Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!


Forgot your password?

Cray Linux Beowulf Clusters 100

An anonymous reader wrote in to say that Cray has announced that they will be selling their own Linux Beowulf clusters. They're apparently gonna be working with Scyld on the software, and they of course have some crazy hardware (of course the name is SuperCluster, but I guess stupid names are nothing new ;)
This discussion has been archived. No new comments can be posted.

Cray Linux Beowulf Clusters

Comments Filter:
  • I'm just talkin' 'bout Cray
  • In the slides/brochure linked from their home page.
  • Beowulf killed Grendel, didn't he? He's a murderer!!!
  • Sure looks like it.
  • How about taking a couple Cray supercomps, and installing Beowolf on all of them. It would be a cluster of Crays.
  • It seems to me that there's a very good reason why Cray chose Linux for this. There is a plethora of source code for Linux apps available for it. Since they chose Alpha processors, they'll need some softwrae for it, and since there are relatively few Alpha binary software distributions, it seems quite logical.
  • Not really- it's just they aren't highly visible. The need for traditional HPC machines continues because there are some computational problems that do not scale well to hundreds of off-the-shelf clustered PC's (even clustered Alpha PC's.)
  • Where exactly do you read Cray will build Beowulf clusters?
    In the product brochure [cray.com]. First sentence....
    "Starting mid-2001, Cray will offer world's first production-orientated clusters, based on Alpha Linux and scalable to 1000s of processors."
  • Actually, with it being January in Chicago, my a/c works wonderful with windows open.
  • In other news, Microsoft has announced [] that they are going to be porting all their software to Wine.:
    SEATTLE--(BUSINESS WIRE)--Jan. 29, 2001-- Global software leader Microsoft (Nasdaq:MSFT - news) today announced plans to offer "Wine" Linux SuperOffice® systems -- highly-robust versions of their current desktop productivity software that combine blazing problem-solving speed with increasingly popular data stability, attractive price-performance and the popular Linux operating system.

    The product line will be formally launched in coming months and is scheduled to begin shipping in the mid-2001 timeframe. The company has received an early order agreement from ApplixWare and expects to announce multiple orders by the time of the product launch. Financial terms were not none of your damn business.

    ``The Microsoft SuperOffice Series targets the need for clusters with higher capability than those available in the market today,'' said Microsoft Chairman and CEO Steve Ballmer. ``Organizations around the world have told us they want the advanced capabilities of our Office productivity suite to be re-architected using leading COTS (commercial off-the-shelf) Windows Emulation technologies. This is an attractive revenue growth opportunity for us.'' The Microsoft P2C2E(TM) (Process Too Complicated To Explain) system holds the world record for sustained computer reliability and was named 2000 ``Super Duper Computer Product of the Year'' by someone or other.

    ``Starting in mid-2001 and continuing over the next two years, we plan to phase in unrivaled capabilities that enable customers in academia, government and industry to advance the boundaries of marketing and management while better managing their large, tedious workloads,'' said Willy Gates, software architect for the Microsoft SuperOffice series. ``We will combine Microsoft's Innovative (make sure you get that word -- innovative -- that's very important -- EYE ENN ENN OH VEE AY TEE EYE VEE EE) Windows-oriented architecture, proven Unicode client operating environment and advanced system software with Wine, the latest Windows-on-Windows emulation technology from those bastard godless open source poopieheads, and the highly scalable Gbberish.NET client-interconnect network from VapoCom''

    Gates said targeted capabilities include:

    • Superior sustained performance on Win32 based Intel computing environments working on various problem sizes and workloads, made possible by marrying the high-availability Microsoft SuperOffice architecture with our current Intel based clients -- based on the world's most commonly available 32-bit processor -- and API's ``network aware'' Windows facility.
    • Superior workplace capabilities, including high availability (cold starts, lukewarm performance, warm feelings, and hot tempers); global checkpoint/restart (in the event of a system interrupt, deletes all users' work and abruptly ends jobs, forcing an angry recovery); global resource mis-management; secure socket layer based network security coming Real Soon Now to even the lowliest customer sizes (thousands of processors); more or less efficient job scheduling (ok not really but who's gonna notice?), low prioritization, creative accounting, and some other gibberish that this reporter stopped writing down because he knew you would have stopped reading by now.
    Etc. Boy, some projects are just screaming to be born, aren't they?

