570


Advanced search

Message boards : Comments and discussion : 570

AuthorMessage
Fischer-Kerli
Project donor
Send message
Joined: Mar 24 05
Posts: 70
Credit: 78,436
RAC: 0
Message 6523 - Posted 2 Aug 2007 12:37:55 UTC

    This is computed in two rounds on my system (WinXP SP2, AMD dual core, BOINC 5.10.13): after reaching some 98%, progress goes back - mostly to negative percent values ... wonder how they are calculated - and then climbing up again. Seems to be a cosmetical issue only, all results have validated OK so far if the quorum was there.

    Achim
    Send message
    Joined: May 17 05
    Posts: 140
    Credit: 544,866
    RAC: 0
    Message 6525 - Posted 2 Aug 2007 18:34:46 UTC

      Same for me.
      At least some of the WU\'s. E.g. this one

      Profile Janus
      Forum moderator
      Project administrator
      Send message
      Joined: Jun 16 04
      Posts: 2848
      Credit: 53,382
      RAC: 0
      Message 6527 - Posted 2 Aug 2007 19:08:21 UTC

        Noted.

        Fischer-Kerli
        Project donor
        Send message
        Joined: Mar 24 05
        Posts: 70
        Credit: 78,436
        RAC: 0
        Message 6528 - Posted 2 Aug 2007 20:18:28 UTC

          Janus - how can a progress of -x% come about? Just curious ...

          Profile Janus
          Forum moderator
          Project administrator
          Send message
          Joined: Jun 16 04
          Posts: 2848
          Credit: 53,382
          RAC: 0
          Message 6531 - Posted 2 Aug 2007 23:33:24 UTC - in response to Message 6528.

            Last modified: 2 Aug 2007 23:36:29 UTC

            Janus - how can a progress of -x% come about? Just curious ...


            The client uses interpolation between the progress updates it gets from Blender to predict the actual progress. It does this to be able to display progress smoothly in BOINC (it updates the estimate almost every second).

            In other words, Blender says:

            .
            10%
            .
            40%
            .
            80%
            .
            100%

            At the first 10% the client knows how long time it took to calculate the 10% and will then be able to estimate how long the next 10% are going to take. So it will do like:

            .
            10%
            28%
            40%
            60%
            80%
            90%
            100%

            But if Blender reports the progress as split into more than one pass (which could very well be the case for an effect such as subsurface scattering) it will go from 0 to 100% and then do another run from 0 to 100%. At the point where Blender switches from 100 to 0 the client sees a negative progress with the time between 100 and 0 as the delta timing value. It will try to figure out how long before you reach -200% and use that to make it\'s next progress estimates...
            Now after a while Blender reports the next progress update (which should then be slightly more than 0) and the progress estimates from the client then become sane again (positive).

            The current code in the Blender status system is a bit of a mess, meaning that there\'s no consistent way to get the total number of passes (since a pass can include another pass etc). So basically I\'ll have to make the Blender patchset know about all the weird ways you can combine effects, fields, layers, compositing and scene inclusions in order to completely avoid this problem. Not an easy task, but not impossible either.
            I\'ve chosen to simply add support for new effects etc. as we go along and people start to use them.

            At the moment the Blender patchset supports correct progress reporting for fields and for motionblur but not for sss, layers, postprocessing effects and scene inclusions.

            Fischer-Kerli
            Project donor
            Send message
            Joined: Mar 24 05
            Posts: 70
            Credit: 78,436
            RAC: 0
            Message 6532 - Posted 3 Aug 2007 0:08:35 UTC

              Thank you!

              Post to thread

              Message boards : Comments and discussion : 570