Truly Off-The -Shelf PCs Make A Top-500 Cluster

SLiDERPiMP writes: "Yahoo! News is reporting that HP created an 'off-the-shelf' supercomputer, using 256 e-pc's (blech!). What they ended up with is the 'I-Cluster,' a Mandrake Linux-powered [Mandrake, baby ;) ] cluster of 225 PCs that has benchmarked its way into the list of the top 500 most powerful computers in the world. Go over there to check out the full article. It's a good read. Should I worry that practically anyone can now build a supercomputer? Speaking of which, anyone wanna loan me $210,000?" Clusters may be old hat nowadays, but the interesting thing about this one is the degreee of customization that HP and France's National Institute for Research in Computer Science did to each machine to make this cluster -- namely, none.

Comments

Can you imagine? (obligatory)

[Andorion]

Can you imagine a Beowulf cluster of these... erm... clusters?

-Berj

Re:Can you imagine? (obligatory)

[zm]

I guess they'd call it "BeoDrake"...

Re:Can you imagine? (obligatory)

[Anonymous Coward]

Damn it! I was going to say that!

P.S. This "Slashdot requires you to wait 20 seconds between hitting 'reply' and submitting a comment." is annoying. I waited 30 seconds and still got this message. Lame Slash 2.0.

Re:Can you imagine? (obligatory)

[Anonymous Coward]

At first I thought this was reported on Yahoo!Serious News. Damned confusing if you ask me.

Re:Can you imagine? (obligatory)

[Anonymous Coward]

Retard.

Re:Can you imagine? (obligatory)

[Anonymous Coward]

Darn it, clustering is not all it's cracked up to be, yet there's always a half-serious "imagine a Beowulf cluster" comment on Slashdot. Yes, it's wonderful for certain types of scientific computing, but it's hardly useful for apps that aren't *specifically* written for it...and having that kind of app (generally *not* real time processing) isn't necessarily the most entertaining thing in the world (unless the research project running on the cluster happens to be your baby).

You can't use the raw floating point power here to play Quake. You can serve up static pages just as quickly on this lowly 486 as you can on a cluster, and most dynamic processing doesn't come close to requiring a cluster (Google and a few other highly specialized sytems aside). You can't *do* much with a system like this, aside from a few specialized tasks.

Yes, it is really neat that things like this can be made, and the research it allows defintely helps all of us. But as for actually working with a cluster like this...why? What was the last project you were working on that could have used a cluster? (I suspect that there *are* a few people on Slashdot that actually would use one, but even so...)

Re:Can you imagine? (obligatory)

[Anonymous Coward]

Massive parallelism would've been useful for a project I worked on over the summer-a simple game using a minmax search. You might not be able to play Quake, but the beowulf cluster of beowulf clusters of the article's clusters would play a nice game of chess. Another thing, haha, I know who you are in real life. See you in class. (no i wont tell you who i am, youll have to guess)

Re:Can you imagine? (obligatory)

[ReelOddeeo]

BeoPlex?

harmony

[vrmlknight]

"We had to face heterogeneity by spreading it over Linux and Windows too," she said. "It's not scientific, but technically it's good experience." its good to see the two getting along...

Re:harmony

[windi]

Sounds interesting.
Joint Linux/Windows cluster are especially nice because then, companies (that haven't switched to Linux) can use all of their desktop workstations in a cluster and thus increase their available CPU-power without having to shell out cash for new machines.

If I had a supercomputer...

[Anonymous Coward]

What the hell would I do with a supercomputer anyhow? Probably a lot of pong, maybe some TuxRacer.

Re:If I had a supercomputer...

[jgomez]

You can crack every FBI, CIA, and any other encrypted file you can get your hands on... hmmm. That makes me uneasy

Re:If I had a supercomputer...

[Anonymous Coward]

Some son of a bitch would die!

Re:If I had a supercomputer...

[Anonymous Coward]

You could impress all the chicks. HAHhahahahahah. heheheh. heh. hahehh. ok but seriously, you could run seti@home nicely.

Pentium 1 cluster

[Mr. Eradicator]

I've got an old POS Compaq with a Pentium 1. I say we all get our old POS's out and make our own cluster to get in that top-500 list :)

Re:Pentium 1 cluster

[Anonymous Coward]

I have a Cyrix 586 chip crammed up my ass right now. Overclocked with no heatsink. I say you reach up there and fish it out; it's kinda running a little raw in there, y'know?

Re:Pentium 1 cluster

[Mike Buddha]

Yeah, I've got an old p-166 lying around here somewhere, we could cool POS with chintzy fans from Wal-mart!

Re:Pentium 1 cluster

[gavinmead]

It can live at my house! I'll feed it and water it and love it! I promise!

Now, just start sending those donations...

Re:Pentium 1 cluster

[AssFace]

I like to call it "distributed.net"

Re:Pentium 1 cluster

[Anonymous Coward]

We'd probably have something that performs more like a dual Athlon than a real supercomputer. The power drain would be consistent with a supercomputer though. :)

It's a pitty

[morbid]

..because the official line of the HP sales people at the moment is that the entire world will be using 64-bit Windows XP within 2 years.

What was even cooler...

[UserChrisCanter4]

Was when the HP-powered cluster started assimilating some of the Compaq multi-Alpha machines as it's own.

Stephen Douglas, slashdotter, dead at 34

[Anonymous Coward]

I just heard some sad news on talk radio - Horror/Faggot Stephen Douglas was found dead in his Virginia home this morning. There weren't any more details. I'm sure everyone in the Slashdot community will miss him - even if you didn't enjoy his cock, there's no denying his contributions to popular culture. Truly an American icon.

$210,000 ??

[a.out]

How about $0 Baldric a student run beowulf at the University of Western Ontario built one on hardware dontations. It's not exactly top 500 but it still kicks ass.

Re:$210,000 ??

[nolageek]

Damn them geeks are cute. Jeff Gardiner, marry me! Um, nevermind.

Practically anyone? I think not.

[Pulzar]

Should I worry that practically anyone can now build a supercomputer?

Unless "practically anyone" has the funds, the storage room, and the manpower to maintain this monstrosity, there is nothing to worry about.

And even if anyone could build a supercomputer, what's there to worry about? We don't live in the "War Games" world where supercomputers play chess, tic-tac-toe, and start nuclear wars for fun.

Re:Practically anyone? I think not.

[pmancini]

Actually I worry that people like Saddam Hussein might get ahold of enough parts and knowlege to build a system specifically for modeling atomic weapons design. He's tried it before when he attempted to get hundreds of PS2's when they first came out.

Re:Practically anyone? I think not.

[Pulzar]

Actually, the worry about the PS2 machines was that their imaging capabilities are strong enough to be used in the missile guidance systems. I think he never actually attempted to get any of them, but US blocked shipments to Iraq just in case.

Re:Practically anyone? I think not.

[Anonymous Coward]

are you nuts? like anyone would fucking bother with your bullshit PS2? Get a life and a clue.

Re:Practically anyone? I think not.

[Anonymous Coward]

whoo, wow, aren't you bitchy ...

Re:Practically anyone? I think not.

[Anonymous Coward]

Yes I am. Rather than kick some co-worker's ass to next tuesday I can take it out the the vertiable plethora of idiots on slashdot who will belive just about anything apparently.

Re:Practically anyone? I think not.

[archen]

think I'd be more worried about someone building a cluster using AMD Athlons, and thus reducing everything for a 500 meter radius into a smoking pile of ash.

Re:Practically anyone? I think not.

[KyleCordes]

My guess is that people who understand how to use computers to do modeling for nuclear weapons design would be somewhat harder to come up with than the appropriate degree of computing power.

Knowing nothing about it, I would nonetheless guess that it's rather non-trivial.

Keep in mind that nukes were invented without the aid of a Beowulf or a Cray.

Re:Practically anyone? I think not.

[BLAG-blast]

>Should I worry that practically anyone can now build a supercomputer?

Unless "practically anyone" has the funds, the storage room, and the manpower to maintain this monstrosity, there is nothing to worry about.

It's only 256 machines, how hard could it be to steel(steal?) 256 computers. Rip off a couple of Gateway (are they still in business) and Dell trucks, maybe rob a few schools. Now that you've gotten (by fair means or foul) a super-computer that is rated in the top 500, what would you like to do with it? How about any brutt force crack that you feel like (it's not like you need some neat algorthyms with that much firepower)? How about generating insane encryption levels that FbI and CiA can't crack?

Of course, about 256 people with laptops and wireless modems, all running mandrake, sitting in the train station solving tough mathematical problems just annoy people....

Re:Practically anyone? I think not.

[Esoteric Moniker]

Mod this up, that was funny!

Or terrorists could use the Athlon cluster for a more insidious plan. Simply plug them in, turn them on and with the juice they draw, black out any major metropolitan area of their choosing, oh the horror! *g*

Re:Practically anyone? I think not.

[Anonymous Coward]

Actually, the processing power of the PS2 is impressive for how much it costs (yes, it is sold as a loss-leader like almost all other consoles). When the PS2 specs were first tossed out, there was concern about import/export of them. I personally know a few scientists who were really wishing those vector units in the PS2 could be harnessed for their calculations :)

Remember that the import/export laws rate machines basically on MIPS. Pentium Pros were bumping up against those numbers when they were the new thing. I'm not sure if they raised the numbers lately, though.

India at one time had a rather nice cluster of Sun 450s connected with Myrinet to do parallel computing. I think it was about 40 or so boxes. Anyway, even that amount of CPU power can do some impressive computation, enough that there was concern about their simulating nuclear weaponry, and this was over 3 years ago.

Anyway, a cluster can achieve pretty good performance these days. It is really hard to control the export of them because they can be bought as individual machines and assembled at a later time.

Re:Practically anyone? I think not.

[Tuonenkielo]

Replying to the question in your .sig, I'd guess fake ID and try to get drunk? (The latter par probably isn't that bad a choice if you can't do anything about world going to Hell in Handbasket.)

Re:Practically anyone? I think not.

[Andreas Rueckert]

Maybe you don't have to steal them. Think of a place where lot's of people bring their highend hardware and conect it via highspeed network connections. It's called a LAN party AFAIK? Why not run a benchmark in between two games, make it into the top 500 list and maybe accelerate this SETI project a bit...? :)

Re:Practically anyone? I think not.

[c_g_hills]

This is quite funny. Most of the PS2's processing power is in the GPU. It is designed for displaying graphics, not processing them. You would be better off using PCs.

Re:Practically anyone? I think not.