  • How beautiful is the snowshine in your eyes, so directly current from the static in your brain.

  • This is not a cluster of Cray mainframes they are talking about!

    Cray have announced that they will be selling a Beowulf cluster made up of a big pile of API networks' CS20s. These are 1U rack-mounted PCs using DEC Alphas running at 750MHz. You could build one of these yourself at home if you cared to - Cray are simply using their name to resell a cluster of alpha boxes.
  • Aren't you something! Post a bucket of piss and bad attitude and don't event sign your name! If you've got a gripe that you think is worth our hearing you should at least be proud enough to tell us who you are. If you're not, you're just another whiny bastard who thinks I should give a sh*t.
  • So when person/company XYZ release Crayon Linux they'll get Cray on the legalese phone saying "We Own Cray-On" =P

    Just think, we could miss out on a linux aimed at two year olds! Oh. Windows. Never mind. Cray-On it is.

  • I hear these can do an infinite loop in 2.3 seconds!
    Infinite loops are one of the applications you really need a conventional supercomputer for. Running an infinite loop in parallel just means you have to run multiple infinite loops, so clustering doesn't help you. As long as people need to run infinite loops there will be a market for conventional supercomputers.
  • I supposed it would be too much too hope for Cray to name this the 'Z' line. Might make marketing tougher: "Yes, sir; I'd like to spend a jillion dollars on a Cray-Z Super Computer. No? I see."
  • My sentiments exactly.
  • Cray making beowolf clusters, huh? Does that seem a little strange to anyone else? I mean, if you are going to buy a Cray, buy a Cray!

    I agree. This is kind of like Jaguar deciding to start selling an economy sub-compact.

  • What you are saying, is that the next MS Office will need a cluster of Crays to run?
  • Hey, Geekiziod, if you don't like Slashdot.org, Then don't type it on your browser! ...or just start another lame copy of it.

    I'm not saying /. is perfect or anything, but I'm not gonna whine about something that I can't/won't do something about.

    Sheeesh! Your little manifesto is wordier and more annoying than JonKatz. Hey, are you trying to con Taco-boy into getting you on the payroll?

  • Now that I think back, you're right about the Origin. I was mislead by the fact that this is the only system that actually employs Craylink. Sort of ironic, no?

    Of course, SGI used to put the Cray label on all Origins with 64 or more processors. But I don't suppose that counts...

    You're probably also right in thinking that Cray #4 has more Cray than Tera. But I think the result is closer to Cray #2 than Cray #1.


  • Actually, the Florida recount was smacked down by 5-4, not 7-2, because the US supreme court was all wishy washy, first asking to Florida court not to impose a standard (as that would be unconstitutional), and then penalize it (and the people of Florida, and the rest of the country and the world, but playing nicely into the hands of the SNL writers for the next few years) because the Florida court, as instructed, did not impose a standard. Well duh...

    The fact is, the Florida court was letting the local agencies involved determine their own standards because that was exactly the US court told them to do. Allowing dimpled chads would have been just fine, and the US court even said as much, but it first halted the count that Saturday and then, Monday night, said "yeah okay you can keep counting but you have to have it done by the day we have arbitrarily decided is really important, which happens to be in 90 minutes. Good luck fellas...".


    I can't beging to count the dirty things that went on in this election, and though I'm a bit more sympathetic with the side that lost, I'm not about to say that their hands are any cleaner. This had nothing to do with upholding democracy or any such high minded claptrap. This was a matter of the rich white guy with the highest placed friends getting what he & his friends wanted. Democracy? No. An elected president? Hardly.

