Follow Slashdot stories on Twitter

 



Forgot your password?
typodupeerror
×
Mandriva Businesses

Mandrake Announces Turn-Key Clustering Distribution 187

joestar writes "According to their website, Mandrake and partners (Bull, INPG/INRIA...) have launched an 'easy-to-deploy easy-to-use Linux Clustering solution,' that has already been tested on a 40-node cluster. Of course, it's published under the GPL, comes with parallel applications, and is available for download as an ISO. It seems the project is financed by French government. It's great because I've always dreamed of having my own supercomputer at home."
This discussion has been archived. No new comments can be posted.

Mandrake Announces Turn-Key Clustering Distribution

Comments Filter:
  • Imagine.. (Score:5, Funny)

    by teamhasnoi ( 554944 ) <teamhasnoi AT yahoo DOT com> on Wednesday October 30, 2002 @05:00PM (#4567833) Journal
    a single PC running this!

    Gotcha!

  • No SCSI (Score:3, Informative)

    by gorillasoft ( 463718 ) on Wednesday October 30, 2002 @05:01PM (#4567841)
    Hardware restrictions

    A fast ethernet network switch

    NO SCSI drives in nodes

    A PXE bootable network card (intel / 3com)

    A 3D accelerated video card for the virtual reality 3D engine


    Seems a bit limiting - no SCSI drives?
    • It seems the project comes in three differents stages, and this is only the first stage. I hope SCSI can be included in next stage!
    • Re:No SCSI (Score:2, Insightful)

      by ltwally ( 313043 )
      I'm just curious how many super-computers out their have fancy 3D grafix cards, and only run on IDE drives...? I think may-haps I'll wait for version 2.. just my two cents.. your mileage may vary
      • One component of the forthcoming TeraGrid is a cluster of computers with NVidia GeForce cards at Argonne National Lab for visualization. So there is one...
    • Re:No SCSI (Score:5, Informative)

      by benploni ( 125649 ) on Wednesday October 30, 2002 @05:07PM (#4567922) Journal
      You've obviously never built a real computational cluster. Real cluster nodes are better off not having any drives at *all*, as they are the only moving part in the mix. It boots PXE, loads a kernel, and nfs mounts root.
      • Re:No SCSI (Score:3, Insightful)

        by SquadBoy ( 167263 )
        You are , of course, right. But they also have a 3D card included in those specs. Seems odd to me. I'm thinking those must be the specs for the master node. (It has been a long time since I've thought about clusters not sure if that is the right term but I think you know what I mean) In any case it looks like someone is confused/wrong on this.
        • Re:No SCSI (Score:1, Insightful)

          by Anonymous Coward
          hehe, wrong again. The 3d-cards is for fast 3d-rendering (doh) - they use the program Netjuggler to utilize opengl on multiple machines.
      • And I assume you are going to dump all your data over the NFS network?

        Gigs and gigs of file I/O are going to kill you, as a lot of parallel applications will dump out scratch files to disk. Tell me, is it faster to dump data to a ATA133 or U160 Disk, or gigabit ethernet to a congenested NFS mount?

        I can cite examples, such as g98, Amber, Fluent, and other scientific apps you might want to run an a beowulf who tend to dump out check point files.

      • yes and no (Score:2, Interesting)

        by ananke ( 8417 )
        yes, they're better off, because you don't need to do jack on the computational nodes.
        no, they're not better off, because where is your swap going to be? or temporary data generated by individual jobs?

