Globus (grid software being developed mainly at Argonne) has a security model built into it. See www.globus.org/security for some details.
Also, anything the DOE would do on this network would be unclassified, and completely non-export controlled. Classified work is done on internal networks separated from the internet by an air gap.
I wasn't referring to the 512 intel chips - I was referring to the MAP design.
In talking to people from SRC, they claim that this architecture (off the
shelf chips + reconfigurable computing) was Seymour's last big contribution
before dying.
And I never said that Cray Research didn't build vector machines. But
Cray Research pretty much abandoned that idea once the T3-D
was designed. The SV-1 is hardly a supercomputer...
This Cray company doesn't not really have anything to do with Seymour (other than the name). The architecture is more inspired by Tera and SGI than Seymour.
Remember back 5+ years ago, there was a Cray Research (that built the T3-E) and a Cray Computer (that was still building vector machines - the last of which was never bought by anyone [Cray-4])
If you want to see where Seymour's influence still reigns - check out SRC Computers . Hint: SRC doesn't stand for "source" - it's Seymour's initials.
Re:Other groups working on similar stuff
on
FPGA Supercomputers
·
· Score: 1
I'll add one more group: SRC Computing (Seymour Cray's last company he started before he died tragically). Check out their MAP architecture.
Because government
projects that involve computers are almost always administered by someone who knows nothing about them and is more interested in
servicing their own career at the expense of the system they're dealing with.
True, but:
You just described an awful lot of private industry, too
You are painting several hundred thousand workers and projects with an awfully broad stroke
Not that the government isn't wrought with those problems - but it's not totally exclusive to them...
Are you really getting paid directly by the gov't though - ie "scale" pay?
Here at Lawrence Livermore, we are actually University of California employees, and as such are not restricted to gov't salaries. Last I knew - it was the same for most LANL employees (ie - you were technically contractors with UC benefits, etc...).
Due to our proximity to Silicon Valley, the salary issue at LLNL was addressed here several years back with a "special action" salary for computer scientists which brought our salaries back up toward the curves generated elsewhere in Northern California. It was a very progressive move at the time, but in retrospect did help stop the bleeding and bring some equity to CS people.
It's been hard to keep up with that, though - as 4-5% raises are normal here, whereas 10-20% are standard elsewhere.
What the gov't needs to do is issue a tracking stock for their fiscal performance so they can offer stock options to all the young'uns.:-)
I'm sorry, but making those comparisons is about as useful as saying a Greyhound Bus outperforms a Ferrari because it can get 50 people around a racetrack faster. They're different types of machines for drastically different types of problems.
People will argue (myself included) that LINPACK is a useless benchmark for the Top500 - but distributed.net would probably run LINPACK ~100 times slower than a single CPU.
LLNL is Los Alamos' sister lab - they do very similar work, but aren't quite as well known.
ASCI White will be "behind the fence", and thus used mostly for classified work.
"Stockpile Stewardship" is the official language. Making sure weapons are designed to be "one point safe" is an example (ie - it won't go nuclear if someone unloads a machine gun or a shape charge into the pit).
Livermore just got the OK to put out an RFP for a 70 Teraflop machine for delivery sometime around 2004. LANL is getting a 30 Teraflop machine in about 2 years which will be built by Compaq. (ASCI White is 12 Teraflops).
The answer to your "what is the fastest single CPU out there" can probably be found at the Hot Chips web site.
My guess is that the Japanese (NEC or Fujitsu) are the current leaders, as they have continued to build highly vectorized processors - along the lines of what Cray used to do in the past.
Another thing to keep in mind is that these machines are very rarely run in a mode where a single application is using all of the machine. I work on these machines (currently ASCI blue), and the real payoff is that a dozen or so people can be running moderately parallel jobs all at the same time.
Each OS will only run 16 processors which make up a "node" of shared memory. There are hundreds of these nodes which are completely distributed memory over a high performance network.
The idea is to run a large problem using message passing (usually MPI), across nodes, and multi-threaded (often using OpenMP compiler directives - easier than Pthreads for scientific computing) inside a node. I am a developer on one of the large ASCI codes, and this is the model we are targetting, although currently we usually run using all MPI (no threads) just because the threads performance is not as good.
Some more info here for those interested in more detail.
This machine is going in at Livermore - but Los Alamos has already contracted for a larger machine (currently called the "Q" machine) which will be designed by Compaq - installed in 2002 (I think).
The folks at U-Wisconsin have been working on a package called Condor for many years. Although it's currently designed to work in a workstation cluster, some of the ideas are worth investigating for someone wanting to take wide area distributed computing to a new level. In particular, they have solved many of the problems of checkpointing a process when a user comes back to work on the machine.