[Wargames]

Chess would be a perfect application for one of these suputers. It's hard to play chess when you are in a nuclear war, I wouldn't call that fun.

Putting them to work.

[nairnr]

Well, it seems like super clusters are becoming very easy to build hardware-wise. If you throw enough commodity at a problem, it becomes easier. I would think the biggest problem with supercomputers is no longer the hardware itself, but networking, and the programming to take advantage of the hardware. These computers still only really work for something that distributes easily. The biggest factors are now the ability to distribute, and schedule work for each node. The more nodes you engage, the more you hope your problem is CPU bound, so it will scale more.

Data transfer and message passing are such a big issue I belive the most important developments are in the networking topologies and hardware for these environments.

That said, I still want one in my basement :-)

Re:Putting them to work.

[spaceyhackerlady]

That said, I still want one in my basement :-)

Yeah. Me too.

This even more strongly underscores the current state of the art: we have astonishing hardware. OK, now what do we do with it? Other than grumble about Windows XP memory requirements? :-)

I wouldn't mind finding out. I bet it will be cool.

...laura

Re:Putting them to work.

[ogrizzo]

That said, I still want one in my basement

That'd be good in the winter, but, hey, it's going to be really hot at your place next summer!

Re:Putting them to work.

[lexcyber]

There is network solutions out and also other networks topologies. But its hard to build a hypercube or hypertorus or something like that from e-pc's. But you can do it from normal pc's where you can put in several NIC's. And if you want to move even further you can go into the realms of dolphin or myrinet. But that is a priceissue.

Dolphin and myrinet will not only solve the pure bandwidth problem but it will also cut down on the latency. Since latency is the biggest problem if you look into the cluster enviroment not the pure bandwidth. 100Mbit is enough in terms of datatransfer in most cases. BUT you get a few microseconds of latency wich is a bigger problem for sync and will produce idle clocks etc. etc.

So build yourself a torus with myrinet or dolphin for the real fast clusters.

Re:Putting them to work.

[Xpilot]

Well, one of the ways programming a cluster could be simplified is with distributed shared memory, a software abstraction layer that sort of emulates shared memory like on a multiprocessor. I'm working on such a system right now....

Imagine...

[SnoopDobb]

a beowulf cluster of these!

Re:Imagine...

[Bonker]

A Beowulf cluster of E-machines?

I dunno. It's kinda lacking when you compare it to all the other Beowulf clusters we've considered.

Re:Imagine...

[Anonymous Coward]

i wish i had mod points right now so i could mod you to -100 for being a complete dumb-ass.

Re:Imagine...

[xkenny13]

• A Beowulf cluster of E-machines?

  I dunno. It's kinda lacking when you compare it to all the other Beowulf clusters we've considered.

Maybe so, but this cluster still made into the top 500 most powerful computers.

Now, imagine a cluster of Athlon 1.4GHz machines doing the same thing ... now there's a drool factor, and probably cheaper to boot!!

Re:Imagine...

[yellowjacket03]

If you really want to impress me, than you should not only build one of these from eMachines, but all eOnes... :)

This just goes to show you

[bstrahm]

How powerful standard desktop computers are. There is only two orders of magnitude between a normal desktop computer (I refuse to call a Pentium III 733 as outdated) and a mainframe computer.

Now all we need are ways of getting local connections significantly faster (Did someone say Gig Ethernet) to allow faster communication between the nodes and we will be able to scale beyond several hundred and break the top 100. I hear 1gig NICs will be falling in price to under $100 US retail soon...

How fast do you connect to your cluster ?

Re:This just goes to show you

[dirtyhippie]

No... What really is needed is not more bandwidth, but lower latency (more bandwidth IS nice, however)... That way we could have hard real-time distributed supercomputers. Wouldn't mind one or two of those myself...

Re:This just goes to show you

[bstrahm]

Well here I thought we were staying with commodity hardware, but if we are going to go beyond commodity hardware into stuff that is engineered to provide low latency, high bandwidth, lets look at technologies like Infiniband which is engineered to avoid almost all OS latencies by delivering data directly to applications from hardware with OS bypass (almost no software between your app and the hardware anymore), and at 2 Gb/sec it isn't too bad on the performance scale. However I am betting that when this stuff comes out it will be a little pricey and go slightly over the 210,000 dollar budget

Re:This just goes to show you

[kindbud]

I hear 1gig NICs will be falling in price to under $100 US retail soon...

I hear Gigabit switches won't be...

Re:This just goes to show you

[Anonymous Coward]

they already are under $100

32bit pci version
http://queen.pricewatch.com/search/search.idq?cr =E G1032&qc=%22EG1032%22*

64bit pci version
http://queen.pricewatch.com/search/search.idq?cr =E G1064&qc=%22EG1064%22*

Re:This just goes to show you

[BrentN]

The problem with Ethernet in clustering isn't bandwidth, its the latency.

The real issue is how parallel-efficient your algorithms are. We do molecular dynamics (MD) on large clusters, and we can get away with slow networks because each node of the cluster has data that is relatively independent of all other nodes - only neighboring nodes must communicate. If you have a case (and most cases are like this) where every node must communicate to every other node, it becomes a more difficult problem to manage. To deal with this, you need a high-speed, low-latency switch like the interconnects in a Cray. The only real choice for that is a crossbar switch, like Myrinet.

And Myrinet is tres expensive.

Re:This just goes to show you

[GlobalEcho]

I'm curious. Given that neigbor communications are the most important, how do you network the machines? I mean, it seems that a useful design might be to have extra network cards in all the machines, to make overlapping network topologies reminiscent of the physical dependencies.

Or is th elatter to problem-dependent to make this practical?

Anonymous Coward, troll, dead at 12

[Anonymous Coward]

I just heard some great news on Kur0shin - Troll/ Crapflooder Anonymous Coward was found dead in his parents basement this morning. There weren't anymore details. i'm sure everyone in the Slashdot community will miss him - even if you didn't read at -1, there's no denying his lame trolling skills. Truly a cockgnome.

--EggTroll

Re:Anonymous Coward, troll, dead at 12

[Anonymous Coward]

Louis Armstrong, trumpet player and Jazz pioneer, died yesterday morning in his Los Angeles home. He was 71. Armstrong's last performance was at James Madison University's Convocation Center on March 24, 2001, where he played to a standing room only 5,000. Armstrong was helped off the stage by his wife of 20 years, and he later told a reporter for the campus newspaper "I don't know how much longer I can do this. This may be one of my last shows." His final song was his biggest hit, Hello Dolly! He is survived by his wife, 3 children and 6 grandchildren.

worry?

[motherhead]

Should I worry that practically anyone can now build a supercomputer?

Yes, you should probably worry that practically anyone can build a supercomputer. But you could mitigate all that fear with the fact that not practically anyone can whip up software that takes full advantage of it.

Thank god there isn't any off the shelf "missile trajectory" software in the CDW catalog. you would hope that any society that can whip together motivated coders to write such code already has access to some pretty spiffy kit.

(yeah i said "kit"... and I'm from Chicago... I feel like such a wanker.)

Re:worry?

[doubtme]

Yes, you should probably worry that practically anyone can build a supercomputer. But you could mitigate all that fear with the fact that not practically anyone can whip up software that takes full advantage of it.

Thank god there isn't any off the shelf "missile trajectory" software in the CDW catalog. you would hope that any society that can whip together motivated coders to write such code already has access to some pretty spiffy kit.

And in tonights' news, Saddam Hussein announces his support for the open source movement, and promises to help coordinate and fund a number of projects in a a variety of areas, including chemical analysis, epidemiology and particle physics.

Seems like a lot of effort

[Anonymous Coward]

just to run Mozilla.

Re:slashdot needs a new server farm

[Anonymous Coward]

Klerck, this is your mom. Shut the hell up, get off that damned Internet, and clean your room! Don't make me force you to lick my fish mouth again.

The amusing implication of this

[WillSeattle]

is that Iraq and Afghanistan could develop their own nuclear bomb programs.

Think about it. Mandrake supercomputer cluster + determination + funds = more fun for us.

And all the attempts by the NSA to restrict code will be meaningless.

Re:The amusing implication of this

[Anonymous Coward]

A cluster of linux pc's, no matter how much mentalmasturbation a nerd or two does, cannot be used to create nuclear weapons. For that, you need lots of enriched plutonium and smart scientists, Iraq and Afghanistan are low on both at the moment.

What is more worrying is the possible aquisition of stolen russian suitcase nukes. Can you say "holy jihad holocaust"? I thought so.

Re:The amusing implication of this

[geekoid]

if they were easy to get, don't you think they would have used one?

An interesting project

[bstrahm]

I have not had the chance to play with Beowulf clusters at all. Do I still get a local desktop on certain clustered computers ???

The ultimate Linux selling tool, every linux box in your company is a node in a cluster, add a few servers for extra speed, add a few computers to provide file I/O and backup capability, and you have one of the fastest supercomputers available to your company without having to spend an extra dime (everyone needs a desktop anyway). Can you imagine the extra cycles available for simulation, whatever when people start going home at 5 PM.

Re:An interesting project

[ichimunki]

And can you imagine the damage to your data set when some dipshit spills coffee all over his/her desktop/node, or tries to open the latest Outlook virus? :)

Re:An interesting project

[bstrahm]

Absolutely none I would hope... The dataset resides on a centrally managed server, and because they are running a Linux desktop I get to laugh at what a trojan horse virii can do to a user account on a Unix box. This can also be removed as a problem by putting a keyboard, mouse and monitor on the desktop and locking the PC into a cabnet under each desk... What the user can't touch the user can't screw-up

That is a serious problem though, and one I assume Beowulf clusters will take care of, what if a node goes down in the middle of processing, how does the cluster respond to it ?

Re:An interesting project

[tolldog]

Software such as Platform's LSF take care of this magicly... it even allows for checkpointing, assuming your task allows it. Because my render software didn't really do checkpointing, I had to add that in to my wrapper code.

We do use desktops at night to work with our render farm. Platform has some cool tools to work with for such environments. I have never tried LSF in conjunction with PVM or MPI but they have support for it, so I imagine it does pretty well.

Hofstadter's Law?

[Anonymous Coward]

Wow, that's weird. Anonymous Coward posts Anonymous Coward's obit. Hofstadter would have a field day.

-AC 4eva! w00p! 4ll j00r b4s3!

Re:Hofstadter's Law?

[Anonymous Coward]

And then Anonymous Coward himself thinks that it is weird posting his own obituary.

Oh, wait...