    You republicans were pissed at the failure of Clinton to win a majority vote, well lookee here -- your boy got neither a majority not a plurality, and if his cronies had allowed the counting process to finish naturally, they would not be sitting on the throne today. Support el presidente if you want to, but I'm disgusted by the whole farce of it, and can't wait for the bastards to get thrown out of there...

  • I wonder if Cray intends to use one of it's proprietary networking (like craylink) instead of 100/1000Base-t. IMHO- that would be quiet bad ass. Even more so because the source code would provide and *amazing* example (the core theory... like how cray manages memory or distributes load) for the open source super computer community.

    Also, could this help Compaq feel better that it's still keeping the Alpha alive? It's really quite sad to see such an awsome chip not in much use.
  • to make it one better...
    a set worker accidentally spilled hot grits on the beowulfed cray machine that was embedding encrypted DeCSS code into the 3-d rendering of Natalie Portman in the Episode II love scenes with Anakin...

  • In those cases you *can* tune your code to the number of available boxes and interconnect performance - perhaps by partitioning your problem space and cascading data exchanges across groups of nodes. Granted, not all problems can be executed asychronously, but in most cases you can rethink the problem in a way that avoids massive cross updates. The situation you describe leaves whatever interconnect in place idle for the majority of the time.
  • Does this make... a Crayowulf?
  • by fgodfrey ( 116175 ) <fgodfrey@bigw.org> on Monday January 29, 2001 @11:49AM (#472434) Homepage
    The big difference is service. SGI is not (and I suspect neither are the other vendors you mention) "just doing a wget for the latest beowulf tarball". When you buy a cluster from us, you can get a whole mess of stuff like service contracts. Also, when you take the thing out of the box (or large wooden shipping crate) it will "just work" and have a bunch of software like load balancing and batch scheduling with it. Yeah, this stuff is almost all open source and yeah, you can get it free on the net (probably off our website among other places) and for the majority of the people reading this, that is enough for them to build their own cluster in their spare time. The problem is businesses don't wanna do this. So the point is to add value to Linux/Beowulf/other stuff and resell it and make a proffit. Since one company does it, more follow.

    Cray has an incredible reputation in the HPC business so I suspect that some places will buy clusters from them simply because they are Cray and have provided excellent service in the past.

  • from the article:
    The company has received an early order agreement from BioNumerik Pharmaceuticals and expects to announce multiple orders by the time of the product launch

    it's kinda interesting. especially since pharmaceuticle companies spend most of their money on advertisement and management [slashdot.org].

    i guess this will be to support the accounting package that would be required to funnel all of that money into management and ad's. the extra cpu cycles can be spent rendering 3d molecules for the commercials.

    use LaTeX? want an online reference manager that
  • Hey Ray-

    Your alt. names are okay, but I just wanted to point out a flaw in your .sig (if you don't mind...).

    Air conditioners work fine when windows are open, they just don't work well. So:

    Computers are like air conditioners; they don't work well with Windows open.

  • they should try airnet.

    i hear that rocks.
  • A company called Tera recently purchased Cray, and renamed itself to Cray.

    The former supercomputer company is now actually a division of a company called cray, who most likely has other products, including this.
  • Oh well, time to start browsing +1...
    Should have done that ages ago.

    I kiss you all!
    Yuck, oh no you bloody don't!

  • Conventional supercomputers aren't just really fast serial machines. They're parallel, just with better interprocecessor communication lines and (usually) shared memory.
  • SGI lost its edge because it was inwardly focussed, in large part because Cray's problems were so large compared to its revenue generation ability, and there was not a manager with enough backbone to kill off the stuff that needed to die.

    The next time someone tells you that SGI ran Cray into the ground, just remember that Cray did most of this to themselves, with a combination of massive ego's, unwillingness to consider other points of view, and allowing themselves to be embedded in a completely dysfunctional organization.