        i run a 100 node computational cluster. the nodes boot with pxe, then if they are already installed, they simply boot to their local scsi drives. only /home is mounted via nfs from a disk array, so each individual node has access to the users' programs. we use fast ethernet for that, since it doesn't put that much strain on the network, and myrinet for the computational nodes to exchange data.
        i've had more problems with ram and myrinet than the 'moving parts', aka drives.
    • Re:No SCSI (Score:2, Interesting)

      by Atev ( 157440 )
      I wonder whether non-volatile storage makes sense on the nodes at all- isn't it the memory and CPU of each node that is abused by large computations? The way I read this (from the press release), this is a solution for number crunching, not for data warehousing, and the nodes boot off the network anyway.
    • Re:No SCSI (Score:3, Informative)

      by bm_luethke ( 253362 )
      well then use OSCAR from here [sourceforge.net]. It supports mandrake (and redhat) and scsi has the tools listed for CLIC, tested up to 128 nodes, doesn't require PXE card (though I would have to say a network switch is kinda needed for networking :) ). Has a fairly large base and good developer base.
    • The nodes should'nt be installed with SCSI because the duplication program (really fast 1Gb in less than 2min !) doesn't handle it but it seems to be fixed soon. But you can have scsi on the server of the cluster ! But if you wan you can make installation using the normal installation on all node but it will take more time :-)
  • Cluster? (Score:3, Funny)

    by sheepab ( 461960 ) on Wednesday October 30, 2002 @05:01PM (#4567844) Homepage
    Can you imagine a beowulf cluster of these things?
    OH WAIT! I CAN!
  • by screenbert ( 253482 ) <screenbert@[ ]mail.com ['hot' in gap]> on Wednesday October 30, 2002 @05:01PM (#4567848) Homepage Journal

    1. This one is nromal.
    - Phase 1: Develop and release a specialized Linux distribution that contains everything needed to quickly deploy a ready-to-run cluster.

    2. I think this means ?????
    - Phase 2 (early 2003): Release of specialized administration, control and monitoring tools for the clustering solution, plus added support for Itanium 2 architecture.

    3. And finally, PROFIT!!!!!
    - Phase 3 (end of 2003): Release of specialized tools and applications for development in parallel environment, final integration and tests, release of the final stable version.
  • Energy Usage? (Score:5, Insightful)

    by Steve Franklin ( 142698 ) on Wednesday October 30, 2002 @05:02PM (#4567851) Homepage Journal
    I always figured the problem with a home supercomputer would be the electric bill. Am I wrong about that?
    • Re:Energy Usage? (Score:1, Insightful)

      by Anonymous Coward
      All depends on if you are in winter and happen to heat with 'lectricity. Think of it from an entropy standpoint; electric coils turn highly ordered 50/60 cycle juice into a random sea of heat energy. Using a fleet of CPU's to do the conversion can at least allow you to have some fun in the process.
  • by GeckoFood ( 585211 ) <geckofood@nosPAM.gmail.com> on Wednesday October 30, 2002 @05:02PM (#4567864) Journal

    If it was financed by the French government, does this mean that Mandrake is going to become the Renault of the Linux world? Eww...

  • by itallushrt ( 148885 ) on Wednesday October 30, 2002 @05:03PM (#4567871) Homepage
    Great, now with my 15 386/486 systems sitting in the garage I can create something almost as powerful as this laptop I am currently using.
    • only at 34x the cost of running your laptop!!!
      only imagine the breathtaking view of your full 2d @640x480 at 53Hz refresh!!! almost ready to run clustered doom!!
    • NOPE! (Score:4, Informative)

      by bhsx ( 458600 ) on Wednesday October 30, 2002 @06:03PM (#4568560)
      Mandrake is compiled for i586, so you're gonna have to at least drop a pentium in to those old mobos...
      • Mandrake is compiled for i586, so you're gonna have to at least drop a pentium in to those old mobos...

        Ugh, you caused me to remember purchasing a Pentium Overdrive chip for a Packard Bell desktop many a year ago... of course the CPU was soldered onto the main board, and I had to return the Pentium upgrade...
  • Preemptive strike (Score:5, Interesting)

    by GreatDave ( 620927 ) on Wednesday October 30, 2002 @05:05PM (#4567900)
    Good. I hope this will convince CEOs in large companies that Linux clusters and scales better than Windows (well, that's obvious) or proprietary Unices. But the penguins can't be too confident, as Sun is up to no good here, [com.com] hawking distributed clustering as a brand new technology when it certainly isn't, and catering to Fortune 500s and other large companies.

