Project Diablos

citrixmetacitrixmeta Montreal, Quebec Icrontian
edited October 2003 in Folding@Home
Citrixmeta, of Team Icrontic, has a crazy idea to bring in an enormous amount of points for the team…

Introducing project "Diablos"

A giant cluster of computers folding for Team 33693! Essentially, all that is needed for each node of this folding monster is a CPU (1GHz or more), CPU Heatsinks, Motherboard, RAM, Power Supply (250W or more), HDDs, NICs, 52u rack with, trays ,hubs/switches, Ethernet cables, Power cables, etc.

Fortunately, almost all of the above is already covered.

All we are missing include:

- CPUs (and heatsinks)
- Motherboards
- Power Supplies

Over time, we will be accumulating parts, probably off eBay, to build it. Also, it would be great to somehow get the whole cluster running as efficiently as possible, so the choice of OS or method of linking these machines together is important.

Do any among the team have spare parts or experience with this sort of undertaking? Imagine the possibilities for the F@H project and the team!


post your comments please.



Join Project diablos Signature.

<img src="http://www.teamicrontic.com/diablosjoin.png&quot; border="0">

Comments

  • trippintrippin Chatt, TN
    edited September 2003
    This sounds like an awesome idea. I think i can help :D. Ive got an extra 1700+ and an extra powersupply for the taking! I also carry with me a small amount of knowedge about FreeBSD :D. I think we should run BUNCH of diskless nodes running a custom trimmed down version of FreeBSD 4.x stable thats say around 30megs or less. I have done this before and will do it again. The server can keep statistics and monitor all the clients to make sure they stay in line. It could ping every machine every 5 mins and if one goes down send out an email. It could generate up to date reports on the progress of each machine.

    This sounds like a great project, and its in the name of folding so it also has a great cause! I'm in for what ever i am worth.
  • karatekidkaratekid Ogdensburg, NY
    edited September 2003
    Oh man, this just sounds awsome. I can see it now, 'Icrontic's Super Mega Awsome Ass-Whopping Folding Machine.' This, seriously, is one of the coolest ideas I have seen in a long time. And of course, this monster will be pumping out protien after protien for a very wearthy cause, which of course is what really matters.

    I don't have any spear parts to donate, but I can send yea some cash. I can tell you this one thing, seems that the Linux folding client is slow and unstable. People are find that running the windows client under Wine gives them much better performance and stability, and that the Windows client even runs faster under Wine then under Windows.
  • citrixmetacitrixmeta Montreal, Quebec Icrontian
    edited September 2003
    oh wow replies already!!

    thanks for the kind words guys.


    well, we will discuss all the technical stuff once a few parts start coming in.

    we are going to run a few tests and decide on the setup.

    thanks guys!
  • MrBillMrBill Missouri Member
    edited September 2003
    I think it's interesting that the team number is 33693. :D
  • edited September 2003
    ...how does this differ from ICX other then it's just a cluster instead of machines spread out at people's houses/work place/etc?

    I can't wait to make my own site and hopefully Tex will be back in charge so he can give me the new team!
  • citrixmetacitrixmeta Montreal, Quebec Icrontian
    edited September 2003
    no m33ps, theres actualy more involved, we have a game plan that we will release later on.
  • trippintrippin Chatt, TN
    edited September 2003
    Shipped tomarrow to canadian Citrixmeta:
    Amd Athlon Tbred 1700+
    Amd Athlon Tbird 1.0ghz
    300 Watt Powersupply

    This should help get us off to a good start.
  • citrixmetacitrixmeta Montreal, Quebec Icrontian
    edited September 2003
    thats incredible!!

    im getting a mobo 2mmorow, we need 1 psu people!! and we got 2 nodes already.

    thanks trippin
  • edited September 2003
    If you want to see what another team has done with folding and clusters, go to overclockers.com and run a search on "Yattamonster" and you should be able to turn up some good info on some clustered machines running Linux. They have a lot of good ideas for this kind of thing.:)

    Good luck with the Diablos, you guys.
  • citrixmetacitrixmeta Montreal, Quebec Icrontian
    edited September 2003
    Yes!! thanks muddocktor!!

    i actualy found a few documents, im working on a diskless boot to win98 (rumor has it *nix folding is slow and unstable)

    thanks alot for the info!
  • edited September 2003
    AKAIK, *nix is actually a little faster when folding Gromacs WU's and stable; it's with Tinker WU's that *nix is slower than Windows based OS's.

    o/c has 1 member that built a 6 node, Yattamonster that only uses 2 hard drives, with the other 2 nodes per layer using a network boot off the master of that layer.

    Thiis is a link to stymee's thread there about his Yattamonster Jr. and he has some nice pics and some links too. Even if you don't want to *nix it, it is still a nice thread to check out and see the workmanship that he put into his cluster.
  • citrixmetacitrixmeta Montreal, Quebec Icrontian
    edited September 2003
    awesome! ok then, we got some work to do now,


    OMG !! what a great HOWTO!!

    im definitly going to try this tonight!

    thanks again muddocktor for the support!




