What is Avg Work Done (Credits)?


Advanced search

Message boards : Getting Started : What is Avg Work Done (Credits)?

AuthorMessage
Profile Brett Kuntz
Send message
Joined: Aug 13 10
Posts: 4
Credit: 1,778,319
RAC: 0
Message 105599 - Posted 17 Aug 2010 22:07:01 UTC

    What is the Avg. Work Done? How is it calculated? Why is it constantly going up even though my computer is not increasing in speed?

    Profile Gary Roberts
    Forum moderator
    Send message
    Joined: Feb 9 05
    Posts: 2927
    Credit: 816,615,132
    RAC: 1,614,369
    Message 105603 - Posted 18 Aug 2010 1:20:55 UTC - in response to Message 105599.

      Hi Kuntz - welcome to Einstein!

      You obviously have a very decent machine with >10K total credits after just a couple of days :-).

      What is the Avg. Work Done? How is it calculated?

      If you head over to the BOINC Website and click on the 'documentation' link under the 'Volunteer' heading near the top of the page, you will be taken to a Wiki where you can find information on many of these sorts of questions. If you scroll down through all the links on the main page of the Wiki, you will find one in the miscellaneous section near the bottom of the page which is directly about credit. While you are on the main page, take a look at all the other available topics as well in case you find other things of interest.

      Why is it constantly going up even though my computer is not increasing in speed?

      Recent Average Credit (RAC) is a moving average taken over 10 days (I think) of your machine's production. On day one, after first joining, its value will be the average of 9 days of zero production and one day of full production. On day two, .... - I'm sure you get the picture.

      It is also being affected by delays in granting credit if your 'wingman' is a bit slower in returning results than yourself. It wouldn't be unusual for it to take a couple of weeks or more to reach a semblance of stability.

      ____________
      Cheers,
      Gary.

      Profile Mike Hewson
      Forum moderator
      Avatar
      Send message
      Joined: Dec 1 05
      Posts: 3320
      Credit: 27,251,173
      RAC: 22,004
      Message 105605 - Posted 18 Aug 2010 1:47:58 UTC - in response to Message 105603.

        It is also being affected by delays in granting credit if your 'wingman' is a bit slower in returning results than yourself.

        Do we ( attempt to ) match 'like with like' with the locality scheduling features ? He's certainly at the top end of performance, low turnaround times indeed. He may peak RAC in the top 100 I think.

        Cheers, Mike.

        ____________
        "I have made this letter longer than usual, because I lack the time to make it short." - Blaise Pascal

        Profile Gary Roberts
        Forum moderator
        Send message
        Joined: Feb 9 05
        Posts: 2927
        Credit: 816,615,132
        RAC: 1,614,369
        Message 105607 - Posted 18 Aug 2010 2:21:43 UTC - in response to Message 105605.

          It is also being affected by delays in granting credit if your 'wingman' is a bit slower in returning results than yourself.

          Do we ( attempt to ) match 'like with like' with the locality scheduling features ?

          No. Locality scheduling just keeps track of which hosts already have which LIGO data files and so tries to minimise unnecessary downloads of additional LIGO data wherever possible. It may well be that a very fast host and a very slow host are paired up with the same LIGO data files and (together with other hosts as well) are crunching on the same series of tasks.

          He's certainly at the top end of performance, low turnaround times indeed. He may peak RAC in the top 100 I think.

          Even if so, there really is no problem with being paired up with an old, slow host. If you think about it, it's much more preferable to be paired up with a slow host that returns one result every two days than with a fast host with an 8 day cache :-). Yes, they do exist :-).
          ____________
          Cheers,
          Gary.

          Profile Ageless
          Avatar
          Send message
          Joined: Jan 26 05
          Posts: 2941
          Credit: 5,251,019
          RAC: 1,475
          Message 105609 - Posted 18 Aug 2010 4:43:47 UTC - in response to Message 105603.

            Last modified: 18 Aug 2010 4:44:00 UTC

            Why is it constantly going up even though my computer is not increasing in speed?

            Recent Average Credit (RAC) is a moving average taken over 10 days (I think) of your machine's production. On day one, after first joining, its value will be the average of 9 days of zero production and one day of full production. On day two, .... - I'm sure you get the picture.