    I used to work for SGI, but my involvement with the Cray side was very limited. So I bow to your interpretation, which is very plausible in any case.

    Still, I think SGI's "lack of focus" is kind of a secondary problem. They've never really had a managment capable of achieving focus, or any broad goal. Current insiders tend to blame this on Wall Street, which went to see Jurasic Park and came away determined to throw money at those who created the technology. Being awash in cash is sure death for any pioneer technology company -- it allows them to avoid addressing all the organizational issues that relate to their long-term survival.


  • It's so fast-

    How fast is it?

    It's so fast, it takes TWO halt instructions to stop it!

    Humor that predates the Arpanet
    www.matthewmiller.net [matthewmiller.net]
  • Heh, yeah well you could say that too, but actually I was thinking more along the lines of "gee, isn't it strange to have this high quality system now being reconfigured to run as an emulation of itself"? (Granted, I wouldn't call Office "high quality" by any stretch of the imagination, but you get the idea.) I realize that some problem sets are so massively parallel that they could be more easily attacked by a massively parallel computing system (Beowulf clusters, SETI distributed computing, etc), but I've always kind of seen these strategies as a way around the fact that the high end stuff is so expensive & thus inaccessible. I find it amusing that the makers of the high end stuff are now going after the people that are trying to emulate them....

  • There is plenty of room for cluster and compute farm companies. Besides technical skills and service which were mentioned in other comments another big differentiator is specific skills or experience in particular domain or vertical market.

    As an example, when not working on bioperl :) my day job is with a company that only builds clusters and linux compute farms for the hardcore biotech and pharma crowd. The reason people hire us is that besides the hardware geeks we also have the PhD level computational biologists who understand the algoritihms, software and underlying science. Knowing what your customer is trying to do with the cluster helps greatly in configuring and tuning it :)

    Profit margins in hardware, especially in the commodity intel platform are almost non-existant. The way you make money is with the services and software you put on top of the inexpensive hardware.

    just my$.02


  • He IS your president.

    No, he *isn't* my president. Presidents are elected, so that rules out Dubyuh here. He's a king, or a prime minister maybe, but the term president, as we've known it for 200+ years now, does not apply.

    And if you think about it (which, we both know, you wouldn't :), shouting "I'd like to kill the King" is as likely to get someone in trouble as shouting "I'd like to kill the president", so I'm not really sure what your little game would prove. I'd ask you to explain, but it's funnier for me to just undercut you & walk away. So ner.

    You are a fag

    Feeling a bit insecure in our sexuality, are we private? There there, no one's gonna ask so you don't have to tell.

  • They floundered for years before SGI snapped them up. In the early 90's, they seemed to be on the front page of the (Minneapolis) Star Tribune's Business section every other week, reporting poor earnings and talking about various approaches toward becoming profitable again, by reducing costs and expanding their reach by using scalable MPP systems. At the time, it was perceived that the HPC market had dried up because of defense cutbacks and export restrictions.

    In hindsight, it may have also had something to do with the big IBM shutdowns in Rochester (MN) at the same time. I disrecall what lines IBM cut there, but it was something along the mainframe line, which is something completely different. It may have also been the ongoing "dumping" lawsuit they had with Hitachi(?) that was regularly featured, too.


  • a beowulf cluster of these? oh yeah - fp!
  • I kinda like the name "Penguin Power Punch". Or the Beo-guin? Or the Slashguin...yeah, I like that one.
  • Don't know much about these big beasts, but the Cray supercluster being described here outweighs something like an IBM S/390 [infoworld.com] running Linux by a factor of?

    Choose your scale: ips, ops, ability to compress a 2 1/2 hour DVD to MPEG4 format, etc.

    Serious question.
  • by G-Man ( 79561 ) on Monday January 29, 2001 @11:23AM (#472450)
    ...he could call it the Bad MuthaCluster.

