Army Contractor To Build A 1566 Xserve Cluster
olePigeon (Wik) writes "MacCentral has an interesting article on a new computer cluster. From the article: 'Apple Computer Inc. will announce on Monday the sale of 1566 dual processor 1U rack-mount 64-bit Xserve G5 servers to COLSA Corp., which will be used to build what is expected to be one of the fastest supercomputers in the world. The US$5.8 million cluster will be used to model the complex aero-thermodynamics of hypersonic flight for the U.S. Army.'" alset_tech was one of the many readers to point to
CNET's version of the story.
Isn't hypersonic flight research better suited to the Air Force?
"[Regarding the 'cloud,'] ownership was what made America different than Russia." -- Woz
The zdnet version I read earlier said it'd run OS X, at least initially, but they might explore running Red Hat or Yellowdog.
I have a feeling that as more time goes on, more and more Apple-based clusters will use OS X. Apple continues to optimize the OS. They also continue to add remote administration features (both GUI and command line) while at the same time keeping the BSD-ness of OS X as pure as possible. (OS X is based on NeXTstep and OPENSTEP, so it does have some oddities when compared to "pure" 4.4BSD or Free/Open/Net BSD).
There are also some Apple software cluster technologies (such as Xgrid) but I'm not sure if they're hardcore enough for something of this magnitude. Apple has mainly been aiming their cluster software and marketing towards the small-scale (10 to 100 notes) research groups.
(Translation:
This is...
O
U
T
R
A
G
E
O
U
S
!
After the article about the renderfarm, I was asking myself why people didn't use the blade for factor to build renderfarms and clusters...
I know there aren't available for mac, but I seem to remember Opterons and Xeon blades were the hot topic some month ago, with dual opteron blades and all...
any reason not to use them blades to build a cluster, each blade bay connected to all other, creating a (sic) beowulf or mosix cluster of some sort ?
It takes 40+ muscles to frown, but only four to extend your arm and bitchslap the motherfucker
Man, at least you might get the A2. My term in the Louisiana National Guard got me shared TA-50 gear (we actually had to share) and i think the exact same 1911 Colt my dad used in 1964....
As the price of processing power keeps dropping these clusters are getting closer to the magical 100Tflop mark, which is what Ray Kurzweil and others speculate is required to run a human-level AI . Maybe we should start worrying about the computing projects that military isn't announcing.
I have fond memories of yelling many a "bang! bang!" and (my personal favorite) "budda budda jam!" during training exercises.
I was active duty USMC from 1992 through 1998 (aptly dubbed "Clinton's Corps"). It's good to know that the more things change, the more they stay the same. Whether you have a Republican globalist in the White House or a Democrat globalist ruling the roost, the people who need it most still get the short end of the stick when it comes to military spending. In the end, the D.C. suits in charge are all globalists with the common goal of the oligarchy in mind.
We had practically no green money (USMC money) for things like training and education, but blue money (US Navy money for the aviation side of the house) seemed to come out of the friggin woodwork. I couldn't get a new three ring binder without filling out two forms (in triplicate!) and a two week wait for the purchase to be approved, but one avionics jockey with a few too many beers in him from the night before drops a $45K helicopter battery on the tarmac and POOF!! a new battery practically materializes out of nowhere with no paperwork and no questions asked.
The Marines are well known for doing the best job with the worst equipment and no preparation. Keep up the good work, and watch your ass in Iraq.
Semper Fi!
-- Stu
/. ID under 2,000. I feel old now.
"I blieve the 25 Tflop figure is a typo. I've read 15Tflop elsewhere ... they're also using plain gigabit ethernet for interconnects, not Infiniband, supposedly because the applications they plan to run don't require a lot of I/O"
Now _this_ makes sense. I can easily believe that a different app could have very different performance characteristics, which could explain a 2x performance difference. That won't affect the Top 500 list too much, though, since it's based on standard benchmarks.
In any case, it's nice to see another Mac supercomputer. It's kinda cool watching the open source world win (since an Xserve is basically an open source machine as far as supercomputing nodes go -- nobody cares about the GUI on a compute node) that's competitive based on raw performance. Go PPC!
Enable 3D printed prosthetics!
Virginia Tech's machine sustained 10280 GFlops and peaked at 17600 GFlops. The Army's new cluster has half again as many nodes, as Big Mac did, so they are predicting a 25000 GFlop peak. If the new cluster works on embarrassingly parallel problems, they might achieve 25 TFlops. If not-- perhaps 12-15 TFlops is a more realistic estimate.
1566 is probably evenly divisible by however many racks they have alloted for the cluster.
-- "Makes Little Debbie look like a pile of puke!" - Moe Szyslak
but why 1556. It seems like a rather odd number
1556 = 1024 + 512 + 20 hot spares.
I'm aircraft maitenance in the army... i think your battalion level commander is more concerned with keeping his FMC (fully mission capable) percentage higher than he is concerned about the binder. The paperwork will get done. Having an aircraft not FMC makes him look bad.
