This looks interesting... let's try it!

Geeky1Geeky1 University of the Pacific (Stockton, CA, USA)
edited September 2003 in Folding@Home
It's similar in concept to stuff that's been done before, but the implementation is very different:

http://www.icronticforums.com/showthread.php?s=&threadid=2323

Comments

  • TemplarTemplar You first.
    edited September 2003
    Who's paying the power bill? :)
  • Geeky1Geeky1 University of the Pacific (Stockton, CA, USA)
    edited September 2003
    Don't look at me. I'm not allowed to have a rack in my room.
  • LincLinc Owner Detroit Icrontian
    edited September 2003
    I'd just as soon stick with the SMx project. It distributes the responsibility and recognition to more team members, and doesn't require as many resources from one person. Besides, the pieces we get slowed down on are the core pieces we'd need for nodes anyway. We have enough HDs and NICs to last a loooong time :)
  • Geeky1Geeky1 University of the Pacific (Stockton, CA, USA)
    edited September 2003
    GK, you forgot about one thing tho:

    Being able to tell people you have a, say 20GHz computer (computer/beowulf cluster... same difference) is fun :D
  • Straight_ManStraight_Man Geeky, in my own way Naples, FL Icrontian
    edited September 2003
    Yeah, I got into that thread, thanks for pointing it out Geeky. Not cheap unless you have someone like Tex with many extra parts, though. THAT is part of Icrontic's leverage here. Problem is , what you get is basicly a LAN with a gateway with Folding, it does not have a clustrerable client to allow one box to split work units which is what a cluster is really optimized for (controlled parallel versus distributed processing). What folding has for a core is basicly an interlocked (semi gridded)LAN, which is much cheaper than writing your own client to really parallel compute.

    To illustrate quickly-- take a 30 blade cluster, assign one percent of the first thirty to each blade, all work at once, then next 30 percent is done. This means you need in fact a 33 blade cluster to do this with one controller that also does the last percent of a work unit, or you need work units broken into the number of parts that your blades can process in one pass with each taking an equal part of the load.

    I would instead LAN this thing with one internet gateway box, advantage would be one place to maintain all thirty in actuality-- analogous to puttign all the SMx's in one hosting faciltiy (co-location) and having one or two folks responsible for checking on them-- could be remote as our admins do here except for emergencies and major Database burps.

    Other reason for a LAN or grid is each box can be unequal in power to any peer. True clusters are matched capacity of processing boxes, or are broken into subclusters of like boxes. This unequalness handling makes a spare parts build-up easier to make happen, as finding 30 totally alike sets of parts on ebay is unlikely. EXCELLENT in theory, Folding is not tuned for the cluster Idea.

    CERN in Switzerland IS suited to a cluster, the data flow from a mass particle accellorator needs true parallel computing to control quickly and get results in reasonable times. so are Los Alamos clusters, they calculate each parts of what an atomic reaction will do under certain very complex scenarios.

    John.
  • Geeky1Geeky1 University of the Pacific (Stockton, CA, USA)
    edited September 2003
    Here's an idea. Primesuspect has a web hosting company thingy, right? Let's make him install F@H on the servers :D:nudge:
  • mmonninmmonnin Centreville, VA
    edited September 2003
    Hehe no I wouldnt do that with a server.

    Its not a reasonable thing to make. It would cost way too much cause we would have to have all the same parts which we would have to buy most of them. Then putting the burdon on one person to host that is too much.
Sign In or Register to comment.