Ported?!

[Anonymous Coward]

Windows.

Is being ported.

And people will actually be running it.

In two years.

Yeah, right. Tell me another. I remember NT4...

Re:Ported?!

[morbid]

Well, that's what they're telling their (potential) customers just now...

Re:Ported?!

[Anonymous Coward]

I tell potential girlfriends I've got tons of cash to buy gifts for them. But, like the customers, they just end up getting screwed.

Saddam Hussein's is faster

[corebreech]

http://www.theregister.co.uk/content/2/15584.html

You'd think a top computer manufacturer would be able to beat out a part-time dictator from the third-world in gigaflops, but I'm thinking it was more a demonstration by HP that they're getting set to embrace Linux and shelve HP-UX.

Yahoo! news

[Anonymous Coward]

shouln't that be Yahoo! Serious News now?

Technicly..

[Anonymous Coward]

Technicly all you have to do is buy a single G4 Mac and you have a super computer. If you buy $210,000 worth of them, and use BlackLab linux from the Yellow Dog folks youll be good to go. Or you can just buy a few of their Briqs.

Re:Technicly..

[flegged]

Yes, we all saw the Apple ads for the G4 being capable of 1GFlop. What you didn't see, was that the Pentium III 500 was capable of ~2GFlop. Now that can run an 1GHz. You also didn't see that AMD's Athlon, having a superscalar FPU, is faster than a P3. And now they can run at 1.6GHz. The P4 has new instructions to speed up certain types of multimedia processing as well. By contrast, the G4 is only now approaching 1GHz. Go figure (as you Americans say.. :o)

An Apple is not a supercomputer.

RISC does not mean faster. It allows for simpler design which can lead to increased speed, but as we have seen, Apple have consistently failed to compete with Intel and AMD (not that they even make thier own chips...). CISC is actually a good idea, since with the huge speed differential between CPU and memory, and the introduction of cache, the bottleneck in any system is the memory bandwidth. Think for a moment : why did Intel add instructions to the x86 architecture in every iteration? Because its faster having one instruction doing something complex, than many simple ones, simply because of the reduced frequency of memory access. In todays computers, RISC doesn't mean anything, since memory, storage and network bandwidth is the bottleneck.

The moral of this story:
1: Don't believe Apple's advertising.
2: Don't believe what a Mac Zealot will tell you about RISC or some other claptrap.
3: Get ppc Mandrake if you're unfortunate enough to have actually bought a G4.

Yes, I use Macs. Daily. And I hate Apple. But my PHB is a Mac zealot. It frightens the hell out of me seeing all our company's work being stored on a Mac (OS9 (no pre-emption, memory protection, RAID, journalling, or anything you would want for a server...)).

Re:Technicly..

[nusuth]

nonsense. It is not the superiority of cisc, it is the superiority of amd and intel. Opposite is also nonsense, cpus do not directly execute instructions for the last 20 years or so, there is always a layer for translation to simpler operations. It is not the architecture that counts- after all every current cpu is risc in a sense- it is the implementation.

Re:Technicly..

[flegged]

there is always a layer for translation to simpler operations

True. Every computer interprets instructions, right the way down to the simplest level of switching a gate on or off.
The difference between RISC and CISC is the instruction set exposed to the programmer (compiler, interpreter, whatever), and thus the number of memory accesses needed to implement something complex. But if you were to believe Apple advertising (which anyone who actually buys Apples must do...), you would think less instructions is better.
I was simply demonstrating that anything Apple tell their lusers is believed, whether it's true, fabricated, or just hiding certain facts (like the relative floating point performance of the G4).

Feel free to disagree.

Re:Technicly..

[Anonymous Coward]

Pentium III 500 was capable of ~2GFlop

Please support your silly claim. (with proof actually)

You do understand that requirement in your part of the world?

Moral of the story:
1. Don't believe anything you read on Slashdot. (by flegged)

Re:Technicly..

[flegged]

A quick Google reveals the following

Enormous FPU performance with three simultaneous instructions and one GFLOP at 500 MHz (1 billion floating-point number operations per second) with 80 bit floating-point numbers. Two GFLOP with MMX and 3DNow! instructions. That at least equals Pentium III's performance with full utilization of Katmai. The 3DNow! engine has even been improved comparing to the K6-3.

Yeah, they're talking about an Athlon, and it's only 1GFlop, but it's quite an old article, the Athlon still being in Slot 1 form.

Or, from Intels site (use lynx because it refreshes to a missing page)
The adder and multiplier were placed on different ports. This allows for simultaneous dispatch of 2-wide addition and 2-wide multiplication operations. This boosts the peak performance two more times when compared to the Pentium II, and hence, it allows 2.2 GFLOP/sec peak at 550 MHz. The new units developed on the Pentium III and modified P6 units are shown in color in Figure 3

Which also reveals that the PII was capable of 1GFlop.

Moral of today's story:
Search before flamebaiting, please.

Spare cycle approach is interesting...

[alexhmit01]

They mention in the end that they are working with Microsoft to support this approach. They also suggest using spare cycles. Unlike SETI@home, where you download some stuff, work on it, send it back, this appears to be a system where the power scales linearly with nodes.

Windows support makes a difference. Take a large company (10,000+ in a single location) that has some intensive projects. In this case, they could just drop the $210,000 (call it $750,000 with installation, support, etc.) and put it in a room.

However, a smaller shop, say, 50-250 employees, being able to install this software on the staff's machines. They rarely use their computers to capcacity, and can probably contribute 90% of the CPU 90% of the time. This approach could let people doing giant calculations do so cheaply.

The real question, however, is who needs that kind of horse-power. For those that need the horse power, is the savings with off-the-shelf components meaningful.

Its a tremendous accomplishment, and I wonder how much of the changes were new (vs. Beowulf clusters that we always hear about). However, if this fills a need, congratulations, its an impressive accomplishment regardless.

Alex

Re:Spare cycle approach is interesting...

[gribbly]

There's a very good Greg Egan novel called Permutation City that takes the concept of shared computing power a long way.

I don't have time to explain -- just wanted to plug a good book of (at least marginal) relevance to this topic!

Is this a story from 1993?

[Anonymous Coward]

Uh, I was doing this in the early 90's (as were many others). The idea of using idle cycles from your workstations is beyond old. Is it somehow newsworthy because HP did it? The article makes it sound like a revelation. I'm willing to bet what I was doing was more sophisticated. My processes would relocate themselves whenever a regular user logged in and would even save the system state to prevent any lost work. Hmmm ... sounds like a nasty virus! And while I'm at it, Beowulf was nothing more than rehash as well. How far back does PVM date? Guess it is just because the name sounds cool.

Re:Katy and Physics test!

[Anonymous Coward]

You are bitter and angry. I suspect you are also a virgin.

[OT] Yahoo news posts

[realdpk]

Argh. Is it possible that when news articles come listed from Yahoo that a non-Yahoo source could be used instead, or at least added as a secondary link?

As soon as I see a blur (pop-behind) ad I quickly close the page I'm on in the naive hope that they track when people stop visiting their sites and don't leave via an ad... ;)

story should actually read...

[Anonymous Coward]

SLiDERPiMP writes: "Yahoo! Serious is reporting that HP created an 'off-the-shelf' supercomputer, using 256 e-pc's (blech!). What they ended up with is the 'I-Cluster,' ...

real winner == open source cluster software

[Anonymous Coward]

you have to admit that the free open source tools, drivers, os code, and yes for the STALLMANnanites - GNU code are the real winners here.

They make it possible.

Kind of like having a linux router project floppy and adding enough to download/accept workloads from another system. No HD required == $150 or less per node.

Re:real winner == open source cluster software

[Anonymous Coward]

I can't WAIT to hear that pakistan used a cluster of linux servers to develop the bomb after they drop it on india, because STALLMAN made it happen!!!

okay um..

[Wakko Warner]

i'll give you $210,000 so you can do exactly what with your new supercomputer?

also, who will pay your power bills?

i don't get this "drool factor" thing some people have for supercomputers... sure, they're cool and all, but they can do exactly nothing you would want or need to do on a day-to-day basis...

Re:okay um..

[dillon_rinker]

Photo realistic first-person shooters, 75 frames per second.

Re:okay um..

[sharkey]

The combined vibration of the coolings fans would make timothy's pr0n browsing that much better.

THAT'S the drool factor, so to speak.

Re:okay um..

[kin_korn_karn]

There's going to be a "photo-realistic first person shooter" released soon. It'll be played on ACRES of screens! It's a tactical shooter, with full one-shot kills, air strikes, artillery support, realistic weaponry, everything! What could be more cool?

*spit*

Re:okay um..

[Anonymous Coward]

DSP's. Being computer musician, I need many many many clock cycles to be able to do what I want with sound. I have made models of analog synthesizers that were too much for my 850 MHz Athlon. Some refining and trimming help, yes, but only to a certain point. A super computer would be immensely beneficial for those exploring sound.

Re:okay um..

[fgodfrey]

Kinda depends on what you do for a living doesn't it? I'd like to see you predict the weather acurately (well, as I guess I should say "as accurately as the standard weather forcast") over the coming week or two without a supercomputer. Or design an aerodynamic car. Or an airplane. Or a new medical drug. Or a spaceship. Or any number of defense/military applications. There is plenty that a supercomputer can do that people do on a day to day basis. You just have to be in the right line of work. (And by the way, a supercomputer *can* do everything a desktop machine could do, it'd just be rather pointless to use a machine that big as a desktop...)

Okay, so I'm curious:

[Saint Aardvark]

Sez the cost was $210k US w/o cabling...why the qualification? What *would* cabling for 225-odd boxen cost?

NEXT:clustering consumer electronics + computers

[Anonymous Coward]

"A more daunting task might be taking the model to a consumer environment, which, Richard pointed out, is full of often dormant processors like those in printers and DVD players.
"

Get funky with recursion!

[ekrout]

Imagine a Beowulf cluster of Beowulf clusters!

what scares me is...

[pXgray]

The individual machines that made up the I-Cluster are now out of date, each running on 733MHz Pentium III processors with 256MB of RAM and a 15GB hard drive. HP introduced a faster version at the beginning of this month and will launch a Pentium 4 e-PC by the end of the year.

this kind of hardware is out of date? unless i'm mistaken HP markets these e-PCs toward home users looking for light processing power, such as the ability to view web pages, read emails, and play solitaire. this looks more like a power-user rig, or something a gamer would have as a decent Q3A machine. how in the world could this hardware be obsolete? i guess i should replace the pentium III 933 i'm running because lord knows it just won't hold up to today's high powered apps! man it's almost a year old, i should start worrying...

pxg

Re:what scares me is...

