Slashdot Mirror

← Back to Stories (view on slashdot.org)

Online Game Cluster

Posted by timothy on from the drool-pools dept.

axehind writes "Carlo Daffara posted to the openMosix mailing list about his game cluster. It's a 6 node cluster using Athlon XP cpus and running linux & openmosix (with some qdisc trickery) for the OS. It is used to host approximately 1000 users playing online games, like Jedi Knight and Quake III. Here's a link (italian) to the pictures."

11 of 107 comments (clear)

  1. Imagine... by SlimFastForYou · · Score: 5, Interesting

    That same cluster spec running 1 instance of a game for someone to play on :). Someday I hope for clustering software to be to the point where when someone at a LAN party goes to sleep, their processing power can be used to help other machines.

    1. Re:Imagine... by neo8750 · · Score: 2, Interesting

      who goes to sleep durning a lan party?

    2. Re:Imagine... by SlimFastForYou · · Score: 2, Interesting

      At two day LAN's?:

      The weak rich bastards with the P4 2.8 GHZ and 1 Gig of RAM :P.

  2. Why use Mosix ? by boaworm · · Score: 4, Interesting

    OpenMosix is used for load balancing, allowing processess to migrate during runtime.. I wonder what happens if you are trying to "gib/frag" someone and the server process migrates to another node .. ;-)

    It seems to me it would make more sense to use only the mentioned queue-system to position the games evenly.

    --
    Probable impossibilities are to be preferred to improbable possibilities.
    Aristotele
  3. Why not make ONE game... by SexyKellyOsbourne · · Score: 4, Interesting

    That uses all CPU resources of a cluster, and leap 10 years into the future?

    The AI would use genetic algorithms, the sound would have every echo and diffusion effect possible, the graphics would use real-time raytracing, and the level count would be as extreme poly as possible. Simply spare nothing when it comes to CPU power, and just let it fly.

    Just use nothing but outright raw CPU power to render the whole thing.

    1. Re:Why not make ONE game... by Screaming+Lunatic · · Score: 4, Interesting
      That uses all CPU resources of a cluster, and leap 10 years into the future?

      The same reason that game companies don't make games for machines with 8 CPUs or 4 CPUs, or even 2 CPUs. People just don't have machines out there to play the game.

      You can do it server-side, because some geek just has to put together a bunch of computers. On the client-side, can you imagine that 12 year old that keeps fraggin you when you're playing RtCW online putting together a cluster of computers? (That was rhetorical)

      The AI would use genetic algorithms

      "True to life" AI, does not imply good "Game" AI.

      the sound would have every echo and diffusion effect possible,the graphics would use real-time raytracing, and the level count would be as extreme poly as possible

      The network latency would kill you. Motherboards are moving to AGP 8x so that they can get more bandwidth to the graphics card.

    2. Re:Why not make ONE game... by FTL · · Score: 3, Interesting
      >That uses all CPU resources of a cluster, and leap 10 years into the future?

      Hmm, that is a good idea. According to my back-of-the-envelope math, 10 years would be seven cycles of Moore's law. Which means we are roughly dealing with a 128 times increase in CPU power. A cluster computer will waste much of its CPU time, so let's say we'd need 256 modern computers to do what you want.

      Not something the average person could do, it is the sort of thing that a bored university student with access to a lab of computers could do...

      --
      Slashdot monitor for your Mozilla sidebar or Active Desktop.
  4. Migrating processes by tribes · · Score: 2, Interesting

    We toyed around with openmosix, "borrowing" some hardware at work. It was interesting to see what types of processes would be migrated to other cpus and which ones were simply not movable. We ended up running distributed.net threads until our hardware hijacking was discovered (hard to miss ~30 2RU servers). I would guess their tweaks facilitate easier process migration?

    1. Re:Migrating processes by realnowhereman · · Score: 2, Interesting

      Clustering is not necessarily used to get performance increases. We're looking at using openMosix to make better use of a room full of lab computers. Certainly during the night their CPU's are completely unused. Getting processes to migrate to them and make use of the CPU is the goal NOT to gain some huge parallel computing benefit. The original poster might not want to get higher performance but rather to "steal" cycles from idle desktops, that otherwise would go to waste.

      --
      Carpe Daemon
  5. Local Cacheing by Catskul · · Score: 3, Interesting


    How about: before posting we see if we can get permission to cache locally (on slashdot). Obviously this isnt always possible, but in the case of small websites it might be practical, and even polite to do.

    --

    Im not here now... Im out KILLING pepperoni
  6. are there tuturials? by justindussonet.net · · Score: 2, Interesting

    are there tutorials out there on this kind of stuff? this seems very interesting to me and I would like to play around at home with a few machines I have.