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

38Re: [genetic_programming] Beowulf Clusters

Expand Messages
  • Douglas Kell
    Oct 1, 2001
    • 0 Attachment
      Yes, the most straightforward solution, which seems to me rather
      convenient, is simply to scale the population size that is dealt with
      by a given machine (whether explicitly in demes or effectively as a
      single metapopulation) according to the machine's effective speed.
      This way each 'generation' takes the same time on machines of
      very different spec. Might have some interesting properties too, and
      there are some especially interesting and (so far as I know in
      parallel machines) relatively uncharted ideas in which you might
      also make the mutation/recomination rate on each machine a
      function of the size of each population, as a way of tipping the
      exploration/exploitation ratio in your favour.

      Interesting thread.
      Douglas.

      On 1 Oct 2001, at 14:56, Lee Spector wrote:

      >
      > Mattias,
      >
      > All of the CPUs in our cluster are the same speed so I haven't had to deal
      > with this, but your idea of scaling population sizes with clock speed seems
      > like a good approach to me. Restarting a node and then importing
      > individuals from late in other runs seems like a bad idea (as you seem to
      > think as well).
      >
      > -Lee
      >
      >
      > >Hi!
      > >
      > >I've been having a bit of a headache with my beowulf cluster code - we
      > >currently have several machines of different speeds, so while one machine
      > >might perform 5 generations, another might only perform 3 generations.
      > >
      > >Would you generally let the fast machine and the slower machine exchange
      > >genetic material? And if one machine restarted it's evolution, started
      > >from zero, if it still allowed others to supply genetic material, it'd be
      > >like it never restarted within only a handful of generations.
      > >
      > >I'm thinking, if they're within 10 generations of each other, I'll allow
      > >the exchange, but that's such a neat solution. Another solution might be
      > >to make the population on the faster machine larger, and on the slower
      > >machine make it smaller, so that they keep about the same pace...
      > >
      > >Any ideas?
      > >
      > >thanks,
      > >mattias fagerlund
      > >
      > >
      > >
      > >To unsubscribe from this group, send an email to:
      > >genetic_programming-unsubscribe@yahoogroups.com
      > >
      > >
      > >
      > >Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
      >
      > --
      > Lee Spector
      > Associate Professor of Computer Science lspector@...
      > Cognitive Science, Hampshire College http://hampshire.edu/lspector/
      > Amherst, MA 01002 413-559-5352, Fax: 413-559-5438
      >
      >
      > To unsubscribe from this group, send an email to:
      > genetic_programming-unsubscribe@yahoogroups.com
      >
      >
      >
      > Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
      >
      >
      >


      ********************************************************************
      *(Prof.) Douglas B. Kell Phone: +44 1970 622334 *
      *Director of Research *
      *Institute of Biological Sciences Fax: +44 1970 622354 *
      *Cledwyn Building *
      *University of Wales, Email: dbk@... *
      *Aberystwyth SY23 3DD, U.K. http://qbab.aber.ac.uk/ *
      * http://www.abergc.com*
      ********************************************************************
      ********************************************************************
    • Show all 19 messages in this topic