    If we want to make our inroads we must do so now.
    • Excactly how is offering a solution to companies and helping them implement it "no good"?
      • > Excactly how is offering a solution to companies
        > and helping them implement it "no good"?

        Granted, if you are a Solaris fan, you'll be cheering for Sun (maybe...), but you'll still do a double-take at the news that Sun is out to try and "invent" distributed clustering. Read the CNet article carefully:

        Sun Microsystems has come up with a way to insulate computer networks from fires, floods and bomb attacks: Split up the machines and put them in different cities. The Santa Clara, Calif.-based server manufacturer on Tuesday will unveil its Enterprise Continuity program, a collection of services and technology designed to prevent network failure by physically separating computers that work together in a unified cluster.


        CNet may be totally off their rocker (which wouldn't surprise me), but I think it's more likely Sun provided this misinformation, verbatim, in a press kit to CNet. Sun's trying to be the first out the door when they're not; they're offering their service at a premium, saying 'this is the only way you can distribute your data centre'. That's what I call misleading advertisment.
  • by SexyKellyOsbourne ( 606860 ) on Wednesday October 30, 2002 @05:06PM (#4567910) Journal
    I recently acquired 30 PCs in the Pentium-II range from a local community college, for free, as they were actually headed for the landfill.

    Just as a weekend project, I was going to use the Beowulf software, but this CLIC software looks quite interesting, considering it's a total package and probably comes with the ease of use of Mandrake, so I'm going to give it a try.

    All I have to do is get the PCs out of the shed, make a lot of CAT5 cables, format quite a few hd's with CLIC, and build my own Beo^H^H^HCLIC cluster out of crappy Pentium II's and do something stupid like calculate PI to a googol digits, even though my power bill will probably be insane.

    Wish me luck, I'll keep you posted!
  • by Anonymous Coward
    Why, for the love of god, why must you idiots always mention "a beowulf cluster". There should be an automated system to mod any post to -1 that contains the word beowulf.

    "A BEOWULF CLUSTER!! (Score:-1, FUCKING RETARDED)"
    • Why are you so upset about BEOWULF CLUSTERS? Have you had some bad experience with BEOWULF CLUSTERS or do you not like BEOWULF CLUSTERS for some other reason? I personally have no problem with BEOWULF CLUSTERS, a lot of people here even like BEOWULF CLUSTERS. BEOWULF CLUSTERS have been known to be useful even.

      I conclusion, do you really think a good way to have BEOWULF CLUSTERS mentioned less frequently is to talk about BEOWULF CLUSTERS? I doubt it.

      Cheers,
      BEOWULF CLUSTERS

    • by Anonymous Coward
      could you imagine the automated modding system you just mentioned running on a beowulf cluster.

      Man that thing would SCREAM, baby!
  • home clusters (Score:2, Informative)

    by clymere ( 605769 )
    I'm helping to adminstrate a cluster here in Youngstown, Ohio thats partially funded by the Ohio Supercomputer Center. As a result of the few things I've picked up, I've gotten it into my head that I'm building a P1 cluster of my own, at home, over the summer. One of the OSC guys has a small(4-5 machines) P2 cluster in his office...its really not extremely difficult...except that I'm using P1's(cause i can get them for far less money). OSC has fairly well automated the install process for anything 686 and above. Using 586's means having to do most things manually. And before someone points it out, yes i'm aware i'm not gonna get a whole lotta power here...i'm doing it as a learning experience, right now noone is going to let me serup a cluster all on my own, so this is the best way for me to learn it i think. Incidentally, one of the things i want to try next is getting GUI desktop-type programs to distribute across my P1 cluster. The idea is to take all of these old systems, and get comparable functionality out of them as you get from your typical desktop computer. I really hare seeing how many systems end up tossed in the trash, i feel like something like this could prove helpful in eliminating some waste. Of course it could also just be a lot of hard work for nothing. Guess i'll find out this summer!
  • by evildan21 ( 548879 ) <evildan19@[ ]oo.com ['yah' in gap]> on Wednesday October 30, 2002 @05:08PM (#4567933)
    So I guess when you do something wrong it will tell you things like: "I fart in your general direction..." and most importantly "go away or I will taunt you a second time".
  • Alright! (Score:5, Funny)

    by papasui ( 567265 ) on Wednesday October 30, 2002 @05:08PM (#4567946) Homepage
    Now I can put those 8 486's to work and roughly have the power of a P2-300mhz and lower my heating bill all at once!
  • with 250+ 2.4 Ghz machines, all already configured to net boot, I think I can pull together some useable CPU power. Especially if they give Me keys to the master machine...