            In your example, the RAC would only go up. But it doesn't. You forgot about the decay. ;-)

            Each time new credit is added, the RAC is recalculated. Once every week it's also divided in half. This latter counts for RAC to go down, especially when no credit is added or when you stop crunching completely.

            For the complete drop-down (albeit somewhat old and perhaps not completely correct anymore versus the new credit system), see RAC in the Unofficial Wiki.
            ____________
            Jord

            Profile Brett Kuntz
            Send message
            Joined: Aug 13 10
            Posts: 4
            Credit: 1,778,319
            RAC: 0
            Message 105612 - Posted 18 Aug 2010 5:23:48 UTC

              Thanks guys! I also found the built-in benchmarker so I will test out a few different configurations tomorrow to see what clock speeds make for the best crunching!

              Profile Mike Hewson
              Forum moderator
              Avatar
              Send message
              Joined: Dec 1 05
              Posts: 3320
              Credit: 27,251,173
              RAC: 22,004
              Message 105620 - Posted 18 Aug 2010 10:46:43 UTC - in response to Message 105612.

                see what clock speeds make for the best crunching!

                Apart from pure speed, decent memory size and bandwidth, a major determinant is floating point efficiency. There has been much work devoted to optimising the relevant code sections. For GW tasks at least, one part is processor bound, the other memory bound. I can't remember where we are with the Intel vs AMD debate, probably on the Intel side last I recall. Of course there is the Power PC or whatever Apple is serving these days. Also while the screensavers are nice, they do slow the crunching.

                @Gary :

                Even if so, there really is no problem with being paired up with an old, slow host. If you think about it, it's much more preferable to be paired up with a slow host that returns one result every two days than with a fast host with an 8 day cache :-). Yes, they do exist :-).

                Well put. I never thought of it that way.

                Cheers, Mike.
                ____________
                "I have made this letter longer than usual, because I lack the time to make it short." - Blaise Pascal

                Profile Brett Kuntz
                Send message
                Joined: Aug 13 10
                Posts: 4
                Credit: 1,778,319
                RAC: 0
                Message 105641 - Posted 18 Aug 2010 19:09:10 UTC

                  Aside from BSOD's and freezing, if a tiny calculation error takes place, what signs or errors should I look for?

                  Profile Mike Hewson
                  Forum moderator
                  Avatar
                  Send message
                  Joined: Dec 1 05
                  Posts: 3320
                  Credit: 27,251,173
                  RAC: 22,004
                  Message 105643 - Posted 18 Aug 2010 19:52:51 UTC - in response to Message 105641.

                    Aside from BSOD's and freezing, if a tiny calculation error takes place, what signs or errors should I look for?

                    I'm not aware of the E@H apps giving BSOD's/freezing per se. The calculation errors ( if any ) get picked up back at the server side ( ie. the E@H validators, after you have returned the workunit ) by various tests including with respect to other quorum members' results for that work unit - a given unit of work is replicated and sent to several distinct hosts for precisely that purpose.

                    Cheers, Mike.

                    ____________
                    "I have made this letter longer than usual, because I lack the time to make it short." - Blaise Pascal

                    Profile Brett Kuntz
                    Send message
                    Joined: Aug 13 10
                    Posts: 4
                    Credit: 1,778,319
                    RAC: 0
                    Message 105654 - Posted 19 Aug 2010 3:43:11 UTC

                      Well my computer is being a poopy pants and not wanting to run at 4.2GHz anymore, I think silicon degradation has permanently forced me down to 4GHz. On the plus side I can now run my memory at 1905MHz CL7 as before it was 1676MHz CL6. I'll work on my memory timings a little bit this week and try to get them down quite a bit to make up for it.

                      Post to thread

                      Message boards : Getting Started : What is Avg Work Done (Credits)?


                      Home · Your account · Message boards

                      This material is based upon work supported by the National Science Foundation (NSF) under Grants PHY-1104902, PHY-1104617 and PHY-1105572 and by the Max Planck Gesellschaft (MPG). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the investigators and do not necessarily reflect the views of the NSF or the MPG.

                      Copyright © 2014 Bruce Allen