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

Pointers for use on a small cluster.

Expand Messages
  • flimped
    I m doing some factoring stuff at my college and I have a small cluster available. I ve pieced together some information online, but if anyone can offer some
    Message 1 of 4 , May 2, 2009
      I'm doing some factoring stuff at my college and I have a small cluster available. I've pieced together some information online, but if anyone can offer some pointers or a document about distributed/parallel factoring using ggnfs that would be great. The numbers I'd like to factor are only 120 digits and it sounds like that should be fairly easy for ggnfs.

      Thanks!
    • Jeff Gilchrist
      ... If the numbers are 120 digits they can be processed fairly fast. For GNFS numbers it would take about 4 days on 1 processor of a Pentium4 3 GHz machine.
      Message 2 of 4 , May 3, 2009
        On Sat, May 2, 2009 at 10:36 PM, flimped <flimped@...> wrote:

        > I'm doing some factoring stuff at my college and I have a small cluster
        > available. I've pieced together some information online, but if anyone can
        > offer some pointers or a document about distributed/parallel factoring using
        > ggnfs that would be great. The numbers I'd like to factor are only 120
        > digits and it sounds like that should be fairly easy for ggnfs.

        If the numbers are 120 digits they can be processed fairly fast. For
        GNFS numbers it would take about 4 days on 1 processor of a Pentium4 3
        GHz machine. So if your cluster machines were dual-core or quad-core
        Core2/Opterons you can process a number in 1 or 2 days. This guide
        explains how to set everything up in a parallel way:
        http://gilchrist.ca/jeff/factoring/nfs_beginners_guide.html

        You can also spread some of the load over multiple machines but you
        would have to write your own scripts to handle distributing the work
        between systems or someone else here might be willing to share their
        scripts if they know what kind of OS/systems your clusters is made up
        of. But with such small jobs it might be just easier to do multiple
        parallel jobs on your cluster.

        Jeff.
      • Luigi Morelli
        ... We at ClusterArch are working on a 16 multicore workstations small cluster, based on a rewritten Beowulf kernel and MPI. Are there open source factoring or
        Message 3 of 4 , May 3, 2009
          > If the numbers are 120 digits they can be processed fairly fast.
          > For
          > GNFS numbers it would take about 4 days on 1 processor of a Pentium4
          > 3
          > GHz machine. So if your cluster machines were dual-core or quad-core
          > Core2/Opterons you can process a number in 1 or 2 days. This
          > guide
          > explains how to set everything up in a parallel way:
          > http://gilchrist.ca/jeff/factoring/nfs_beginners_guide.html
          >
          > You can also spread some of the load over multiple machines but
          > you
          > would have to write your own scripts to handle distributing the
          > work
          > between systems or someone else here might be willing to share
          > their
          > scripts if they know what kind of OS/systems your clusters is
          > made up
          > of. But with such small jobs it might be just easier to do multiple
          > parallel jobs on your cluster.
          >
          > Jeff.
          >

          We at ClusterArch are working on a 16 multicore workstations small cluster, based on a rewritten Beowulf kernel and MPI.
          Are there open source factoring or primality checking applications that already run on MPI framework we can test on it?

          Thanks :-)

          Luigi
        • flimped
          ... Jeff- I ve been using your guide as a starting point and it has been a great help. I noticed that factMsieve.pl accepts some arguments called client_id and
          Message 4 of 4 , May 3, 2009
            --- In ggnfs@yahoogroups.com, Jeff Gilchrist <jeff.gilchrist@...> wrote:
            >
            > On Sat, May 2, 2009 at 10:36 PM, flimped <flimped@...> wrote:
            >
            > > I'm doing some factoring stuff at my college and I have a small cluster
            > > available. I've pieced together some information online, but if anyone can
            > > offer some pointers or a document about distributed/parallel factoring using
            > > ggnfs that would be great. The numbers I'd like to factor are only 120
            > > digits and it sounds like that should be fairly easy for ggnfs.
            >
            > If the numbers are 120 digits they can be processed fairly fast. For
            > GNFS numbers it would take about 4 days on 1 processor of a Pentium4 3
            > GHz machine. So if your cluster machines were dual-core or quad-core
            > Core2/Opterons you can process a number in 1 or 2 days. This guide
            > explains how to set everything up in a parallel way:
            > http://gilchrist.ca/jeff/factoring/nfs_beginners_guide.html
            >
            > You can also spread some of the load over multiple machines but you
            > would have to write your own scripts to handle distributing the work
            > between systems or someone else here might be willing to share their
            > scripts if they know what kind of OS/systems your clusters is made up
            > of. But with such small jobs it might be just easier to do multiple
            > parallel jobs on your cluster.
            >
            > Jeff.
            >

            Jeff-

            I've been using your guide as a starting point and it has been a great help. I noticed that factMsieve.pl accepts some arguments called client_id and num_clients... I'm currently running a factor job on a few machines with the directory shared via NFS. Ive then run factorMsieve.pl like: ./factorMsieve.pl test.poly 1 3 (machine 1) ./factorMsieve.pl test.poly 2 3 (machine 2) etc. Does this handle everything for the clustering for me or do I need to combine the sieving results manually?


            Thanks again for all your help!
            -Ryan
          Your message has been successfully submitted and would be delivered to recipients shortly.