    I wonder if there's a multi-threaded Folding client......
  • by nocomment ( 239368 ) on Wednesday October 30, 2002 @05:09PM (#4567954) Homepage Journal
    Imagine a beowolf cluster of these!!! ;-)

    Seriously though. How cool is Mandrake? First to LSB, full GPL'd version of cd avail for download (damn you SuSE!!), lots of freebies, like the Single network firewall and the old Corporate server(i loved this back in the day), their prosuite stomps any other distro (and for $135 without doc's!), oh and mandrake runs on XBox!! :-)
    Now a cluster??

    My only complaint is they are an RPM-based distro and I like apt-get. Something like the freebsd ports collection would be nice too though, but i just hate those rpm's.

    Just so you don't think I'm a linux leech I have bought 3 version of linux inlcuding mandrakes prosuite, i can't donate code but i can donate ca$h.
    • mdk got the apt-get thing too.

      The tool is urpmi and the equivalent to debian unstable is cooker.

      Other interesting sources is contribs (unsuported contributions) and plf (stuff with legal issues to be official mdk packages).

      Dont compare apt-get to rpms. Compare debs to rpms.
      • Mod the parent up! urpmi is awesome, I guess i've been stuck in tgz hell for too long, i should have poked my head out to look at the new tools :-)

        Use urpmi, unless you're a debian geek.

        i sense a disturbance in the force, a new erra ...the apt-get vs urpmi debate begins........NOW!
    • Finally: a postive comment rather than the retarded dinosaur-era recycled jokes bashing the French and their government.
    • My only complaint is they are an RPM-based distro and I like apt-get.

      From the command line, urpmi will give you similar functionality (mainly deal with the dependencies for you). And even nicer, the GUI for Software Management in MandrakeControlCenter is just beautifull. Really painless software management, as long as you install software from the installation CDs or urpmi-aware repositories. Security updates are just a few clicks away and you get to see the advisories and decide what you want to install and what you don't. When installing from CD's you are prompted to insert the CD's in the order they are needed. You can search for packages (in names, files and descriptions). I NEVER had any problems, never had to manually solve dependencies (with ML 9.0). It works like a charm ...

    • In the longer term, set up all the desktops in your orkplace wth this Mandrake distro and use it for storage (Aka freenet) as well as processing, nothing unique to any particular desktop so you can add/delete from the LAN at will.

      But more immediately;

      So how about combininng all the no-fuss PVM setup stuff with the Gibraltar bootbale Debian CD?

      OK, maybe not as neat as PXE, but imagine being able to convert the office LAN into a Beowulf cluster every night when people go home.:o)

      Xix.
      P.S. Anyone notice how many more new ideas are popping up compared to "clone product X" ideas?
  • by tellezj ( 612044 ) on Wednesday October 30, 2002 @05:18PM (#4568061)
    One could argue that those businesses that really need a cluster will likely buy one preconfigured. Those that do the preconfiguration probably have an established technique (distro, tools, etc.) and won't likely go for something from Mandrake. The hobbiest who builds one to learn about clusters more than likely wants to do everything himself (like start with a strip down install of slackware and build it up from there). That really only leaves the hobbiest that wants to do parallel programming but doesn't want to build the cluster. Not a very large segment.
    • by bhsx ( 458600 )
      This is a freebie from the French governement really. They hired MandrakeSoft to create it for them, and MandrakeSoft turned around and said something along the lines of "You realize we're going to give this to everyone to play with and use don't you?" To which the French government replied "I don't care, I'm not gonna sell the thing, I just need it for virtual weapons testing, did i say weapons testing, I meant happy camps."