[Fencepost]

It's probably out of date because processors that speed are either already unavailable or will be shortly. They could presumably underclock, but it makes more sense to just tweak the model number slightly.

Re:what scares me is...

[Anonymous Coward]

What makes it ridiculous is that even the 1 GHz Pentium III is *also* considered out of date. (In fact, Intel has stopped fabbing both for desktop use; any new processors below 1.2 GHz from Intel are intended for notebook/embedded duty.)

This has resulted in some rather odd dichotomy in processor pricing: a 1.4 GHz Pentium 4 now costs *less* than a *slower* Pentium III. This happened before with the introduction of the 100 MHz CPU-clocked Pentium II @ 350 MHz; at introduction, it cost $USD6 *less* than the 66 MHz CPU-clocked P-II @ 333 MHz (quantity one).

Re:what scares me is...

[Darth Hubris]

Ok, now ... My parent's rig is a K62-500 with 64 Megs of RAM. One of my freinds has my old P233 with 64 Megs. Those are surfing machines. The $200 machines from Compgeeks.com are surfin' machines. 733 Mhz Celerons are bit more than just surfin' machines.

Hell, the 233 and 500's are servers with Linux!

It's at #385 in the list

[CormacJ]

The latest top 500 list is here : http://www.top500.org/list/2001/06/

The cluster is at #385

Re:It's at #385 in the list

[Anonymous Coward]

Avalon was at #113 in 1998
http://www.top500.org/list/1998/11
It used completely off-the-shelf hardware.
http://cnls.lanl.gov/avalon/

Why not just a g4?

[Anonymous Coward]

You could spend $210,000 or just get a g4 from apple for about $4,000! G4s are sure super computers too. And then you can use OS X instead of Mandrake Linux. Much better.

-Adam

Re:Why not just a g4?

[Anonymous Coward]

See above, RE your rendundant comment.
Please read before posting.

distributed supercomputer cluster

[maxpublic]

What I'd like to see is a shot at a distributed supercomputer cluster utilizing the spare cpu cycles of computers on high-speed internet connections (cable or DSL). Since efficiency would be remarkably degraded by slow communication times and the fact that many of these computers would be running Office (ahem), you'd have to scale up at least one order of magnitude.

Technically I can't see why this wouldn't be feasible. It would be beyond SETI and protein folding in that the 'control center' could change what problem was being worked on at any time. It may not be incredibly practical compared to setting up specific machines in a single large room, but it would be free and have a potential user base in the hundreds of thousands or millions.

Imagine: instead of the same SETI screen output time and again, you'd get a message on your SS saying "would you like to see what your computer is working on right now? How about high-pressure fluid dynamics in environment x?"

Max

Re:distributed supercomputer cluster

[The Panther!]

Actually, this is exactly a project I've been working towards this last 6 months. Granted, I'm not an expert in the arena of parallel algorithms, but I've done some reading and have some legitimate exposure to load balancing complex jobs across workstations in my current employment environment.

That being said, my approach has been to design a C/C++ like language that intrinsically understands multithreading and remote processing, but has a minimal standard library, and is by default intrinsically secure by design. People running the client software can allow default security privileges for anyone to run under, or can set up specific execution profiles for certain people or programs, controlling cpu load and even restricting the hours they share out their machine, etc. Basically, think SETI@Home, except with any program that might be written in the language automatically taking advantage of all available clients.

Trust me, there's a _lot_ of technological wrinkles to iron out in the language to make it safe to run on another person's machine without authorization, as this is the heart and soul of the project. Rather than reinvent the distributed processing technology for many different purposes and try to get many people to run several clients, it's better to get a single client that works for all sorts of problems. I think the time is now to prepare to capitalize on spare processing power. As we plunge further into the information age of cheap hardware and don't seem to require much more power to do the same tasks (unless you run Windows XP), the ratio of spare cycles to used ones will only increase over time.

The topology of a network of distributed clients over (potentially) slow connections makes it undesirable for solving some kinds of problems, though. What it's _not_ good for are things like converting video files from one format to another, like the HP I-Cluster is. Making it good at such tasks means slackening the security aspect a bit... And even then, slow transfers can invalidate even a fast remote CPU.

Of course, I'm always looking for funding. I'd love to make this a full time job in the future. Know anybody with loose purse strings? Didn't think so. :-/

To Divx-Encode DVD rips in minutes?

[Anonymous Coward]

. . . so you can do exactly what with your new supercomputer?

A cluster of Rippers

[sabinm]

Richard said that supercomputing power could come in handy for certain tasks, like converting large video files from one format to another, that currently take a good amount of patience.

Any one else notice that this seems to be a very elaborate (and expensive) project just to bootleg encrypted DVDs?

Re: RISC

[Turq]

While I agree with what much of flegged said, his/her post implies that modern Intel/AMD CPUs -are- largely CISC devices. This simply isn't the case. Both (the AMD moreso though) make heavy use of RISC-type design and technique.

RISC does matter, or Intel and AMD wouldn't be using it.

Power Usage

[Xunker]

I wasn't able to get hard facts about this, so I'm going to throw out the question for general "gee whiz" value.

I was pondering the computrons per watt of a cluster such as this versus a real honest-to-Bob supercomputer (Something from Cray/Terra/SGI, for example). we can assume that each machine in HPs cluster uses probably 60-80 watts (because they're sans monitor), so youre looking at about between 1.2 and 1.8 kilowatt hours to power this thing. I'm not sure what a Cray TSE uses, but I have to think it's nowhere near that because of all the redundancy that PC clusters use (one Power supply, chipset, etc per Core).

Though, I'm sure if you can afford either a Cray or 256 PCs, you can afford the power bills, too. If you have to ask how much it will cost you, you can't afford it. But while CIP (Cluster of Inexpensive PCs) is cheaper, is it as efficient?
.

Re:Power Usage

[WinterSolstice]

Crays use rather signifigant voltage, but what gets you is not the power. It's the heat...

Here's some info: http://www.cray.com/products/systems/t3e/>

Regarding efficient, in a word: no.
These systems, much like an OS/390 are written and built around a whole different theory of architecture. They are also often custom made for specific solutions. The system for 'missile guidance simulations' would have different cards, different chips, and way different software from the system for 'global thermo-nuclear warfare simulations'.

Cool to play on, though.

-WS

Re:Power Usage

[Kinetix303]

there is a certain amount of storage redundancy as well

with an efficient cooling system of network booting pcs (no hard drives, no moving parts except for fans) then you have a bank of systems which is not only incredibly cheap and off the shelf but efficient as well

MS Costs

[Anonymous Coward]

So if this solution was done with MS licensing.. what would it cost? 225(machines) * $1800(Advanced Server) + $210000(initial cost) = $615000. Ouch. And I believe theres a licencing cost for Cluster Services too isn't there?

On the preponderance of "Kit"

[ColGraff]

As we all know, "kit" is a british slang term for computer hardware. What many people may not know is that it is also the secret weapon in a British campaign of cultural assimilation.

Yes, you heard me right. Cultural assimilation. The brits are sick of seeing Mickey Mouse and Donald Duck and the sexy chick from Enterprise on TVs all over the world, and they're going to do something about it.

The British invented the English language, and in many circles certain British accents are percieved as more sophisticated or upper-class. They're capitalizing on that by inventing slang terms - "kit" being among the forerunners - that other English-speaking peoples appropriate. Thus is begins.

Soon, British TV will move off of PBS, where it belongs. British computer games and hardware will surpass American in popularity. And there is nothing - absolutely nothing - we can do about it.

(In case you hadn't realized it, yes this is a joke. And yes, I know it's offtopic and will be moderated as such. But this was fun to write.)

Re:On the preponderance of "Kit"

[Anonymous Coward]

thanks, i enjoyed reading it.

Re:On the preponderance of "Kit"

[Seanasy]

Your plan is futile.

Like most Americans, when I hear the word "kit" used in a technological context I instantly think of the popular 80's television show "Knight Rider" starring David Hasselhoff.

Now if you were German...

Re:On the preponderance of "Kit"

[Eccles]

What many people may not know is that it is also the secret weapon in a British campaign of cultural assimilation.

A secret weapon? Smeg!

Re:On the preponderance of "Kit"

[pma]

Sorry, but when I hear kit I think of the products from the late lamented Heathkit company. I think it was building my own VTVM and Oscilliscope that sent me down the road to geekdom.

what about...

[Phalkin]

using a bunch of those 1U dual athlon rackmount boxes for this? seems like it would reduce the overall footprint by several orders of magnitude, as well as easily doubling (if not tripling) the power. comments, anyone?

Re:what about...

[wadetemp]

Rackmount boxes aren't usually considered off the shelf components... sure they raise the power and reduce the size, but they also raise the cost significantly.

Unless you know where you can get rackmounts case and powersupply for less than $50 each... you can't technically make a Beowulf cluster with them.

Air strikes against computers?

[Dr. Spork]

I bet you HP and many other tech companies have people who called the government telling them they should bomb "enemy" computers because they are "weapons systems." With this cluster, we see this justification could apply to any computer whatsoever.

Then, the US gets tired of bombing, and HP sells them new machines. Soon thereafter, we decide their new "good" dictator is just as bad as their old "bad" dictator, and the cycle begins again.

Re:Air strikes against computers?

[WillSeattle]

So the question then is, is this good for Open Source computing? I mean, this gives us dollar metrics like the IDG and other measurement people want, but the end work product couldn't be described as beneficial, so it's really not that good that this happens.

Of course, the same could be argued about a Win2K or WinXP hacked clone - but the utility in solving nuclear equations and modeling explosions is not as high.

Even a suitcase bomb requires that you have:
1. component parts - detonator, nuclear material, shaping material
2. supercomputer to model the charge shape and impact velocities
3. willingness to deliver the material (even if lead shielded)

We know that they have #3, this HP open source supercomputer may give them #2, now they only have to pick up #1 - maybe Pakistan or the Taliban have such and will sell them to raise cash or create more problems for us.

Re:Air strikes against computers?

[Dr. Spork]

I honestly doubt that terrorist organizations would go through all the trouble of sending off people to get educated if physics, having them build a supercluster and compose simulation software, throw up and test some designs, etc. Much more likely, they'll buy finished bombs, or they'll at least buy pre-tested blueprints.