The stuff i work with is on the same level, even an order of magnitude larger. The budget is out of this world. And about the money... every part, servicable or not, is worth 70%+ of what was paid for it and is turned back in.
I work on apache longbows, i almost dropped a battery last time in the field. There was a nasty component faliure that was hard to trace and was causing the battery controller to charge all the time, not trickle. It was still hot enough to boil an egg after 4 hours sitting on the ground.
The majority of rich people don't hoard their money. That idea is an untrue stereotype. The majority invest it, start new companies, hire more employees, expand their businesses, buy expensive cars, boats, homes, etc. and, in general, keep the economy moving.
Unlike the poor people, who, when given a tax break, hide the extra money in mattresses because they don't know what to do with it.
Okay, enough sarcasm.
The difference between a poor family, or even a lower middle-class family, and a rich family is that when the rich family saves $200 on taxes, they buy another big screen TV. When poor or middle-class family saves money on taxes, they buy *groceries*. Bush cut taxes, maybe, but the bottom 50% or so isn't any better off.
Should the top 40% pay 95% of the taxes? The top 30%? The top 20%?
Yes -- you make the money, you pay the taxes on it. Should the top 40% pay 90% of their income above, say, $100,000 in taxes, like they did in the 30's and 40's? Doubtful. Should they pay more than they do now? Definitely.
The top 50% *may* pay 95% of the taxes (doubtful) in terms of the government's total tax intake. The top 50% are not paying anywhere *near* 95%, or even 50%, of their *income*. Remember, the tax system is a bracketed system, so if the tax rate for the lowest bracket gets reduced a couple percent, *everyone*, from Jane Welfare to Bill Gates, pays less in taxes on the income in that bracket. I realize that wealth naturally accretes in the hands of the few -- I'm a realist about economics -- but I don't think we need to help that process along any. Since money naturally trickes *up*, and economic health is determined by the movement of money, why the hell are we giving the tax breaks to the people who would get the money anyway? Keynesian economics requires none of the hand-waving you need to make Reaganomics seem sensible. Giving tax breaks to the rich to "stimulate the economy" is like pouring water into the ocean and waiting for it to flow to the mountains.
How much money do you need to live, anyway? $30,000 a year? $50,000 a year? $100,000 a year? There's a certain point at which you can purchase pretty much every basic thing you could ever need (food, clothes, and shelter) -- above that, it's gravy. You sure as hell better be giving some of it back to help people who aren't able to pull the big bucks in through their jobs. Maybe the rich use less in government services -- that's mostly because they can afford to get theirs elsewhere. The more the poor are able to afford their own medical care and groceries, the less they have to rely on the government for that.
Try living within spitting distance of the poverty line, and *then* tell me that the rich deserve their tax breaks. How many plasma screen TVs and yachts do you need, anyway?
Is this really a supercomputer? Sounds more like a... supercluster to me.
At what point does linking together a bunch of off-the-shelf fully-self-contained PCs become a supercomputer? If doing so is the case, wouldn't it be a heck of a lot cheaper to link together whitebox machines, much as datacenters (the type that rent servers) tend to use whitebox servers rather than rackmount boxes?
I just feel like the term "supercomputer" is being sullied by so-called supercomputers that are nothing more than a simple cluster. Of course, I'm probably a moron, as I said earlier.
The true radness in this is how cost effective it is to biuld xserve clusters. Not only will they save lots of money biulding the thing they will save lots of money supporting it. RAD!
411 Y0UR 8453 4R3 8310NG 70 U5!! -NSA
Can anyone with an insight explain how one can come up with 1566 units for the supercomputer? It seems such an odd (as in strange, oh, you know what I mean) number? Is it determined by the budget? The topography of the network? Some magic number?
I'm sure that's not the only solution. What I suspect when I see you type that is that you like the violence.
http://www.rootstrikers.org/
They needed the thing - the first Apple knew about it was when one of their online store employees phoned up VT to check whether they'd made a typo on their request for 1100 dual G5 machines.
Volume of penetrator =~530cc
Uranium density=19g/cc so the penetrator weighs ~10kg
Kinetic energy = 0.5*10*(1500)^2 =~11MJ
Dynamite is 4.3GJ/ton, so this is 0.0023 ton or 4.6 pounds of dynamite.
11MJ are applied in roughly 5e-4 seconds, so total power is 1.65GW. Cross sectional area is about 7cm^2. Not quite as extreme as you have-the penetrator is a lot heavier but a lot slower.
I've got an older M392A2 spin stabilized sabot round in my office. Heavier than it looks :^)
"Seven Deadly Sins? I thought it was a to-do list!"
I've asked this question before and been modded as troll, but I'm serious: Is Apple the new Sun? It seems that while Apple doesn't have the broad product line that Sun does on the high-end server market, they are nonetheless making inroads into that very market. Further, Apple is sleek and sexy and has a lot of goodwill going for it, whereas Sun mostly brings out ambivalence.
I'm not saying they are direct competitors, but they are competitors in at least some respects. And it seems that Apple is profiting from sales of its products whereas Sun's biggest revenue inflow recently has been its $1b settlement with Microsoft, not from its product lines.