    /me waves at trippin!
  • trippintrippin Chatt, TN
    edited September 2003
    Check out the Project Diablos' new site.

    http://diablos.icrontic.com

    This site will be updated regularly with information from this thread and the donation thread.
  • edited September 2003
    I can feel the breeze from the electric meter already... I guess the upside is that wherever this folding machine is going won't have a heating bill all winter... :D
  • citrixmetacitrixmeta Montreal, Quebec Icrontian
    edited September 2003
    :D;):D
  • trippintrippin Chatt, TN
    edited September 2003
    Originally posted by godzilla525
    I can feel the breeze from the electric meter already... I guess the upside is that wherever this folding machine is going won't have a heating bill all winter... :D

    That accually is a serious issue, but luckily it seems that citrixmeta's server room is air condioned, well wired, and has a strong circuit breaker. If it wasnt we would have some problems, because this thing is gonna draw some amperage. 10 nodes will draw probly around 1500 watts or more which is substanial!
  • karatekidkaratekid Ogdensburg, NY
    edited September 2003
    Originally posted by trippin
    10 nodes will draw probly around 1500 watts or more which is substanial!
    In the words of the great Kang "Holy Flurking Shnit"
  • CammanCamman NEW! England Icrontian
    edited September 2003
    saw a link to this somewhere, I dont remember where, probably at S-M, but anyway, its great to see some innovation and exciting projects in F@H, good luck with your Diablos cluster, I cant wait to see it coming together!
  • a2jfreaka2jfreak Houston, TX Member
    edited September 2003
    If anyone has any evidence about which OS is faster for which type of WU, could you please post it here?

    Also, Citrix, if you can get W98 to work diskless, can you post a link to a how-to here?
  • citrixmetacitrixmeta Montreal, Quebec Icrontian
    edited September 2003
    you could say we are still in testing Phase.

    this week we should have the 1st node (bsd) running and booting off a BSD box.

    then we can bench/test.


    trippin isnt following any documents, he is writing one :D

    as for the win98 diskless, here it is



    http://www.nelliott.demon.co.uk/distributed/windows_diskless/how_to_build.html
  • edited September 2003
    Anyone with Linux experience might want to look into the CLIC project that Mandrake has going. One of the first subdistros they did that was not for PCs or Single servers was a Linux Cluster implementation. This uses one controller blade and the rest as peer nodes. Controller in essence delegates work. BUT, I do not know how you would get work from folding fully automaticly for 15-30 blades from one ID unless the controller was managing as many users as peer blades.

    Easier, would be to make the manager box a server, and delegate to clients. If you want pure peer to peer and have a router, it could mitigate network connects or you could build a Linux router to interface to web, then each blade could get its OWN by connecting via LAN and using router as gateway. In this last case, if you used dialup to send and receive data, you might be able to have the router box only do routing, hook modem to router box whihc could be BSD or Linux, and then let each machine vie for a connect since the clients will resend in Linux if modem is down or connection to internet fails. to the blades under the router box's umbrella, you could set them to use a gateway to internet with the router bnox as a gateway for them and the dialout the router box's gateway.

    That is basicly how we got Windows for Workgruops to dial out way back when and is fairly simple to do if someone knows PPP (AKA PTPP) and how to set up a single subnet lan.

    IF you want a wild idea, my Linux box now thinks it is a peer surfing on a worldwide LAN\WAN with End Point Security(it guards itself), no cable modem driver it knows of-- Zoom 5041 will run Linux transparently on most Cable companies in US. THAT part alone I could help with well, as well as defining a subnode, but I do not use Dialup on Linux as that is more complex due to the variants of ways to dialup used by different ISPs.

    with multiple boxes, easiest way is a router box ot outside world, which does NOTHING but route and can be slower than any of the blades it is routing for, with a cluster semi-protected by the router'ssecurity with each balde doing nothing but fold.

    Depending on how big the blades are, most of the cluster could also be KVM'd with each blade run headless as far as monitor (technically, you COULD skip a video card per peer\client blade with a good blade cluster and SSH from one box with X on it(or not, if there is aUnix admin or BSD admin or Linux admin in the house) and have the blades each talk to the controller blade as to data-- inverse of a Server to dumb terminal star lan structure as far as being headless, but that is one way to do that part-- software called vnc is used there with ssh logins to keep secure). COULD even be set up for remote, as Linux does have mail transfer agents included and email notifies could be used once things were stable with a remote connect-in to fix problems.

    John.
  • citrixmetacitrixmeta Montreal, Quebec Icrontian
    edited September 2003
    - update



    mobo and psu have arrived, we are waiting for the CPUs that trippin sent and were ready for Node1

    :D
  • citrixmetacitrixmeta Montreal, Quebec Icrontian
    edited September 2003
    -Update!

    CPU's are in!!!


    1700+ and a duron!! node 1 should be up tonight/tommorow

    :D
  • edited October 2003
    you could save power by using headless and diskless nodes. LTSP seems to work well for folding. Once you confirm that folding starts once the node boots you could remove the video card
  • trippintrippin Chatt, TN
    edited October 2003
    Originally posted by Bionic_Redneck
    you could save power by using headless and diskless nodes. LTSP seems to work well for folding. Once you confirm that folding starts once the node boots you could remove the video card

    We were planning on running a diskless headless nodes. No need for a video card at all. If I need to make sure folding is running I will just ssh into the box.
  • mmonninmmonnin Centreville, VA
    edited October 2003
    So what the names of these nodes going to be? I dont recognize one on stanford yet.
Sign In or Register to comment.