      So Mandrake didn't really do this to make more money off it, they're already getting paid for it, we just get a nice new toy to play with.
      • Congratulations, you just gave a three-lines summary of the general economic foundations of open-source.

        This is what should be replied to the usual "if we don't sell software, how do programmers get paid ? " whinning.
  • a) [Some French joke or another]

    b) [Some Beowulf joke or another]

  • This sounds like a very cool and nifty idea. The one thing that would hold me back is the fact that I would have to update from their network and would have a hard time installing packages of my choice.

    I may be ignorant but, whenever I tried to update anything or install anything, I was told that a dependency was not found. With a cursory exploration, I found that the dependency was there but low and behold everything has been renamed!

    MDK is a pain in the a$$!

    If they have stopped this practice of renaming files I am unaware as I have moved on to a more standard linux distro.
    • I am SO in agreement with you here. I wiped Gentoo1.3 last month to try Mandrake9 -- more or less just to see what it was like. I won't have a spare weekend to put Gentoo1.4 back on this box for two weeks so until then I just have to suffer.

      I can't wait to go back because each and every time I have tried to compile something nifty the process has failed because due to irresolvable dependency issues. This is what originally pushed me over to Gentoo from Mdk8.2, and once again I'm being pushed back.

      Mandrake is, like all the "latest and greatest" attempts at being a simple install (e.g., lindows, lycoris, redhat, xandros, etc.) FANTASTIC so long as you never plan on using anything not included in the distro, and never plan on upgrading any software package.

      That's fine for an office worker I suppose, but given the choice I'll take Debian or Gentoo any day. At least I know that when I get return to Gentoo when I type configure/make/make install I won't have to start crying!
  • until the the Xbox Linux Project [slashdot.org] ports this version to Xbox?

    It would be great to be able to have a bunch of friends bring their Xbox over and cluster them together.
  • When I think of clusters, I think of the active-passive Win2K database server we have at our co-location facility. It requires special cluster-aware hardware (e.g. the disk array) and cluster aware software (e.g. Win2K AS, SQL Server). I get the impression from people's comments that this is a different type of cluster. Rather than being about high availability, it is about massive parallel computing. Is this a correct assessment?
    • by Havokmon ( 89874 ) <rick@h[ ]kmon.com ['avo' in gap]> on Wednesday October 30, 2002 @05:29PM (#4568186) Homepage Journal
      When I think of clusters, I think of the active-passive Win2K database server we have at our co-location facility. It requires special cluster-aware hardware (e.g. the disk array) and cluster aware software (e.g. Win2K AS, SQL Server). I get the impression from people's comments that this is a different type of cluster. Rather than being about high availability, it is about massive parallel computing. Is this a correct assessment?

      Right. Netware 6 has kick-ass "clustering" that allows a Server to go down, and a 2nd server to beome your file server. You can stream a video (from FILE), down a server, and after a second, your stream will continue - from the 2nd server.

      Most of us call that failover, but Microsoft and Novell are calling it clustering.

      • So, what kind of HA (High-Availability) or Failover turnkey clusters are available for Linux? I'm curious, not trolling... I'd love to get one. I know Red Hat Advanced Server does it, and there are some projects out. (SuperMonkey? LVS?)
        • So, what kind of HA (High-Availability) or Failover turnkey clusters are available for Linux?

          To do what? File Serving? SQL Server?

          I believe NFS is 'HA', and I believe Postgres SQL would also be HA. There are also Linux-based load balancers available, so if you wanted to do something as simple as have 2 servers with static web pages, the load balancer would NAT those two systems, and direct requests to whatever one had the lowest load. That includes being completely down too :)

          One nifty HA application is Email. Take a look at Matt Simersons FreeBSD email toaster [simerson.net] (The Same apps are available on Linux) for a way to setup a HA Email cluster. It describes one system, but add NFS and MySQL replication, and you have HA.