Using screensavers is a cool idea and all - but you can only have one screensaver set to run at a time, no? Can I run SETI@home and distributed.net simultaneously? (Not that I'd want to - but I might want to schedule some priorities so each would get equal time while I'm gone for a weekend).
Maybe if condor shipped with linux distribs, it'd make it easier for this technology to take off?
I just went to a meeting about that last week... For awhile, both machines (ASCI Blue - the 3Tflop and ASCI White - the 12Tflop) will be available. This is because it usually takes many months for a system that size to become stable enough that users can do "production" work on it. So although they're assembling it at LLNL now, it probably won't be used by your everyday user until December or so.
When the machine is ready for "general" use ("general" as long as you have a Q clearance!), then the plan is to move the current machine to the unclassified side, and open it up for use by the ASCI alliances and other unclassified users.
They should be able to simply add it on cluster style as you suggest, since the current machine on the unclassified side is basically the same architecture. I can't tell you for sure that it's what they'll do - but if they do, they should have about a 4-5 tflop machine for unclassified use by the end of the year.
As far as what will happen to ASCI White when they're done with it - it's only being rented from IBM - so it'll go back to Kingston or whereever...
As someone who will be working daily on this machine at LLNL - maybe I can shed some light. Firstly, nuclear explosions are not the only thing which will be simulated. However, the funding for the machine is primarily for that. (The program is "ASCI" - Accelerated Strategic Computing Initiative). Because they are doing these calculations, the machine will be "behind the fence" (classified). As a result, the only people who will be able to use it will be those with a Q clearance (DOE Secret). By default, this makes it difficult (read: impossible) for your average uncleared researcher to gain access.
It is possible for unclassified simulations to be run on the machine with a Q-cleared "proxy" user running the code on behalf of someone else - but in this day and age of ultra-tight security in the wake of Wen Ho Lee and missing disk drives - that is highly unlikely to happen.
However, there is a mighty impressive machine available on the "open side" which will be used for things like weather sim, drug design, etc... In this case it will be the machine ("ASCI Blue") which is currently behind the fence, which will be moved outside. 5000+ PowerPC 604e's, and not a bad parallel environment to work in, I must say...
The ASCI program is also funding 5 university ASCI centers. These centers are targeted with solving unclassified "grand challenge" type problems which involve similar complexities to nuclear simulations. ie - solid rocket motor simulations (Illinois), Astrophysics (U Chicago), accidental fire scenarious (Utah), turbulence (Stanford), and material modeling/response (caltech). These centers get time (or "fight for time" if you asked them) on the unclassified ASCI machines.
The unclassified machines are usually just one step behind, or slightly smaller, so they don't make splashy headlines. They are, however, still very very impressive machines, and there is lots of groundbreaking research being done on them which would probably not have been possible (yet) without ASCI
I (and others) don't particularly like the fact that these machines are used mostly for nuclear simulations - but it's better than the alternative (craters in Nevada), and it's definitely helping push the envelope of parallel computing - which is all I really care about.:-)
Los Alamos has a similar contract with SGI to supply large machines for them. Sandia has a large machine from Intel, and have subsequently been concentrating on massive linux clusters (ala CPLANT) as their future.
Slashdot Mirror
Also, anything the DOE would do on this network would be unclassified, and completely non-export controlled. Classified work is done on internal networks separated from the internet by an air gap.
I wasn't referring to the 512 intel chips - I was referring to the MAP design. In talking to people from SRC, they claim that this architecture (off the shelf chips + reconfigurable computing) was Seymour's last big contribution before dying. And I never said that Cray Research didn't build vector machines. But Cray Research pretty much abandoned that idea once the T3-D was designed. The SV-1 is hardly a supercomputer...
Remember back 5+ years ago, there was a Cray Research (that built the T3-E) and a Cray Computer (that was still building vector machines - the last of which was never bought by anyone [Cray-4])
If you want to see where Seymour's influence still reigns - check out SRC Computers . Hint: SRC doesn't stand for "source" - it's Seymour's initials.
I'll add one more group: SRC Computing (Seymour Cray's last company he started before he died tragically). Check out their MAP architecture.
True, but:
Not that the government isn't wrought with those problems - but it's not totally exclusive to them...
Here at Lawrence Livermore, we are actually University of California employees, and as such are not restricted to gov't salaries. Last I knew - it was the same for most LANL employees (ie - you were technically contractors with UC benefits, etc...).
Due to our proximity to Silicon Valley, the salary issue at LLNL was addressed here several years back with a "special action" salary for computer scientists which brought our salaries back up toward the curves generated elsewhere in Northern California. It was a very progressive move at the time, but in retrospect did help stop the bleeding and bring some equity to CS people.