This might be different in for a country like Iraq who already have many educated physicists, and they have a realistic chance of actually doing all this work from scratch. Of course, the IEEE is doing random inspections there all the time, but maybe they could "disguise" their number-crunching supercomputer as a 256 separate workstation terminals for all the government clerks who write email. By night, thought, it's Linuxtime.

You're right about the missing material, but I'm sure someone somewhere will be willing to sell 10kg of it... (the more people we bomb, the more likely that seems).

spork

An idea for a business plan

[motherhead]

Hey remember all those completely and hopelessly out of work Russian PhD CS grads sitting around and starving and writing strong crypto software for the Russian Mafia? You might even have heard that the Russian Mafia is always looking to explore new business ideas and strategy.

Well hell wouldn't this be a great business opportunity for both of them?Call it RMBM (Russian Mafia Business Machines), and then build cheap super-clusters and turnkey code for "specialized" clients. The possibilities are endless.

This is where you get them now: Support. You sell them the machines at a 25% markup and then charge a ridiculous annual service agreement.

From the presentation:
"Using "borrowed" Post-CCCP Mi-8TV assault/commando choppers RMBM support staff can be deployed to your corner of the desert in a matter of hours! Lets see IBM match that! Not even Larry Ellison and his personal Mig can touch that! (canned laugh track)"

I don't know, maybe not.

Q3 Hmmm

[Anonymous Coward]

I wonder how many FPS you could get in Quake 3 running on a PVM? Or for that matter, install Free-M-Ware (or VMware if you like) and install XP on it. Heck, it might even run at a half way usable speed, judging from reports from people that may not be to far from the truth..

Re:Q3 Hmmm

[Graspee_Leemoor]

The video card's bandwidth would max out with two processors- if you have a good pc and a shit video card it might already be maxed out.

When will people realize they can't just install Windows on a beowulf cluster ?

graspee

The British invented the English language?

[glrotate]

They did? I always thought it was the amalgamation
of german and french after the battle of hastings.

Re:The British invented the English language?

[Anonymous Coward]

whenever you see the word "french" in history, replace it with "latin"

it might not be accurate, no. but at least it disassociates the modern world from the stinking french.

Downturn clustering

[Daniel Quinlan]

I guess this is what you do with all of that extra inventory. Clusters coming from Gateway and Dell next.

Re:Downturn clustering

[Tablizer]

Right On Name!

Not to mention all of the dot-com fire sells.

Anonymous Coward, fag, dead at 92

[Anonymous Coward]

I just heard some sad news on Slashdot - faggot Anonymous Coward was found dead in his boyfriend's ass this morning. There wasn't any more jizz. I'm sure everyone in the gay community will miss him - even if you didn't suck his dick, you probably enjoyed it up your ass. Truly an american fag.

Notice

[poot_rootbeer]

Anyone who posts a comment containing the word "Beowulf" will be shot.

Including me.

Uh-oh.

Re:Let me get this straight...

[epsalon]

Mod this down please...

How do "Clusters" help the 6000+ people that died as a result of being burnt to death, ripped to shreds, smashed, blown to bits, jumping from 80 stories to their death to avoid being burnt alive or being pinned between thousands of tons of steel and concrete for a week?
And how does repeatedly mentioning this tradegy help them. There comes a time when we need to move on. Nothing will help those poor people, so me might as well talk about clusters and thank God we weren't there.

Unless you are going to discuss something worthwhile to humanity, such as ridding the world of disease, then shut the fuck up and stop discussing this useless shit, please.

Ahem, actually clusters can be used for genetic research will will help rid the world of disease. It's quite likely that some /.ers will be able to assist in building the machines nessecary to save lives.
What have you done to save lives today?

You should all be glad that you are alive and well and able to masturbate as frequently as you like while looking at goatse.cx, and not stuck under 100 stories of a collapsed building with your penis sliced off from shrapnel.

Most people on this site are disgusted by goatse.cx. I guess you aren't. Nevermind.

By the way, we all know that Linux is a fucking piece of shit operating system thats only used by so-called "elitists" because it looks "complex" and childish technical messages scroll accross the screen when something goes wrong. The fact is, they should be running the superior operating system, Windows, on this cluster. It could be much more efficient.

It's below my level to respond to this junk. Linux is not "complex". It's beauty is it's simplicity. It's not childish-- quite the contrary. It doesn't try to hide everything from the user, treating him like an idiot, but rather lets the user decide what she wants.
I guess you are not smart enough to realise that. Good for you. Have fun with your overpriced buggy OS with great marketing.

Anonymous Coward, fag, dead at 65536

[Anonymous Coward]

I just heard some sad dick on talk radio - Gay/Faggot Anonymous Coward was found dead in his Vagina home this morning. There wasn't many more dick. I'm sure everyone in the homosexual community will miss him - even if you didn't enjoy his dick, there's no denying his contributions to popular ass. Truly an american fag.

elephant penis

[Anonymous Coward]

Can you imagine a bewolf cluster of these?

Re:Let me get this straight...

[Anonymous Coward]

oh, the "president" said "move on" ? big deal, if he said jump off a cliff, would you?

Making a cluster work (OSCAR?)

[brechin]

I'm a first time poster, so please bear with me.

Anyway... I just got a job working with the NCSA as part of a project called OSCAR. It's basically an open-source solution to the problem of creating a cluster. I'm part of the team working on documentation and training materials for people trying to implement OSCAR. I can say, from my own experience, that installing a cluster (even only with 4 PCs) is not a simple task. OSCAR is still young as far as software development goes, but it will do the job well once it is finished.

For those interested, OSCAR 2.0 is on its way soon.

Kit == Rig

[kindbud]

As we all know, "kit" is a british slang term for computer hardware.

No it isn't. That's just the only context in which you've heard it used (translation: you read too much Slashdot, and should get out more often). "Kit" is the British equivalent of the American "rig" when used in this context. It is not used specifically to refer to computers.

Yahoo! got this story from CNET

[jlitvin]

Yahoo! didn't report this story, they just licensed it. Give credit where it's due . :-)

Ripping Not That Processor Intensive

[Anonymous Coward]

I could rip a DVD in less than 30 minutes with a single, off-the-shelf computer. Throwing more processor power at it wouldn't make it significantly faster, though a RAID might help.

You must be thinking of converting the rip to DIVX, which is indeed processor-intensive and can take 6 hours to more than a day depending on the movie and your CPU.

Re:Let me get this straight...

[apraetor]

no, but i might give YOU a nudge..

--matt

amd for less

[Jaeger-]

i think i could build a better supercomputer
for less money with amd procs/mobos/etc

1gig tbird $100
decent cheap amd mobo w/integrated vid/snd/net $100
256meg ram $25
15gig ide $50
floppy drive (needed??) $15
cdrom (needed??) $25
decent nic $20
cheap case $40

total $375
subtract 10% (due to quantity purchase) gives less than $350 total each

pay a bunch of college kids $10/hour
they'd build 2 machines/hour
so 125hrs total to build comps is $1250

$350 x 250 machines is $87,500
add in (8) good quality 32 port switches @ $200 each and you're up another $2k
add in 250ish cat5 cables for another $1k (who wants to make them, buy for $3-4 ea)

your total cost is way under $100k

or even better
use the new SMP durons, 1gig each

not much more $$ since durons are cheap
add like $50 for the 2nd proc (total $150 for 2 duron 1gig smps, unsure if thats reasonable pricing) and another $50 to mobo cost for dual smp mobo

thats $450 ea box
250 x $450 gives us $112.5k for the boxes
add in networking stuff etc
less than $125k prob

man i want to do this
need someone with $$ =P

--wayne =)

Re:amd for less

[Graspee_Leemoor]

why in god's name would you want a cd and a floppy for each node ?

Also why have a mb with integrated sound (and video) - this is a beowulf cluster, not a John Cage album...

graspee

Re:amd for less

[Jaeger-]

the point was simply that it could be done with just as much/more stuff in each box for much less...

i did put (needed??) next to floppy and cdrom, which obviously means "only if deemed necessary for whatever reason"... but apparently you can't read. personally i don't see them as necessary in a cluster, but in the spirit of matching (and beating) the other cluster in terms of components/power vs cost, i figured i'd include them in the calculations.

same thing goes for snd/vid... i'd rather have more than i need in a box, esp when it comes to debugging a box that won't boot but i can't get a terminal on it since i have no vid card...

but gee thanks for the comment.
--w

Re:amd for less

[ph0n3]

You don't need the hard drive. It's much cheaper and more efficient to use BOOTP or something and NFS similar with a centralized (non clustered) data storage server, preferably RAIDed out the ass.

Benefits include:

1. Fast drive access because of RAID stripping (net lag on a LAN isn't all that important compared to hard drive access times).
2. Data integrity (if using RAID mirroring).
3. Lower power consumption
4. No moving parts in the nodes
5. BOOTP makes upgrading the kernel on all nodes a linear time operation (the upgrade is once, but the reboots are O(n) ).
6. NFS mounted root (and other) filesystems make upgrading software a constant time operation.

However, you won't be able to use most integrated net cards for this.

I agree with taking out the floppies and cdroms, put them in the file server.

If you go SMP, don't use durons, get the full on athlons. Besides not being certified for SMP by AMD, the cache size in the Durons sucks ass for SMP (also why the dual celeron thing a while back licked).

Last but not least, spend the extra money and get some decent cases that don't take up too much space. Considering how bloody empty the case will be, the smaller the better.

again... Supercomputer?

[Boone^]

Maybe I've got a different view on things, but I find it pretty funny when a guy gets all excited about calling a Mandrake cluster a supercomputer, then gives a blech! when he announces they used e-pc's. Does it matter?

So maybe by using e-pc's the peak performance went down some, but anytime you tie anything together in a clustered environment the sustained performance dies (not just takes a hit) too. Only way to make it hit close to peak is assign each node a process that doesn't require any interaction between processes/nodes. In that case, you wouldn't need to tie them together to make a top-500 cluster. Just assign some IP's, cross mount their filesystems and call it a morning.

Besides, government agencies (and scientific companies) are beginning to realize that when you cluster 500 boxes together, you're still administering 500 boxes. When real supercomputer companies make real supercomputers, you've only got to administer one computer. Maybe that's why the term "Supercomputer" isn't plural.

If clusters keep on being called supercomputers, you might as well call "the internet" a supercomputer too since it's an environment where lots of computers are connected and running processes that don't depend on another. "Wow! Look at the sustained power of all 5000 Counter-strike servers out there! It's a super-counter-strike-computer!"

Well, I think I can type...