          • the load balancer would NAT those two systems,


            Of course you would need at least two loadbalancers to. Available here.

            http://www.linuxvirtualserver.org/

            The stuff you can do with these kernel patches cost about $1000 for a hardware solution. And more if you want failover.

      • that allows a Server to go down

        hell, i'll allow a server to go down on me, he he he...
  • MPI/LAM (Score:3, Interesting)

    by Wolfier ( 94144 ) on Wednesday October 30, 2002 @05:24PM (#4568126)
    Does MPI/LAM run on it?
  • Urpmi parallel (Score:4, Informative)

    by imann ( 538879 ) on Wednesday October 30, 2002 @05:27PM (#4568171)
    I'm on the mailing list of this project and there was a parallel feature for urpmi that really rox !

    This tool allow people to deploy RPMS to a group of linux hosts using an intelligent parallel copy.

    How does it works:
    You create a group of hosts so the server can ask the nodes (using urpmi) to prepare for an update/install of packages.
    Each computer tell the server the packages it needs then the server copy in parallel (using ka-tools) the rpms on the nodes (that's very fast even for a hudge number of nodes).After that, nodes update their system using local rpms !
    This feature seems to be designed for clusters but should be used by admins !

    Another point of comparaison between urpmi & apt-get
  • by jukal ( 523582 ) on Wednesday October 30, 2002 @05:29PM (#4568182) Journal
    a project publicly funded by the French Agency for New Technologies (RNTL).

    I have lately started being more and more convinced that one of the key issues to success of open source - in a business sense, will be the fact that it is much easier for government organizations and other similar-type organizations to fund them without being guestioned and having put their moral in doubt. Why? Every single time that Microsoft, Adobe or some other closed source company is looking for government money, the politicians are facing a rather guestion: "In what light will this put us?"

    Now, it would be interesting to see some years further and see how this all changes. I am convinced that there will much more public discussion on the subject whether this is discrimative against the non-open source companies. In my opinion it basicly is not, because being open source, it benefits everyone and not just a single company - but still, there's still someone who benefits most.

  • Will I be able to get JLH in my computer just like they show here [mandrakesoft.com]?

    Or does that require setting up a cluster?
  • by BroadbandBradley ( 237267 ) on Wednesday October 30, 2002 @05:51PM (#4568403) Homepage
    Now they need to let Mandrake Club members pool their resources to create an internet wide cluster for all Mandrake users. Join the club, get to play on the club cluster.

    I don't know what the hell I would use it for / get out of it, but it would be a nice way to support your favorite distro by donating unused cpu time to MCC (Mandrake Club Cluster)

  • Now I can impress chicks... or not as the case may be....
  • by Anonymous Coward
    That's the problem with the economy right now... there are too many of these government financed open source projects. How can we expect the markets to operate properly, products to sell, and jobs to be created, when communists like the French keep putting out competing products like this? Or those communists at the NSA, working on "secure Linux" when Microsoft had a perfectly good C4 security clearance for Win NT 4?

    Fnord.
  • by artoo ( 11319 ) on Wednesday October 30, 2002 @06:01PM (#4568532)
    I can see it now. Microsoft continues to loose money due to the fact people are buying up XBoxes and modding them to run this.

    Bill: What's going on here. Sales of the XBox have risen 500%, but I'm making no money because no one's buying software!!

    MS Lackey: It's the French, sir.

    Bill: What?! How so?

    MS Lackey: Those Mandrake linux people released an ISO image of linux for the XBox, sir, and then they released clustering software. XBox supercomputing clusters are now springing up all over the place.

    Bill: Quickly, we must couteract this. Launch a shared source initiative, and I want the records showing everyone who has puchased an XBOX without buying any games for it!

    MS Lackey: Right away, sir.

    Bill: Oh, and while you're at it, buy France, jail the government, and burn down the Mandrake offices. That'll teach them.
  • I think we should ban marketing speak in the headlines. I get enough emails offering me "exclusive" PDF White Papers from companies talking about "synergy", "best-of-breed", "turn-key" solutions that will help my company "achive success in the competitive and challenging market of today".