    (insert rimshot here)
  • to what those supercomputing gurus can make out of alpha/linux combination. I hope they use compiler designed by them, last time I checked gcc was pretty bad on alpha processors. Would be cool if they ported some of the unicos massive parallel machine code to linux/cluster.
  • (the f stands for "fourth")
  • by tstorm ( 227535 ) on Monday January 29, 2001 @11:23AM (#472453)
    Come on Taco, what's next, a story about how a set worker for Episode II accidentally spilled hot grits down Natalie Portman's pants? Christ, why not just rename this Trolldot and be done with it?
  • For some applications if you have a slow interconnect, it won't bring anything adding other boxes to the cluster, even by tuning the code. If every ten cycle every node exchange its data with evrey other node, there will be a performance degradation if the cluster has too many nodes with an ethernet interconnect.
  • by hexdef6 ( 141919 ) <crotalus@@@gmail...com> on Monday January 29, 2001 @11:21AM (#472455) Homepage
    Cray making beowolf clusters, huh? Does that seem a little strange to anyone else? I mean, if you are going to buy a Cray, buy a Cray! Of course, I guess being able to say that you have a Cray AND a Beowolf cluster is serious bragging rights

  • With so many companies selling Bowulf cluster, you got to wonder what makes them so diffence? I mean you have Compaq, VA Linux, Penguin Computers, the mom and pop store up the street and now SGI all doing a wget for the latest beowulf tarball and running scripts to build the package. Sorta pointless really.
  • damn you!
  • i'm crayyyyyyy linux beowulf clusta man. gimmie some canday

  • In case you haven't noticed, we in These United States don't live in a Democracy. In fact, when you pledged your socialist allegiance to the flag in whatever school you went to, you said, "...and to the Republic, for which it stands."

    In a Republic, the tyranny of the majority (say, for example, the majority that democratically elected Adolf Hitler Chancellor of Germany) doesn't count. It's the Constitution that trumps all. And the Constitution says that the Electors elect the President, not some tyrannical majority of the people.

    Sorry that your schooling in this was lacking. I suspect it probably took place in a government indoctrination center monopoly public school.

  • Where exactly do you read Cray will build Beowulf clusters?

    For what I'm understanding of the article, Cray will be selling a platform capable of running Beowulf. Quote: (Don Becker) ``Scyld believes the SuperCluster is a pioneering initiative that will produce a strong, differentiated platform capable of fully exploiting the best available commercial technologies, such as Scyld's Beowulf operating system. [...]" In other words: SuperCluster sounds like hardware to me.

    The article is very sparse on details, though.

    However, I am very exited about the fact another Big Name [tm] joins the Linux revolution ;-)

  • I guess it's a case of if you cant beat em, join em. Cray must be getting hit pretty hard by people building their own cheap supercomputers, so why not get into selling them?
  • In case you haven't noticed, we in These United States don't live in a Democracy. In fact, when you pledged your socialist allegiance to the flag in whatever school you went to, you said, "...and to the Republic, for which it stands."

    That's funny, I could not count the number of times I've heard the term "The Greatest Democracy on Earth" used by Americans in reference to America. The phrase is rife in American films, as well as political speaches.

    I guess they really mean "The Greatest Republic on Earth" then. That really makes a lot more sense. I always thought the USA had a lot in common with France, after all, they are both republics, and they're both nutorious for testing nuclear bombs.

    Good thing that the tyranny of the majority isn't needed to elect another Hitler, in the USA it takes a lot less, in fact a few good friends in the supreme court can get the country goose stepping in no time, no need to wait for a full election.

    Tyrrany of the majority indeed! Thanks there mister wea@allmax.com ... oops!

  • Wow more open source at the high end. And we are seeing more at the low end. Feeling squeezed micro$oft.

  • by thz ( 217824 )

    I thought this was kinda funny...