[xanadu-xtroot.com]

-1 : Off Topic

I'm sitting at work with no one here. A full T-1 and a Mandrake box under my desk (not to mention the 4 or 5 other *NIX boxes around me). I'm typing this because people like to flame those who are drinking.

I'm drinking Gin&Tonics. I grabbed a Lime and Tonic water at the local convience store (WaWa, for you folks on the East-Coast). I stopped by the Liquer store after that for a small bottle (plastic) of Gorden's Gin. Then I came in here to work. Nope nothing's wrong here, just a fast pipe. :-)
I keep a knife at my desk now. I use it to cut the lime.

I'm drinking. Screw karma. I finally got above 5, and now I realize I don't care... :-)

-

Oh, and to keep this somewhat on topic, this is being typed on a Mandrake box...

Beowulf Athlons?

[Starship Trooper]

I think having a cluster of Athlons would create a miniature sun with all that heat.

Beowulf Politics?

[Meefan]

I'm a student at a vocational college in New Hampshire, and I'd like to get a Beowulf cluster set up here. The obvious question, though, is how do I convince the local administration to go for it? Any suggestions? I'm thinking of having it be on a donation basis, although of course any support the schol can give will be supported. ;D I'd deeply and sincerely appreciate any suggestions. Dave

Re:Beowulf Politics?

[Meefan]

Mental note: Check for typos next time. lol.

Dave

Yawn. Another One.

[4of12]

You know this Beowulf business is getting to be pretty staid and routine by now.

In fact, I'd almost say it would be newsworthy if there were any organization (university, company, govt lab) that had not yet built "a supercomputer from the COTS components".

What I'd like to see now is more metrics (some of which the article does, admittedly, reveal).

1. hardware cost per FLOP (everyone already tells you this)
2. FLOPS per human time to build
3. FLOPS per sysadmin time to maintain
4. FLOPS per kilowatt of electricity
5. FLOPS per cubic foot of rack space
6. can it run smoothly if Bad Andy goes behind the rack and unplugs a few network connections, a few power cords to some nodes?

Everyone knows that you can spend your own time scouring dumpsters for cast-off computers and coaxing them to life, bringing up an old 486 with an ISA 10bT card as a member of your cluster. Unless you're doing it for your own educational benefit, it's just not worth it.

Don't get wrong. I love these clusters and want to use them. It's just that, in 2001, their mere existence is no longer as exciting as it was in the mid 1990s.

Now days, I care more about ease of use and ease of maintenance, taking the low cost of a Beowulf cluster as a given.

With the size of these clusters going up and the ratio of hardware cost to human time constantly decreasing, I'd be more impressed to see how a system with many hundreds of nodes was brought up in a short time, never rebooted for a year, even as 13 of the nodes developed variously problems and become unproductive members of the cluster.

"Clouds" of compute power

[termchimp]

A more daunting task might be taking the model to a consumer environment, which, Richard pointed out, is full of often dormant processors like those in printers and DVD players.

HP imagines "clouds" of devices, or "virtual entities," which could discover and use the resources around a user.

Anyone else reminded of A Deepness in the Sky by Vernor Vinge? In that story, IIR, one of the protagonists controlled a ubiquitous cloud of invisible compute "particles". Each particle was networked to the rest through its neighbors floating in the air around it.

Ok, so I thought it was a cool idea.

Re:Let me get this straight...

[Anonymous Coward]

I've started looking forward to finding your post under each story to see what dumb schmucks fall for it. Keep up the good work.

News worthy?

[tolldog]

I fail to see what is impressive about this.
It looks like the wheel reinvented several times.
For cluster installs on several machines, use system imager .
For using and controlling a cluster of machines for various taskes, use LSF .

The number of machines is pathetic too ... 225 @ 733 mhz? That makes it to #325?
How sad. I need to bench mark our render farm (200+ boxes, 120 are dual 1ghz) and see what we can come up with. I know it is higher than that... and we have a smaller install for the industry.

I looked for info to spec our machines but I couldn't find any info.... any help?

Incorrect citation

[kindbud]

Yahoo! News is reporting ...

No, it isn't. Yahoo! News is repeating a story which, if you'd bother to read the byline they wrote, was

By CNET News.com Staff CNET News.com

The article on CNET's site should be getting the Slashdot treatment, don't you think?

This Cluster and Beowulf not exportable?

[Tide]

Since you could now build a Supercomputer with OTC stuff (well, Im sure you could before today as well), will that make software like this (read: Beowulf) fall under export laws preventing exportation to countries like Iraq and China?

Any insites?

BAAAAANG!

[Anonymous Coward]

Well, at least your didn't make an AYBABTU reference.

Yes, it was = you're right

[ColGraff]

I refer to "the british" meaning the residents of the island now known as England, not in the sense of citizens of the modern political entity. I probably shouldn't have done that, but hey - it wasn't meant to be accurate anyway.

Re:Yes, it was = you're right

[pmc]

But there is no island now known as England.

Thanks for the info

[ColGraff]

You're right, I should get out more.

Of course, you know that makes the threat presented by the word even more insidious. If non-techies can use it well - I shudder to think of the potential for linguistic infiltration!

Nukes must also be TESTED

[nufsaid]

It is true that the first nukes were developed without "a Beowulf or a Cray"- BUT, to develop good nukes without doing lots of tests (and the U.S. led the world in sheer number of tests) you need fast computers. To develop small nukes you definitely need fast computers... Hence the paranoia over supercomputers in the Wrong Hands(TM).

why not tftp

[lesburn1]

why not boot ftom a tftp server?
no need for cdrom/dvd, disk drives etc.
lower power needs.
just a CPU, and enough ram for the
os and job at hand.

For those wondering about nuclear testing...

[segfaultcoredump]

No, a beowulf cluster is the last thing that one would use for nuclear simulation.

While great at highly parallel tasks that require very little synchronization between threads (think code cracking), nuclear testing (and almost all other fluid dynamic problems) generally requires all of the cpu's to have high speed access to all of the memory. So one needs a huge shared memory system (think Cray or Sun StarCat).

And for this reason, I find the top 500 list to be a bit misleading in these days of massively parallel systems. Its great as a test of how many flops the system can crank out, but it does not take into account the memory bandwidth between the cpu's, and that is often more important than raw cpu horsepower.

Re:For those wondering about nuclear testing...

[tolldog]

Ahh... Somebody else who gets it...

I find too that people assume that an "X" type of cluster will solve all problems, regardless of what they are. Each cluster type serves a purpose. Cray and then SGI spent time developing the Cray Link for a reason. Sun, IBM, HP and others have gotten into the game as well. Sometimes you need a ton of procs with access to the same memory, sometimes the task divides well.

I see this from almost the opposite side of the spectrum with rendering. To render a shot, you can divide the task amongst ignorant machines. They just need to know what they are working on. The cleverness goes into the managment of these machines. A place where the massively parallel machines would be nice is rendering a single frame. After the renderers initial load of the scene file and preping for render, the task can be divided amongst many processors on the same machine. To divide it beowulf style would throttle the network with the memory sharing over the ethernet ports.

So from my experience:

big data, long time ... massivley parallel machinebig data, short time ... generic cluster with smart master
little data, long time ... beowulf style cluster
little data, short time ... generic cluster with smart master

Re:For those wondering about nuclear testing...

[Anonymous Coward]

Starcat would be useless for fluid dynamics.

The USIII is the slowest of any of the high end CPUs. IBM's POWER, 21264's and even the P4 2Ghz whup it soundly. And the interconnect in the StarCat is respectable, but hardly impressive. SGI and IBM are years ahead here.

But marketing sells, even to people who like to think of themselves as technical, and Sun realized that a while ago.

Re:For those wondering about nuclear testing...

[castlan]

Just a nitpick...

"Craylink" is an SGI technology. As I understand, many SGI engineers were upset over that misnomer, even if it was better for marketing purposes. Well, that was the story, anyhow.

While the Craylink was used for the Onyx2/Origin(200/2xxx) series, they rechristened the interconnect for the Onyx/Origin 3xxx series. It is now called NUMAlink, referring to the ccNUMA architecure utilized by modern SGI systems.

I'd have to agree with the rest of your post, and add that whether long or short timeframes are involved, a ccNUMA style large unified memory is good for large datasets, whether the system has many or few CPUs. 64 bit architecture (64 bits for addressing memory) also comes in handy in such situations. Less data shuffling for huge datasets is a Good Thing.

Re:For those wondering about nuclear testing...

[Anonymous Coward]

With a bit of thought, CFD problems are highly partitionable and well-suited to clusters like this. Rendering graphics for this sort of thing is what really requires the large memory spaces. See

http://www.fluent.com/////software/fluent/fl5ben ch /new.htm

and also

http://www.hp-partners.com/fluent_public/

e-machines

[Anonymous Coward]

256 e-machines... what is that, like 768 years of compuserve/msn membership?

Power consumption of old CPUs

[muffel]

I'm wondering: if you use many old CPUs (486, early Pentiums) vs. not so many recent (PIII/Athlon, ~1GHz) wouldn't you pay for your elecricity bill more than you saved on the hardware?

Is there anything like a MIPS/Wh rating for CPUs? (Would thermodynamics dictate a certain minimum?)

With a seperate power supply and hard disk per CPU (i.e. complete box) I would imagine that old PCs generate a *lot* of heat per CPU cycle.

Has anybody done measurements/calculations on this?

Re:Power consumption of old CPUs

[jmv]

Actually, the best MIPS/Wh is probably with the slower versions of the current laptop chips. Maybe portable G3/G4?

Also, I don't think you'd get much useful stuff done with early Pentiums and 486. Consider that a P4 2 GHz has 20 times the clock speed and probably does twice as much per cycle, so it's ~40X faster. Now, if you connect 40 P100 together, unless your problem is completly parallel (like breaking keys, as opposed to most linear algebra), you're going to lose at least a factor of 2 there. This means that in order to equal 1 P4 @ 2 GHz, you'll need almost 100 Pentium 100 Mhz. This means that 10 P4 would be like a thousand Pentiums. At these numbers, it's going to cost so much in networking and power...

I'd say (pure opinion here) the slower you'd want to have today is something like a Duron 1 GHz and the best MIPS/$ is probably with a bunch of dual Athlon 1.4 GHz (A dual is not cheaper that 2 single, but you get more done because of parallelism issues).

