Slashdot Mirror
Sandia, Compaq, and Celera To Build Petaflop Machine
Fact-o-matic writes: " Compaq, government weapons facility Sandia National Laboratories and genetics researcher Celera Genomics are teaming up to build a petaflop computer -- one that will process 1,000 trillion operations per second.
To listen to an audio playback of today's press conference announcing the project, Celera has set up a phone-in recording: call (800) 642-1687, and enter the conference ID: 818790
You can read the joint press release or the
Compaq press release"
It's obvious people didn't read the press release before hitting their reply buttons. 1st: Compaq provides the hardware. Sandia and Celera cooperate on getting biotech applications to utilize the beast. Each company is throwing 4 million/year at the project for a few years. 2nd: A distributed, loose, confederation of machines, like SETI@home, can't solve the problems such a machine is designed for. There is way too much intercommunication required. 3rd: The press release indicates a 1000T machine, potentially. It doesn't give that as a goal. I believe the top of the indicated range is actually 100T. 3rd: Not in the press release, but allegedly reported in an industry rag (Computer World did someone say?) is that it's Linux. That's not what is planned. Someone's thinking Cplant. Cplant is reaching for a "top 10" spot on the top 500 list, one of the releases reports. Cplant doesn't even need to deliver 1T to accomplish that right now. 4th: Blue Gene is a special purpose machine. It's designed to solve one problem -- Protein folding. It is about as useful for Celera's problems as the IBM chess machines would be. Finally, as to the gov't controlling biotech for some implied dastardly deeds: Bunk! Can't you imagine folks at the national labs wanting to make positive contributions to humanity? If not, read their web pages: www.sandia.gov, www.lanl.gov, www.llnl.gov
"The Sandia labs and Celera Genomics Group in Rockville, Md., plan to work together on the project under a joint research and development agreement, with Houston-based Compaq as their technology provider. The planned system will be built around future versions of Compaq's AlphaServer SC supercomputer line and is being designed for use in complex applications in the fields of computational biology and life sciences...."
"The prototype supercomputer will likely use 10,000 to 20,000 of Compaq's Alpha processors and is being budgeted at $150 million in current costs, according to Blake. He added that the first system could eventually lead to the development of a so-called "petacruncher" -- a machine capable of 1,000 teraflops -- by the end of the decade."
you do realize, don't you, that distributed systems aren't the answer for everything, right?
projects like distributed.net and seti@home are perfect for distributed systems b/c the data analysis can be easily split into manageable sets which can be passed off to willing participants.
If you are referring to something more along the lines of a beowulf cluster, then you should take a look at 3. of the beowulf faq.
In order for a cluster of machines to make a difference, that application has to be built for such a thing.
hmm...wait a second, did I just respond to a troll?
-Peter
great so all of the people that are going to be interested in hearing the press conference SHOULD have an internet connection and could have easily DLed an mp3 of the conference.
how big was their phone bill for this?
Well according to Computerworld it runs Linux. http://computerworld.com/cwi/story/0%2C1199%2CNAV4 7_STO56666_NLTpm%2C00.html
-- You can be a geeklord too
The drug industry is the only private industry
segment that can afford $100 million computing
initiatives. Many of those fancy new drugs cost
hundreds of dollars per month per prescription.
The US government generally is the only client
with the funds to push computings' edge, i.e.
that is systems over $20 million dollars.
That buys 10 terflops now and by 2010 a petaflop.
Now and then an industry will be doing good enough
to buy the large machines- drugs, oil, Hollywood,
but that is transient. The government has been
supporting advanced computing since computers
were invented around WWII, instutionalized in
DARPA and the National Labs.
Been studied for many years since the mid 90s
when the teraflop barrier was surpassed.
Moore's Law predicts a factor of a thousand more
or less in 15 years. The issues were whether
conventional hardware and software development
would make this next jump of a thousand,
on radical new inventions would be needed.
Currently it looks like the existing trend
should squeak by, but physics will impede the
next jump of a thousand to exaflops.
Celera's attempt at the human genome is to break up the genome into extra-small fragments, sequence them, and use a computer to reassemble them into the contiguous genome.
The problem with this is that the human genome has lots of repetitive sequence which make it hard to identify what fragment goes where.
Compare to the government-funded approach of first breaking up the genome into manageable chunks of known location and size, then break those up into smaller extra-small chunks and then re-assemble those. Then you know what you've already done, and where you've got gaps.
Obviously taking fragments from all over the genome and assembling is going to take a significant amount of processing power, much more than it would take to just do one smaller manageable chunk of the genome.
Here we have Sandia, a government-funded laboratory working with Celera, who wants to privatize the genome, patenting interesting bits and selling access to their database for the rest at extremely large dollar amounts per user.
Now correct me if I'm wrong, but I'm seeing something wrong here.
So technically not only are our tax dollars funding the government-funded Human Genome Project, we are also paying a national laboratory to help develop a supercomputer for commercial interests, correct?
Sandia Lab and the Human Genome Proejct are both run by the US Department of Energy, so I certainly seem to see a conflict of interest here... anyone else?
I really am sorry, I just can't help myself...
... but Imagine a beowulf cluster of these...
This looks like a bigger, faster, beter computer like the ones that are used to simulate bombs which is the primary job of Sandia labs and the DOE's research. This box does not look like it will be all that fast for the types of things needed for geome matching.
Geome matching needs a nice long word (65k bits?) registers and a fast word sized barrel shifter/comparator. That is what is need to search for sequences that appear in several places.
Cray computer company (not Cray Research) were building a device like this for the nice spooks at the NSA when they decided they didn't need it anymore an cancled the contract. That box looked like it was designed from the ground up to do research on very long bit streams (say RSA type keys) but they canceled the project after paying something like 90% of the box and the bits that had been built were distroyed. Too bad that machine didn't see the light of day, it would have been great for finding patterns in DNA sequences.