    Global supercomputer leader Cray Inc

    Gee, Cray is definitely the "Global supercomputer leader." Their fastest computer is an incredible .18 times as fast as IBM's.

    [According to the latest TOP500 Supercomputer Sites list [top500.org], Cray's fastest computer is ranked tenth.]


  • I think i am now stupider now than i was before reading this post!
  • Actually these are going to be API Networks CS20 dual Alpha 21264 machines in 1U cases.
    Here is the the API press release:
    http://www.apinetworks.com/pressreleases/pr01290 1. shtml
  • So if you can't beat 'em, join them, huh? Guess what'd happen if M$ caught up on this...

    "...Fear the people who fear your computer"
  • by fgodfrey ( 116175 ) <fgodfrey@bigw.org> on Monday January 29, 2001 @11:40AM (#472469) Homepage
    Yes, it is. Which is why it probably won't be connected by ethernet. There are a variety of interconnects such as Myrinet coming on the market now that do something called "OS Bypass". Without too many boring details, basically this allows you to map a page on the local side that "pushes" data to the remote side when you do a write to it. It allows you to send data over the network without doing a call to the OS every time. That is actually what kills you on ethernet, not the fact that ethernet itself is all that slow (especially gigabit).

    That said, at least for the time being, a single memory image system like the Cray T3D/T3E or the Origin line from us (SGI) has better latencies by a lot than Myrinet.

    The interesting thing is that as these "OS Bypass" interconnects develop, they are going to get more and more like a standard memory interconnect in a single memory image system and we'll come full circle. But I digress.....

  • This could be nice. I'm sure it will be pretty popular in more than a few markets. Just think about how shweet it will be to get a Linux/Beowulf cluster that:

    - Is prebuilt and tested
    - Works right out of the box (Or at least with very little hassle.)
    - Comes with backing/support from an old, well known vendor.

    I like the idea of being able to justify linux to management...
  • Nah, I'm still here.
  • Of course, but you just have to tune your code to use the available interconnects efficiently. The worlds largest 'supercomputer' is Seti@home's NOW (network of workstations). You view their program as a huge cluster with massively slow interconnects - yet they still manage to process huge amounts of data. Code written for a 4gb/s backplane machine isn't optimized for a 1gb/s backplane, nor a 100mb ethernet 'backplane'. - Josh Siler
  • Lets a bunch and make a beowulf cluster out of em!
  • Oh, wow... Could you imagine a Beowulf cluster of these things?

    Oh... nevermind!
  • No, this is actually from Tera/Cray. Check out their website [cray.com].
  • Well it is off the shelf components just not any shelf :)
  • I mean, if you are going to buy a Cray, buy a Cray!

    Bear in mind that there have been at least four entities called "Cray". All but one would be perfectly at home building clustered micros.

    1. The original Cray, pioneer of vector supercomputers. Probably the company you're thinking of.
    2. The same company after it branched out into microprocessor-based high-performance computing, system integration, and consulting. This entity not only made MIPS, SPARC, and Alpha-based systems, they resold Sun and SGI workstations.
    3. A loosely-defined entity never completely assimilated by SGI. The only parts of Cray SGI really wanted was Craylink [sgi.com] (as much to keep it away from Sun as for themselves) and maybe the compiler software. The rest they more or less ran into the ground.
    4. Tera Computer [cray.com], which bought the Cray name from SGI, together with the Alpha-based and vector supercomputer lines. SGI had already sold the SPARC-based line [sun.com] to Sun and kept the MIPS-based line [sgi.com] for itself.


  • I made a beowulf cluster too.

    You can read more about my beowulf cluster at this site [alignment.net].

    Slashdot is a good place to spam your own beowulf website!