Re:Power consumption of old CPUs

[Paul Komarek]

My experience doesn't suggest that the P4 does twice as much per cycle. I'm seeing P4s do a fair bit less than the P3 per cycle, and the P3, P2, and PPro cores didn't seem *that* much faster per clock than the original Pentiums. My gut tells me that the P4 doesn't do any more than the original Pentiums per clock cycle, and the only thing they have going for them is Intel's ability to manufacture them at high clock speeds.

If you really want a cpu that does a lot in a single cycle, look at the IBM POWER series. IIRC, on the floating point side, a 2xx MHz POWER III is darn not too far from an Alpha 21264 at 733 MHz. And now there are 1.1GHz and 1.@GHz POWER IV chips, in the new IBM p690 machines. I don't know how they compare to the POWER III per cycle, though, because the POWER IV opens a whole new (good) can of worms.

-Paul Komarek

Re:Power consumption of old CPUs

[jmv]

I'm seeing P4s do a fair bit less than the P3 per cycle, and the P3, P2, and PPro cores didn't seem *that* much faster per clock than the original Pentiums.

For "unoptimized" applications, that may be approximatly true. However, if you're going to build a cluster, you're also going to optimize your code for it. What kills the P4 is branch misprediction. However, by carefully writing your code, you can avoid most of these problems. Also, most of the big clusters are for numerical code, for which branch predictions does well (plus you can do lots of loop unrolling).

Another thing that the P4 (and PIII) has is SSE. On a P4 2 GHz, you can do theoratically 8 gflops. In practice, if you write good code, you'll bet between 1 and 2 gflops. On a plain pentium, the FPU is not pipelined, so a P100 (I'm guessing here) probably has a theoretical maximum of ~25 mflops with a performance for real code around ~10 mflops. That means the P4 is probably 100-200 times faster at floating that a P100.

Of course, you're right in saying that other architectures are probably faster that P4. ...and by the way I'm not saying that the P4 is great... but if you're doing numerical stuff and using SSE, it's VERY fast (in my experience, 3DNow! has been faster than SSE at the same clock rate, but 2GHz is too much higher than the fastest Athlon).

Re:Power consumption of old CPUs

[Paul Komarek]

My experience with optimizing research code for a particular platform: it will never happen, unless you're starting from scratch and expect your code to have a short lifetime. We've got libraries that are several years old, written to be portable across various unix and Windows platforms, running on MIPS, Alpha, x86, SPARC and PA-RISC. These libraries aren't optimized for any particular platform, and nobody has time to mess with platform-specific optimizations.

I've never tried optimizing for SSE, but someone in lab did once. He reported higher performance when doing his computations element-at-a-time than vector-at-a-time. His conclusion, for his particular application, was that memory latency was killing SSE. He was better off doing lots of work on a few numbers, than he was doing fancy stuff to a lot of numbers with SSE. On the other hand, some people have had some luck with SSE optimized FFTs, or so I've heard.

At 2GHz, I'll bet that you're better off doing element computations than vector computations because of the radical difference in memory-versus-processor performance, if the P4's L1 take more than a cycle to feed the registers. Otherwise, do whatever fits in L1 and can be prefeteched -- like elementwise computations on long vectors. Anyone have any real or anecdotal evidence to refute or support this theory?

In the end I think that platform-specific optimizations are a waste of time for research code. I seem to remember some people eventually including hooks in BLAS or LAPACK to allow the user to specify cache sizes; and FFTW does some runtime experiments for optimization. But my guess is that, overall, SSE, 3DNOW, and even AlitVec are irrelevant to most computer researchers. I'll bet their highly relevant to most embedded engineers or many robotics researchs, i.e. people targetting specific hardware.

-Paul Komarek

Re:Power consumption of old CPUs

[jmv]

I've never tried optimizing for SSE, but someone in lab did once. He reported higher performance when doing his computations element-at-a-time than vector-at-a-time. His conclusion, for his particular application, was that memory latency was killing SSE.

Maybe his problem was really special, but most likely he didn't know how to write SSE code. First, if you write your code correctly, the worse you can do is a bit better than the x87, because you can use SSE with scalar and take advantage of the linear register model (as opposed to stack).

The only time I've converted some code for SSE, I got a 2-3x improvement in speed. There is one thing you really have to be careful when writing SSE code: ALIGNE EVERYTHING TO 16 BYTES. That's very important. The "movaps" (move aligned packed single) is more than twice faster than the "movups" (unaligned) when data is unaligned (movups on aligned is not too bad). That makes all the difference. Also, sometimes you need to change the way you do your computations.

For the case I have (optimizing a speech recognition engine), just by changing the order of of the loops (inner vs. outter), we got a 2-3x improvement (still with x87) because of increased L1 hit rate. Then when switching to SSE, it ended up to a 5x improvement over the original code. Had I just re-written the code in SSE (with cache optimization), the gain would have been around 25%, because memory would still be the bottleneck.

As for libraries not being optimized, just look at FFTW (www.fftw.org) or ATLAS (www.netlib.org/atlas) and you might change your mind.

Were they shipped with OEM Windows?

[nolife]

They could reduce the cost by another 10% ($89x225=$20025) if they would return their *unused* OEM Windows licenses to MS..

You can run wine on a Beowulf cluster

[Anonymous Coward]

hell, run bochs

Glad I'm not alone.

[FU_Fish]

It makes me happy to read about another Mandrake-based cluster. I was starting to think that I was one of the only people in the world to take advantage of Mandrake's optimizations and superior (IMHO) distribution. I just finished building a 16+1 node Athlon 1.33Ghz, Mandrake-based beowulf cluster. Granted, we didn't make the Top500 w/ just 16 nodes, but it's fun none-the-less.

Re:Why not just a g4? Powerpc best , IDIOT

[Anonymous Coward]

most of the entries in www.top500.org list of clusters use POWERPC chips... almost NONE are intel.

check it out and download all 500 and read them and see , moron.

The reason????? they are the fastest.

Re: mandrake

[createaccount13]

The Linux operating system was born in 1991 and was created by one man, a Finnish student coincidentally named Linux Torvalds. Since these humble beginnings, a multi-million dollar industry has sprung up to exploit the commercial potential of Linux, but until recently Linux has eluded mainstream acceptance. However, due to the recent economic downturn together with uncertainty over changes to Microsoft's pricing policy, Linux is now being touted as a serious contender to Microsoft Windows. While there are many other alternatives to Windows, including BSD which is based on SUN's (Stanford University Network - correction by bc) server-grade Solaris operating system, none have commanded the same level of media attention as Linux. Linux Mandrake is just the latest in a long line of quirkily christened versions of Linux. Previous versions of Linux have been named Red Hat, Slack Ware, Storm and Coral. In stark contrast to the mundane names such as 98, ME or NT preferred by Microsoft, the crazy names of each Linux release hint at its renegade nature.

My foray into the world of Linux began by downloading a "CD image" from the Linux web site. But don't worry, this isn't software piracy, it's perfectly legal! Linux is shareware, meaning that it can be freely redistributed without fear of a visit by the Business Software Alliance. The free availability of Linux is a major reason for its popularity among cash-strapped students and self-styled anti-capitalist hackers.

Before installing new software, it is always advisable to read the documentation. Unfortunately, an unpleasant surprise was in store for me in the "required configuration" section of the manual. I was shocked to learn that Linux Mandrake only runs on Pentium processors, meaning that my hopes of testing the water with my old Gateway 486 were dashed. Furthermore, a whopping 32 megabytes of memory are required to run Linux! Although the advocates of Linux self-righteously boast the efficiency of their chosen operating system and deride the "bloatware" produced by Microsoft, it appears that their claims are blatantly incorrect. Although my humble 486 will happily run Windows 95, it seems that Linux requires far more powerful, and more expensive, computer hardware. Is this really the sign of a lean, mean operating system? Of course not.

Sadly, not even being able to install Linux is just the first of my many complaints. A brief perusal of the features of Linux Mandrake reveals that Linux is sorely lacking many crucial productivity applications. For example, why isn't the industry standard web browser, Internet Explorer, included with Linux? Despite the best efforts of the experts at the Internet Engineering Task Force to encourage adoption of the Internet Explorer standard, the creators of Linux seem to think that they know better. By refusing to adhere to recognised standards, Linux is simply undermining its own credibility.

Similarly, almost all of the world's most popular and widely used software is completely incompatible with Linux! It may surprise you to learn that your copy of Microsoft Office, Outlook Express, or Lotus Notes will not work under Linux. Those who wish to use their computer for recreational purposes are also out of luck, for almost all of the most popular games are unavailable for Linux. Although a wide range of software is freely available for Linux, these pitiful offerings are mostly unfinished, unreliable and do not bear comparison to their commercial counterparts.

Computer security is also an area that seems to have been overlooked by the developers of Linux. In these times when hacking and viruses are commonplace, it defies belief to learn that no anti-virus software is available for Linux. To add insult to injury, there is no Linux version of the popular ZoneAlarm firewall. By using Linux, you are issuing an open invitation to the hordes of ne'er-do-wells on the Internet.

The shortcomings of Linux are obvious. Without even installing Linux Mandrake, I have exposed several fundamental flaws. Surely it is not too much to expect that, after ten years of development, the creators of Linux would have addressed these problems? The real question that the prospective Linux user must ask himself is, "Why bother?" After all, Microsoft Windows comes free with most PCs and there simply isn't a need to replace it, particularly not with a product of inferior quality.

Although it is always tempting to support the underdog, Windows XP will be the deserved victor in the battle ahead. I recommend that those Adequacy readers who are hoping to upgrade their operating system patiently wait for the release of Windows XP, rather than foolishly wasting their time, effort and money on Linux.

crushed ego

[paulydavis]

"The individual machines that made up the I-Cluster are now out of date, each running on 733MHz Pentium III processors with 256MB of RAM and a 15GB hard drive."

I run a 750MHZ machine I guess ill soon be out of date. MY GOD i have to buy one one so i can eek out that one extra frame rate in quake (even though i cant see it)

Re:Katy and Physics test!

[Anonymous Coward]

You are a dork and a loser. I suspect you are also a faggot.

Re: mandrake

[cgleba]

Linux has nothing to do with this subject at all.

Besides, if you decide to flame Linux, at least know what you are talking about.

1) Mandrake, Slackware, RedHat, etc are not versions, but atonomous distributions.

2) The Mandrake distribution is specifiaclly compiled for a pentuim. Get Red Hat if you want to run it on a 386+.

3) There is no anti-virus software for Linux because there are no viruses (the concept of users and limited file access to a user prevents viruses from doing much harm).

