initial replication 4


Advanced search

Message boards : SZTAKI Desktop Grid : initial replication 4

AuthorMessage
Sadir
Send message
Joined: Oct 19 05
Posts: 4
Credit: 1,668
RAC: 0
Message 2320 - Posted 25 Mar 2006 16:35:06 UTC

    Why does SZTAKI have initial replication 4 and minimum quorum 3? I think that much better is initial replication 3 and minimum quorum 3 even if there may be a little delay with delivery of result but no wasting of computing time.
    ____________

    Profile Daxl
    Send message
    Joined: Jul 14 05
    Posts: 8
    Credit: 65,544
    RAC: 0
    Message 2322 - Posted 25 Mar 2006 17:20:03 UTC

      Hi,

      look at SETI@Home:
      minimum quorum 3
      initial replication 4
      max # of error/total/success results 5, 10, 5

      look at SZTAKI:
      minimum quorum 3
      initial replication 4
      max # of error/total/success results 3, 6, 6

      I think that the settings of max#errors and max#total
      are the Problem for the errors:
      \"Too many total results\" and \"0 granted Credit\"
      If three user cancel there WU\'s the Number of error=3 and all other get no credit.

      @Sadir be patient and let ADAM work on this problem.
      (Some Credits are lost in the big Black Hole of BOINC) *ggg*

      DAXL


      ____________

      Profile kadam
      Project administrator
      Avatar
      Send message
      Joined: May 25 05
      Posts: 589
      Credit: 38,614
      RAC: 0
      Message 2352 - Posted 27 Mar 2006 10:24:32 UTC - in response to Message 2322.


        (Some Credits are lost in the big Black Hole of BOINC)


        If you claim credit for an invalid work than I\'ll grant the credits for it, so those are not lost...
        ____________
        If you like BOINC, you may also find CaretCursor to be appealing.

        Sadir
        Send message
        Joined: Oct 19 05
        Posts: 4
        Credit: 1,668
        RAC: 0
        Message 2367 - Posted 29 Mar 2006 22:20:48 UTC

          I just wanted to say if there would be minimum quorum of 3 and initital replication 3, there would be increase of efficiency of about 25%, won\'t be?
          ____________

          Profile Bob Carlton
          Avatar
          Send message
          Joined: Jan 13 06
          Posts: 18
          Credit: 9,224
          RAC: 0
          Message 2369 - Posted 30 Mar 2006 1:56:02 UTC - in response to Message 2367.

            I just wanted to say if there would be minimum quorum of 3 and initital replication 3, there would be increase of efficiency of about 25%, won\'t be?


            No, Sadir, the efficiency would decrease by 33%. Instead of needing 50% of 4 wu\'s to agree, you would need 67% of 3 wu\'s to agree, the chances of which, with the checkpointing problem, are pretty slim in a lot of cases. And what if you send out an initial replication of 3 and one does not return by the deadline, or one is a ghost wu, or a lot of other possibilities.

            It is always better to send more than you need than not enough or just how many you need. There are a lot of things that can go wrong.
            ____________
            member of

            Profile Keck_Komputers
            Avatar
            Send message
            Joined: Sep 18 05
            Posts: 16
            Credit: 7,898
            RAC: 0
            Message 2371 - Posted 30 Mar 2006 22:21:21 UTC - in response to Message 2369.

              I just wanted to say if there would be minimum quorum of 3 and initital replication 3, there would be increase of efficiency of about 25%, won\'t be?


              No, Sadir, the efficiency would decrease by 33%. Instead of needing 50% of 4 wu\'s to agree, you would need 67% of 3 wu\'s to agree, the chances of which, with the checkpointing problem, are pretty slim in a lot of cases. And what if you send out an initial replication of 3 and one does not return by the deadline, or one is a ghost wu, or a lot of other possibilities.

              It is always better to send more than you need than not enough or just how many you need. There are a lot of things that can go wrong.

              Both of these answers are correct, it depends on how you define efficiency. The total CPU time used would decrease by 20% to 25% but the wall time for an average completed workunit would increase by about 30%. A project that has limited supply of CPU time would want to decrease the initial replication to match the qurom. A project that needs quick turnaround may increase the initial replication, for example LHC sends 5 tasks with a quorum of 3.

              Studies have shown that generally the best compromise between CPU time and wall time is to have the initial replication 1 more than the quorum. However then needs of the project are the determining factor.
              ____________
              BOINC WIKI

              BOINCing since 2002/12/8

              Post to thread

              Message boards : SZTAKI Desktop Grid : initial replication 4


              Home | My Account | Message Boards


              Copyright © 2017 SZTAKI Desktop Grid