  • Is that a Cray SuperCluster vibrating in your jeans pocket, or are you just happy to see me?
  • We are doing the public roll-out of the new Beowulf system that we've been working on for the past two years. Drop by our booth if you would like to see a demo.
  • by Fleet Admiral Ackbar ( 57723 ) on Monday January 29, 2001 @11:25AM (#472481) Homepage
    Well, everyone, I have been trying for a couple of minutes now, and I am ashamed to admit it, but I am completely unable to imagine a Beowulf cluster of these.

    I think perhaps my Beoimpotence may have something to do with watching 75% of that Christopher Lambert "Beowulf" movie. Thank G-d the videotape broke. If it had been a DVD, I might have gone insane.

  • 'nuff said
  • [...] (of course the name is SuperCluster, but I guess stupid names are nothing new ;)

    That's pretty funny coming from the guy that came up with "Slashdot"? ;-)

  • by Durinia ( 72612 ) on Monday January 29, 2001 @12:51PM (#472484)
    Good lineage post - I'm going to have to nit-pick, though...

    The Origin (MIPS-based) line was never a Cray product. SGI developed it in conjunction with the DASH project at Stanford.

    Also, interestingly enough, Cray #4 is actually very close to Cray #1...through the sale and un-sale to SGI, a lot of the extra stuff was stripped back off. Tera bought the name, yes, but they also brought their employee total from ~50 to ~950.

  • by kilgore_47 ( 262118 ) <kilgore_47&yahoo,com> on Monday January 29, 2001 @11:27AM (#472485) Homepage Journal
    I hear these can do an infinite loop in 2.3 seconds!
  • No, the other guy (I didn't support him either :) was elected by the majority of the people in all 50 states, almost definitely including Florida. The doubt about the true outcome of the race will never be known, but I know without any question that the way George II acceded to the office, by a 5-4 endorsement of the supreme court, is completely without precedent in this country, and most definitely not how we go about choosing our presidents. Call this guy whatever you want ("moron" works for me :), but he has not earned the title that has been bestowed upon him, and I refuse to give him the honor of the name "mister president". He simply is not that.

    I realize that early electors were appointed by the legislatures, but that isn't what happened this year. You must have been watching Rush instead of the news, or you would have saw that whole thing about the supreme court appointing him. It was big news, I'm surprised you missed it...

  • The only thing I wonder about this is latency.. assuming they're interconnected with ethernet, wouldn't latency be _terrible_ compared to a typical multi-processor machine?
  • Cray used to be big shit here in Minnesota, just like Control Data (well, that IS where Cray came from), and we've been used to the notion that they've been dead for almost a decade.

    Nobody needs million$ of dollar$ in capital for HPC anymore -- unless their software really kicks ass or they have some sort of edge in price or support, they're just another cluster vendor.


  • With the advent of Beowulf and other Linux clustering technologies, supercomputing is being redefined. There is more than one way to scale, and clustering is far more economical than traditional big-iron approaches.

    There may always be a market for the old-style supercomputers, but it is almost certainly going to be much smaller in the years to come. This announcement seems like Cray acknowledging that fact, and trying to keep up with the changes so they will still have a market when and if the big-iron approach dies.

  • you know, if you spammed with a real account, most people (as opposed to us freaks that browse at -1 thresh) might see it.
  • Good point!
  • Isn't it a requirement that Beowulf clusters be made from off the shelf components? If so, these couldn't possibly be Beowulf clusters.

  • I suspect that within the next year or so you will start seeing Infiniband based clustering solutions. Using RDMA writes, you have very fast remote memory access that does not require host processor. Imagine PCI but faster with a network like topology. Can't wait...
  • LOL = laugh out loud

    Its not really a hacker thing, more of an old internet (or perhaps usenet thing).

    HTH. HAND. (Hope that helps. Have a nice day.)

    "That fat, dumb, and bald guy sure plays a mean hardball."

  • In those cases you *can* tune your code

    I'm not sure it is allways possible.

    The situation you describe leaves whatever interconnect in place idle for the majority of the time.