4) Linux has an excellent firewall implementation called "iptables". Furthermore it's a stateful packet filter / NAT that does many things most firewalls can't.

5) IE is not a standard. It's a web browser.
The IETF NEVER encouraged the "adoption" of IE.
Standards are CSS, HTML, etc which the Mozilla, Netscape, Konqueror, Galeon, Nautilus web browsers all stick to.

6) As for the apps, you have many choices. Here are some of the MS counterparts:

IE: Mozilla, Netscape, Konqueror, Galeon, Nautilus
Outlook: Evolution, Kmail, Netscape Mail
Excel: Gnumeric, Kspread, OpenOffice
Word: Abiword, Kword, OpenOffice

Ignorant idiot!

I'm sorry.

[Ziviyr]

But I have to ask you how much you know about the bandwidth of a 747 full of CDs.

How many CDs can be crammed into a 747?

Re:I'm sorry.

[Anonymous Coward]

If the cd's have a radius of 6 cm and a thickness of 1 mm, what's the volume of a 747 cabin? we could calculate it then.

Re:I'm sorry.

[Ziviyr]

If the cd's have a radius of 6 cm and a thickness of 1 mm, what's the volume of a 747 cabin? we could calculate it then.

If you can find 6cm CDs, most are 12. A slight few are 8. I'm betting efficiency is in the court of the larger ones.

Then you just have to consider how many CD drives of a specified speed are necessary to read/write all those CDs within the time it takes for the 747 to leave and return. Of course making sure the systems hosting those drives have the bandwidth to forward that data to whatever the goal is. :-)

Re:I'm sorry.

[c_g_hills]

Forget CDs, use DVDs. 'bout 13 times more storage.

Just as cool as it was in 1995

[marcelk]

When we built a 256 node system out of 75MHz off-the-shelf pentiums. This was done during the International Olympiad for Informatics at the University of Eindhoven in an attempt to make the biggest arcade game. The system ran a custom-made game that could be played by 256 players sitting in front of this monster. For some impressive pictures, check out

this page

Monstrosity is right!!!

[Anonymous Coward]

The sight of some of these clusters is terrible!!! Yuch! You will be divorced before one of these babies takes up any of your living space! And if you got one of these going you will soon be divorced.

Games

[Anonymous Coward]

What kind of gaming could clusters support?

Re:Katy and Physics test!

[Anonymous Coward]

You two would make a perfect couple in that case.

Embeded wrist watch cluster.

[Anonymous Coward]

Two hundred million wrist watch communicators running Linux is the #1 on the top 500 cluster list. The cluster was created to make sure husbands got home in time for dinner.

My sister has a "supercomputer"

[SomethingOrOther]

Should I worry that practically anyone can now build a supercomputer?

Listen mate, my sister has a supercomuter.
Its a PIII with a 1gig clock speed.
As far as I'm concerned, that is a seriously powerful bit of kit that nobody outside a serious mathematition would ever need. My only worry is that she only uses it to run M$ Word and Lookout Express. (Like eveyone else who is talked in to buying such a ridiculously powerful computer for such mundate tasks)

You cant stop progress!

I think I actually have a use for a cluster

[King Of Chat]

I haven't used Beowolf at all so I'd appreciate any advice you guys could give. Years ago I worked with Transputers - which have similar problems: throwing the data round the transputer network is the bottleneck.

I work on actuarial calculations - which tend to be just a shedload of FP calcs done over and over again. Monte Carlo simulations are popular at the moment - which means really just running the calcs a few thousand times with some random elements changing, then getting a few percentiles out at the end. The result being, you end up doing 5 billion plus FP ops to get 4 numbers out the end. We'd like to produce this as a web service ('cos we have the client data - they just need a simple SOAP request saying what numbers they want) with a 5 seconds response. Yeah, I know.

Assuming I can design so that bandwidth isn't too much of a problem, does anyone think this will work? Is there latency inherent in the clustering software itself? Would I be better off looking at some sort of hardware vector processor, or writing my own stuff to send the data round?

I'll probably get modded down for ignorance and having the nerve to ask an actual question, but if anyone's got any insights, I'd appreciate it.

CPU not the limiting factor

[maddogsparky]

You have to consider I/O when you make your statement about the P4 2 GHz being 40x faster than a P 100MHz. To my recolection, the buss speed is 25-66MHz on a P100 and a P4 is, what, 250 MHz? That is only 4-10 times as fast. Unless your application has (reletively) little data and requires lots of computing, a cluster of 40 P100s could beat a single P4 2GHz by a significant margin.

For example, the P4 would probably take prize for a chess game (unless it requires a lot more data than I am thinking it does). On the other hand, for a weather simulation, I would bet on the cluster.

Re:CPU not the limiting factor

[jmv]

On the other hand, for a weather simulation, I would bet on the cluster.

No way. Weather sinulations involves lots of linear algebra on huge matrices. When you parallelize that, you need a lot of communication between the nodes. With a cluster of P100, communication will kill you right there (10x of so penalty). It's not that much the network bandwidth, but the latency. Weather simulation is one of the hardest problems to parallelize and that's why until recently, SMP was prefered to MPP (and of course clusters of small workstations).

As for the memory bandwidth, depending on the problem, sometimes the L1 is really effective.

Why 256 machines ?

[da5idnetlimit.com]

Last time I looked, they told me it was the limit of the swith : 256 ports.

But I thought you could scale the switchs...
using 256 / 512 / 1024 computer coud be possible with a 256 ports + fiber interconnect.

So am I wrong assume you could pile up 2-3 256ports switchs, or is it possible ?

Will they Open Source ?

[da5idnetlimit.com]

Or do will we send them the EFF in addition to the Marines ?
8)

Wrong.

[Carbon Unit 549]

...nuclear testing (and almost all other fluid dynamic problems) generally requires all of the cpu's to have high speed access to all of the memory.

You are simply wrong here. In computational fluid dynamics we use clusters to solve elliptic PDE's by partitioning the domain and iterating.

As for nuclear testing, I guess it's just a coincidence that DOE created C-Plant and Avalon.

Shoot this guy!

[Kayser_Sose]

Yeah, you heared me.

Yes there is

[ColGraff]

The United Kingdom consists of Wales, England, the Isles of Man, Scotland, and part of Ireland. So There!

Hail Brittania!

Re:Yes there is

[decoydog]

I believe that England, along with Scotland, Wales, and Northern Ireland form the United Kingdom. England, Scotland, and Wales are on the island of Great Britain. Now if UK-English defines England as an island, then I stand corrected and would like someday to visit the Island of Scotland and the Island of Wales.

Athlon Cluster?

[Anonymous Coward]

Check out the 439th entry at the top500.org then.

Here's a snap from the in-house sauna
http://matsu-www.is.titech.ac.jp/~sohda/news/2001/ 09/presto3/P9290183.JPG

Fuel for the debate

[dasunt]

Intel has a list of specs on their processors, including MIPs and other stuff for some processors.

An example:

100 MHz [Intel Pentium] (166.3 MIPS, 3.30 SPECint95, 2.59 SPECfp95 on Xxpress 1M L2)

Juno.com already does this

[Anonymous Coward]

Um, Juno.com already does this.

check out "http://help.juno.com/privacy/agreement.html" and do a search for "Computational Software"

Legality... what does this do now?

[ImaLamer]

I know selling a supercomputer to 'rouge' or communist nations is illegal, but what if a nation just orders a bunch of PCs?

Question: I know there was a story on /. about PS2's [sony, not IBM] being made Taiwan or some area and the threat of high end chips going to China. There has been talks of Linux running on PSX or some such cheap hardware so that it may be clustered in Iraq.

[now the q:] Where is the line drawn? Aren't there computers in nations such as Iraq? I just can't believe that he/they can't get 10, 200, 2000 PCs and hook them up. People smuggle drugs, can't they stow a pc? You wouldn't even need to transport the case, - hell you could build a case from wood or adobe [Native Americans made...]

So where is the line drawn? And why can't I store buy a simple cluster? Anything?

I know chip prices fall usually because AMD, Intel, etc stop making them and go onto new lines but why not 4 Athlon's [Slot A, cheap] onboard at 750 MHZ? I know the limitations; software needs to be writen for that, speed isn't 750Mhz x 4, etc.

I want an almost super computer!

Tell my Kernel to use one for disk i/o, one for net, one for X... one for seti i guess. Advaced crontab, AT 6PM RUN FETCHMAIL ON CPU2! Damn, I wish I could write code!

Re:Let me get this straight...

[Anonymous Coward]

Even though I *do* run Windows (2000 Professional at the office and XP Professional at home), I wouldn't make comments like yours (and no Windows user I know would, either).

And by the way, Windows 2000 Datacenter Server *does* support clustering.

These were NOT E-Machines

[Anonymous Coward]

idiots, these werent e-machines, they were the hp e-pc found here -> http://www.hp.com/desktops/professional/e-pc/e-pc_ 40/

Real Research

[Anonymous Coward]

It is very plain to see that you are very ignorant in your comments regarding the Macintosh. If you are really interested in learning and understanding parallel processing why don't you take a look at:

http://exodus.physics.ucla.edu/appleseed/applese ed .html

After you are finished reading this you will understand why some of us, who are doing real work, use Macs for some parallel processing projects.

Re:Real Research

[flegged]

I never said you can't make a supercomputer out of Macs, in the same way you make a supercomputer out of e-machines. I was responding to the fallacy that a single desktop computer is a supercomputer. By definition, a supercomputer is one which is far more powerful than a desktop.

A single Mac is not a supercomputer, despite what Apple say. A single G4 is not even as powerful as a single P3. But string either of them together, and you can have a nice cluster.

Unfortunately, Apple shipped SMP machines long before any SMP-capable OS. And didn't tell their lusers that that second G4 was idling at 500MHz. They just run photoshop filters and say 'I'm sure that one went faster, just then'.

The reason people think Macs are good for Graphic Design is because Graphic Designers think they're good.

We have a nice setup here at work

[Yunzil]

We have 72 1U rack-mounted dual 1GHz PIII boxes running Red Hat 7.1. Most of which are sitting idle at the moment, although that will change in the near future. :)

I wonder where we would be on that Most-Powerful-Computer list.

so now what we need is a central bank of ram

[Anonymous Coward]

All the cpus could share a bank of super quick ram. Shared data would be copied there. Direct pci interface. read access to all. regions of writability for each cpu. all cpus have access to all data at super fast speeds.

throw those network cards in for tcp/ip.

so, EEs get out there and make a pci ram interface and a 40GB ram box. send me one when you've got it.