It's been hard to keep up with that, though - as 4-5% raises are normal here, whereas 10-20% are standard elsewhere.
What the gov't needs to do is issue a tracking stock for their fiscal performance so they can offer stock options to all the young'uns. :-)
I'm sorry, but making those comparisons is about as useful as saying a Greyhound Bus outperforms a Ferrari because it can get 50 people around a racetrack faster. They're different types of machines for drastically different types of problems.
People will argue (myself included) that LINPACK is a useless benchmark for the Top500 - but distributed.net would probably run LINPACK ~100 times slower than a single CPU.
ASCI White will be "behind the fence", and thus used mostly for classified work. "Stockpile Stewardship" is the official language. Making sure weapons are designed to be "one point safe" is an example (ie - it won't go nuclear if someone unloads a machine gun or a shape charge into the pit).
Livermore just got the OK to put out an RFP for a 70 Teraflop machine for delivery sometime around 2004. LANL is getting a 30 Teraflop machine in about 2 years which will be built by Compaq. (ASCI White is 12 Teraflops).
My guess is that the Japanese (NEC or Fujitsu) are the current leaders, as they have continued to build highly vectorized processors - along the lines of what Cray used to do in the past.
Another thing to keep in mind is that these machines are very rarely run in a mode where a single application is using all of the machine. I work on these machines (currently ASCI blue), and the real payoff is that a dozen or so people can be running moderately parallel jobs all at the same time.
Each OS will only run 16 processors which make up a "node" of shared memory. There are hundreds of these nodes which are completely distributed memory over a high performance network.
The idea is to run a large problem using message passing (usually MPI), across nodes, and multi-threaded (often using OpenMP compiler directives - easier than Pthreads for scientific computing) inside a node. I am a developer on one of the large ASCI codes, and this is the model we are targetting, although currently we usually run using all MPI (no threads) just because the threads performance is not as good.
This machine is going in at Livermore - but Los Alamos has already contracted for a larger machine (currently called the "Q" machine) which will be designed by Compaq - installed in 2002 (I think).
Using screensavers is a cool idea and all - but you can only have one screensaver set to run at a time, no? Can I run SETI@home and distributed.net simultaneously? (Not that I'd want to - but I might want to schedule some priorities so each would get equal time while I'm gone for a weekend).
Maybe if condor shipped with linux distribs, it'd make it easier for this technology to take off?
When the machine is ready for "general" use ("general" as long as you have a Q clearance!), then the plan is to move the current machine to the unclassified side, and open it up for use by the ASCI alliances and other unclassified users.
They should be able to simply add it on cluster style as you suggest, since the current machine on the unclassified side is basically the same architecture. I can't tell you for sure that it's what they'll do - but if they do, they should have about a 4-5 tflop machine for unclassified use by the end of the year.
As far as what will happen to ASCI White when they're done with it - it's only being rented from IBM - so it'll go back to Kingston or whereever...
--Rob
It is possible for unclassified simulations to be run on the machine with a Q-cleared "proxy" user running the code on behalf of someone else - but in this day and age of ultra-tight security in the wake of Wen Ho Lee and missing disk drives - that is highly unlikely to happen.
However, there is a mighty impressive machine available on the "open side" which will be used for things like weather sim, drug design, etc... In this case it will be the machine ("ASCI Blue") which is currently behind the fence, which will be moved outside. 5000+ PowerPC 604e's, and not a bad parallel environment to work in, I must say...
The ASCI program is also funding 5 university ASCI centers. These centers are targeted with solving unclassified "grand challenge" type problems which involve similar complexities to nuclear simulations. ie - solid rocket motor simulations (Illinois), Astrophysics (U Chicago), accidental fire scenarious (Utah), turbulence (Stanford), and material modeling/response (caltech). These centers get time (or "fight for time" if you asked them) on the unclassified ASCI machines.
The unclassified machines are usually just one step behind, or slightly smaller, so they don't make splashy headlines. They are, however, still very very impressive machines, and there is lots of groundbreaking research being done on them which would probably not have been possible (yet) without ASCI
I (and others) don't particularly like the fact that these machines are used mostly for nuclear simulations - but it's better than the alternative (craters in Nevada), and it's definitely helping push the envelope of parallel computing - which is all I really care about. :-)
Los Alamos has a similar contract with SGI to supply large machines for them. Sandia has a large machine from Intel, and have subsequently been concentrating on massive linux clusters (ala CPLANT) as their future.
More information here
Actually, the workers at Los Alamos are contractors - not government
employees. They work for UC Berkeley, as do Lawrence Livermore
employees.