"government weapons facility Sandia National Laboratories and genetics researcher Celera Genomics are teaming up"
Does that phrase scare you as much as it does me?
I assumed it would be for nuke simulation till I saw that. This strikes me as somehow worse.
The real Eight Star misses Technocrat.
lsmvcprm.com, Tools for geek power
How long 'till the 800 number says that all lines are full?
Moore's law is a flippant statement that happens to hold true. It is no more "law" than the Monroe Doctrine was, it just happened to work.
We can double the transistor density whenever we want. Economic factors will effect it as will government research policies.
There is no need to exceed Moore's law, you could just increase the number of processors. However, if we need to increase faster, we can, we just pump money into engineering research.
There are laws of physics - these can't be broken (although the law could be changed if we had a problem)
There are laws of society/traffic - these can be broken but you can get in trouble
There are historical trends with cool names that include law... these can be broken whenever someone is in the mood too...
> do that can't be done by a distributed system?
Not every possible computation can be made distributed without major performance loss. Remember, this baby IS vastly parallel (10K-20K processors), but the inter-processor communications are way faster than any network, and some problems that can be parallelized aren't easily distributed.
Also, i doubt you want nuke research done on the same set of systems running SETI@Home :-)
-DVK
"The right to figure things out for yourself is the only true freedom everyone shares. Go use it"-R.A.Heinlein
I do a lot of research into Biological Molecular Dynamics Simulations of Protein-Substrate interactions. When you consider that there are often several thousand amino acids in a protein, each amino-acid has a goodly number of atoms. Each of those atoms occupy multiple dimensions of space and at least one of time. Each of those atoms also have other properties such as velocity, electrostatic properties, physical size, covalent bonds to other atoms etc, and each of those atoms have to interact with an large number of other atoms.... (And I'm not even taking into account Quantum Mechanics). Forget galaxy collisions, Biomolecular Dynamics is a tough cookie. Believe me one of these babies would be really quite useful.
Problem is, the doin't know what the genes actually do, for most of them.
So the extra computing power has to go into sorting this out, and figuring what they mean, the grammar of the genes.
The only thing I can think of, that would be like that, would be the old translating of the rosetta stone.
Now that was a political text. Now what if the text had actually be a discussion of the subatomic particle reactions that take place in a matter/antimatter reactor? The task of translating for Napoleons archeologists would have been much harder, even *if* the ancient greek had been in "clear text", because archeologists did not know higher mathematics, etc. Probably, it would have been seen as a *really* obscure alchemy or religious text.
So now we have a similar task. Knowing the letters and a few "words" of the genome does not mean anything like knowing the design principles that are incorporated into a strand of DNA. DNA is all compiled coded, and we are trying to manually de-compile it. Then reconstruct the source.
"It is a greater offense to steal men's labor, than their clothes"
just a thought...
don't sweat the petty things and don't pet the sweaty things
petaFLOP == 1,000 trillion FLOATING POINT Operations Per Second.
It turns out that the author of the article linked to uses the term to mean 1,000 trillion ops/sec too. I think this is correctly called 1,000 TIPS (trillions insturctions per second), although IPS is usually not that commonly quoted.
PETA won't stand for this. I can already see the lawsuit's coming, along with the spraypainted keyboard's and silicon is murder signs. *sigh*
---
-
An ethernet driver
-
A simple networking layer
-
Standardized disk drivers for virtual memory and input/output set storage
These things are glorified calculators -- you don't need a 'real' OS -- just build on top of a simple embedded/RT systems OS! This would likely yield small performance gains and save on maintenance, setup and support costs. Of course, you would have to standardize on hardware, but if you're building the thing from scratch (as opposed to just clustering a bunch of PCs around the office/lab), then you probably would anyways.Those who can, do. Those who can't, post on Slashdot about it.
IBM is creating a petaflop supercomputer, called "Blue Gene", for the Argonne National Laboratory near Chicago. It also is being used for genetics research and, more specifically, simulations of protien folding. These systems are not just lots of processors and memory, but huge arrays of disks and tape farms. The IBM systems was priced at $100 Million US Dollars.
And that would be one quadrillion. Come on, don't be so afraid to say quadrillion. I know you can.
"Ancillary does not mean you get to rule the world." --U.S. Circuit Judge Harry Edwards, speaking to the FCC's lawyer
Since when did it require those kind of system resources to run Gnome. Op... no wait... Genome... nevermind.
This sig intentionally left blank.
I suppose this computer, financed in part by Celera Genomics, will be called the "Celeron".
Yay!
I'm glad all that crazy animal rights group flopped!
I'm sure glad Intel doesn't have a hand in this. Holy Global Warming!
I'd rather be a unix freak than a freaky eunuch
Ewige Blumenkraft!
You know, I worked out by Sandia for a while and the type of people there are not hip leet computer builders who want to make this thing for the same reason people climb Everest and real investigative science happens.
They want the government to determine the course of genetic research, or at least the course of the distribution of information produced by genome research.
They don't care about the petaflop. Petaflops will be designed in many forms. And one that takes this many processors is not industrially replicable, anyway. So Compaq gets some advanced r&d but very little salable out of this collaboration.
Goat sex free since 2001
I'm what you might call a formulations development chemist. That means, when a client comes to us with a drug, I figure out how to get the proper release profile from the depot formulation that is to be used. That means I'm required to know organic chemistry, pharaceutical chemistry, pharmacokinetics, and polymer chemistry. Oddly, it does not require driving a truck.
Hi! This is the Sig, blatantly attached to the end of this comment.