    Bah! Such rubbish!

    Anyone agrees with me?

  • The real point (Score:2, Insightful)

    by compjma ( 591836 )
    I think you're all missing the most important question. When are we going to see video games for super clusters?
  • TerraSoft [yellowdoglinux.com] of Yellow Dog Linux fame has offered the same thing for the PPC arch for some time already. The product is called Black Lab Linux and Terrasoft has done quite a few demos with this powerful clustering software, videos and other promotional material are located on the website. The software was also recently tested with a cluster of Apple Xserves

    /me drools...

    And no I don't work in anyway for TerraSoft.

  • I've got machines all over the place, and have long fantasized about setting up a cluster for the fun of it. I've read a little about Beowulf and one or two other open source clustering technologies, but have never had the gumption to crack the books and set one up. Mostly, I think, because I'm not sure what I'd do with it when it was ready.

    Would anyone care to post a 2-3 paragraph summary of what Joe PC Hobbyist (e.g. someone like me, with many semi-obsolete PCs lying around the house) might be able to do with the CLIC software? Would it only be able to run applications written for CLIC (oh.), or would the cluster behave like one giant, amalgamated Mandrake box (cool!) ?

    • Well, I don't know how to use CLIC, but I set up a cluster using some surplus P2s, RH 7.3, and openMosix [openmosix.org].

      For kernel 2.4.18, openMosix consists of two kernel patches - a kernel patch to actually do the clustering, and a kernel patch for administration. You patch your kernel, change a few settings (all documented on the site), reboot (gasp!), and you're done.

      You can get more fancy with net installs, but you don't really need to for a small number of nodes. Anyway, I just used a Fast Ethernet switch - the boxes all came with Ethernet cards, luckily for me.

      Mine is a pretty craptacular cluster, but hey, it's pretty cool to have one. The cluster behaves like one powerful computer in the sense that processes on one machine will migrate to another. However, in order to achieve any noticeable affect, the program has to take advantage of multiple CPUs. If you're going to program your own, look into forking, message-passing interface (MPI), and parallel virtual machine (PVM). Sorry, I don't have the addresses for them on hand, I think www.beowulf.org [beowulf.org] links to them.
  • I've seen a lot of people comment on how much energy this would use, but no real numbers. I know it's subjective as to where you live, but generally, how much energy does a computer use? How much would this use?
  • Now all of those crap boxes that "walked home with me" from last tech job will be put to use. This is remarkable to me, I had no idea that such a project was in the works. A question though, do you think that this little freebie from Mandrake and the French Government will affect the profits of a company like Linux NetworX or the use of Beowulf?
  • by teaserX ( 252970 ) on Wednesday October 30, 2002 @07:05PM (#4569112) Homepage Journal
    This seems pretty easy to me. Add the follwing to your RedHat distro, a little scripting, a little autorun info and you got it (for your hardware at least).
    You will need:
    Then:
    1. Install RedHat
    2. Install openMosix rpms
    3. Install kernel source to /usr/src/linux and patch with openMosix-2.4.18-4.gz
    4. Compile and install openMosixUserland-0.2.4
    5. Configure /etc/openmosix.map
    6. Reboot new openMosix kernel
    7. setpe -W -f /etc/openmosix.map
      on each node
    8. ...and you're clusterin'
  • Maybe I'm missing something, but how is this different from MOSIX [mosix.org]?

    I have a small 6-node cluster of P2s running openMosix [openmosix.org]...
  • It's great because I've always dreamed of having my own supercomputer at home."

    So uh, get a playstation 2. Or a G4. Ha ha.

    You know clustering has been around for a god damned long time, and (low end) PCs have been cheap for a less long but still considerable amount of time. You could have built yourself something decent a couple years ago.

    If this is the answer you've been waiting for all this time, I'm not sure I want to know what the question was.


  • Awesome! Just when I was about to declare Red Hat the winner, Mandrake comes up with a "cluster in a can". Great news.

Get hold of portable property. -- Charles Dickens, "Great Expectations"

Working...