Loading ...
Sorry, an error occurred while loading the content.

49Re: [genetic_programming] Beowulf Clusters

Expand Messages
  • Sean Luke
    Oct 2, 2001
    • 0 Attachment
      On Tuesday, October 2, 2001, at 12:53 PM, Martin C. Martin wrote:

      > If you use 1 byte per node, then your individuals are only a few k each
      > at most. Depending on how long an eval takes, I think communication
      > won't be the bottleneck.

      I think the trend in EC is towards rather long evaluation times. If
      evaluation time is the dominant factor, then heck, even on systems which
      are piggish in transfer size, communication lag *still* won't be the
      bottleneck. If memory isn't a problem, I would optimize for an internal
      representation which evaluates rapidly and can be easily modified rather
      than one which can be stored and transferred efficiently.

    • Show all 19 messages in this topic