    The problem is not having interconnect idle for the majority of the time, the problem is having idle nodes because communication needs too much time.

    If you want a big cluster (several hundreds or thousand nodes) for high performance computing for programs with important communication, the cluster needs a low latency high bandwidth network like Myrinet [myri.com], Giganet [giganet.com] or TNet [www.scs.ch] for example. Choosing a cheap network will waste a lot of CPU cycles.

  • by Durinia ( 72612 ) on Monday January 29, 2001 @12:20PM (#472496)
    Hang on there, cowboy. Perhaps you should look at some data before pronouncing the Cray-ons fit for burial.

    They've won two straight Supercomputing product-of-the-year awards with their SV1 and T3E lines, they have a couple of very highly anticipated (in the HPC community) product releases coming in the next year or two (the MTA-2 and SV2), and, unlike their ex-parent company (SGI), they're actually profitable.

    The "dead for almost a decade" you're thinking of probably is related to the fact that they were sucked into SGI for the last 5 years of the 90s. It's hard to hear anything about "Cray", when nobody calls them "Cray" anymore.

  • What is a Beowulf cluster?

    No, seriously, I see the trolls mentioning this all of the time, and yet I have no clue what one is..

    What is it used for, anyways?
  • Uh, duh, talk about re-stating the obvious.
  • Right, but I meant binary-only software. Sorry.
  • True that businesses don't wanna just build the cluster themselves. . . I think it's more than just that they dont want to do it, though. From the eyes of a business, it's not an option. Businesses want security, and they want to know that if the system breaks, they can have an expert show up when they need them. Otherwise, they'd have to pay to have a staffer full-time who knows how to fix a down Cray Beowulf Cluster, which is fiscally stupid. Much wiser to pay the extra money in the first place to have a prepackaged system with a service plan so that you have the experties only when you need it.
  • Of course, this is just a matter of course. Everyone, of course, suspected that it would just be a matter of time before they chose this course.

    ps. Now, of course, we're going to see more course posts by ACs on Beowulf Clusters. Yay. I, of course, am going to continue browsing at thresh 1 as a matter of course.

  • It must be me, but I am just reading SuperCluster will be based on Alpha/Linux. Nothing about Beowulf. Beowulf isn't even mentioned in the brochure.

    Yes, I do believe the system will be capable of running Beowulf, but the article and the borchure are only talking about hardware and some possibilities of the software. Nothing Beowulf specific; the hardware may even be running a Linux kernel with numa patch for what we know.

    Beowulf isn't the only cluster technology out there...

  • by Durinia ( 72612 ) on Monday January 29, 2001 @01:02PM (#472503)
    In the early 90s, they got too big. When they started out, HPC was a big GROWING market. Around that time, it started to flatten out, and investors aren't real thrilled when you anticipate growth, and it doesn't show. The market for their machines was (and is) still there, it just wasn't growing nearly as quickly.

    I think the "new" company has much better focus, and knows what its strengths and weaknesses are. Hopefully, with this new Linux/Alpha clustering, they aren't starting to branch out too far again like they did back then.

  • by Anonymous Coward
    to see a non-clustered version of this
  • Well it is off the shelf components just not any shelf :)

    Ouu..I need to find their pantry.

  • I was hoping this was about a Beowulf cluster of Crays! Now that would be something.

    Still, if you're gonna buy a Beowulf cluster (as opposed to just making one yourself), one with a "Cray" label is way cooler than anything with "IBM" on the box. At least it's an Alpha cluster; it won't say "You-know-what [intel.com] Inside".

  • It's a way to build a supercomputer by networking lots of standard computers together. For example, if you had 10 Pentium3s you could link them and get 10x the power of a Pentium3 (on certain special tasks). See beowulf.org [beowulf.org] for details. Beowulfs are used when a supercomputer is needed on a low budget, and can rival conventional supercomputers at some tasks.

"An open mind has but one disadvantage: it collects dirt." -- a saying at RPI