For example, suppose we have seismic waves traveling through an earth model that is sliced into horizontal layers, like a stack of pancakes where each processor does the number-crunching in a single pancake and the top pancake is n=1. At the end of each time step, after updates have been calculated within each pancake, you'd have a section that says (to translate it to English):

I am processor n.
Send the wavefield leaving the top of my grid to processor n-1
Send the wavefield leaving the bottom of my grid to processor n+1
Wait for the waves coming down from processor n-1
Wait for the waves coming up from processor n+1
Go to the next time step
(with special instructions for the top and bottom pancake...)

Things are forced to synchronize because n can't move on until it's received the swap from n-1 and n+1.

Of course, there are all kinds of fun games you can play. How many pancakes is too many, so that wait/swap time becomes longer than the time you save by cutting the computation up? And what if a small part of the grid has all the action and is taking up too much time so that all the other processors are always waiting? Should the model division be dynamic, so that it changes each time step to keep things moving efficiently?

It really is fun stuff to code, and tough. If you've applied for 30 hours on the top-notch cluster at Oak Ridge, you have to have all this worked out in advance so that your code will run in the time you have.

I once saw the output of a code written at Los Alamos that modeled fluid mixing. Say you have a layer of water *over* oil, which is not stable. So you tap it and it starts to mix, the water sinking and oil rising. They wrote a code that adjusts the grid size where the action happens. Swirly places: very fine grid. Out where it's pure water or pure oil: course grid. You could visualize the mixing just by plotting the grid size. Very cool. I have no idea how they made that work on a parallel computer.

NOTIFY: Y | REPLY | REPLY WITH QUOTE | TOP | HOME

Monday, June 18, 2012 4:08 PM

1KIKI

Goodbye, kind world (George Monbiot) - In common with all those generations which have contemplated catastrophe, we appear to be incapable of understanding what confronts us.

I truly appreciate you taking the time to answer.

I was wondering if they'd all be keyed to the same model time-step (wait till all are complete, swap data, advance all 1 msec) or what. But that's just me muddling along with not a whole lot of the basics.

I can see that far better minds than mine have been doing way cool stuff. It's so cool that you are doing that and even cooler that you post about it.

So, THANK YOU!

NOTIFY: N | REPLY | REPLY WITH QUOTE | TOP | HOME

Wednesday, June 20, 2012 5:02 AM

CAVETROLL

I'd much rather they be testing nuclear weapons in a virtual environment than in a physical environment. Although, think of the electrons!

Kwindbago, hot air and angry electrons

NOTIFY: Y | REPLY | REPLY WITH QUOTE | TOP | HOME

YOUR OPTIONS

SHARE THREAD

NEW POSTS TODAY

USER	POST DATE

USER

POST DATE