Results 1 to 17 of 17

Thread: Rosetta installing new credit system today

  1. #1
    Ancient Haggis Hound Angus's Avatar
    Join Date
    Jan 2002
    Location
    Seattle/Norfolk Island
    Posts
    828

    Rosetta installing new credit system today

    The site is currently down while they make changes to use a new fixed-credit-per-model credit system.

    It appears that the anti-optimized-BOINC-client side of the controversy has won.

    They also were talking about making this credit change retroactive back to February 2006. That should make some interesting swings in the stats if a significant portion of certain teams were all using optimised clients which supposedly inflated their credits.

    Let the gnashing of teeth and screaming begin.

  2. #2
    Administrator Bok's Avatar
    Join Date
    Oct 2003
    Location
    Wake Forest, North Carolina, United States
    Posts
    24,451
    Blog Entries
    13
    I started reading some of the threads this morning, but they were just way too long to be bothered..

    Retroactive application would be just a waste of time I think. Doubt it would change the rankings too much, though someone like 'Housing And Food Services' would benefit from a lot of the top 10 being demoted as I know Ethan ran the standard Boinc client all along.. If they do that, I'll take a snapshot of the stats as they last stood for posterity

    Bok

  3. #3
    Free-DC's Prime Search
    Join Date
    Apr 2004
    Posts
    2,518
    So everyone cheated?!?!??!
    You guys should forget BOINC.

    So how do you cheat on eOn, SOB, DPAD, the classic projects? Very difficult, in the math projects you have doublechecking. That's why I hate BOINC, sounds like monopoly of DC projects.

    Carlos
    Last edited by em99010pepe; 08-16-2006 at 06:17 PM.

  4. #4
    BOINC has a lot of good points. It's fairly easy to roll out, it's easy to run lots of different projects with it and is highly configurable. But the scoring is problematic.

    Optimised clients did serve a purpose at one point, and in theory they are fine. The problem with them is on projects with no quorum or high caps on work unit values. If you use an optimised client on say... LHC, it doesn't matter (so much anyway).

    But on Rosetta, the only way I had of remaining competitive was to use an optimised client. So I end up getting more credit than those who download the standard BOINC client, which isn't fair on them, and I still get less credit on an Athlon XP than I do on an Athlon 64 - even though on Rosetta (which is FPU intensive) they are clock for clock almost identical in performance - all because SSE2 optimisations hugely inflate the benchmark scores. Lack of quorum means they never get balanced out either.

    I'd support them backdating the credit change. I'd probably lose some myself, but I'd rather see those who were getting more than is fair be evened out. A fixed credit system works great on Climate Prediction, so I don't see why it shouldn't work on Rosetta.

  5. #5
    Quote Originally Posted by Biggles
    But on Rosetta, the only way I had of remaining competitive was to use an optimised client.
    Same here Biggles, but after spending the last 24 hour's or so running under the new system without using any Optimized Client & taking just what they are giving for Credit it's time to drop the Project for me I think.

    There isn't almost any other BOINC Project I couldn't go to and get double or even more Credit without using an Optimized Client than what Rosetta's giving me right now.

    All my PC's are 3.06 Ghz in stock configuration & higher, some of them are Overclocked & 4-5 Credits per hour just isn't getting it for me.

  6. #6
    Quote Originally Posted by PoorBoy
    All my PC's are 3.06 Ghz in stock configuration & higher, some of them are Overclocked & 4-5 Credits per hour just isn't getting it for me.
    If you are just in it for the cross-project stats, then sure the new system isn't as good for you. But even if you are just in it for the stats on Rosetta, you aren't getting any less than anyone else doing the same amount of work.

    What they did is basically level the playing field so that everyone gets the appropriate credit for the actual amount of work they have done. Other projects have a way to double check results or some send out WUs multiple times to balance this out but extra WUs is just a waste of CPU on this project.

    If seeing 4-5 credits per hour is too low, why don't you ask the project admins to add a 10 or 100 multiplier so that everyone would ge 40-50 or 400-500 credits for the same work. You won't be any further ahead than anyone else but the numbers would look bigger for you if that makes you feel better.

  7. #7
    Quote Originally Posted by Angus
    The site is currently down while they make changes to use a new fixed-credit-per-model credit system.
    I thought the new credit system was just on RALPH but the news page makes it sounds like it will be on the regular project as well. I guess I will have to watch my credits to see if they change much. For the stats I guess you don't need to update the client at all, it does everyone on the server end to figure out the new credits?
    Last edited by Digital Parasite; 08-18-2006 at 09:23 AM.

  8. #8
    Quote Originally Posted by Digital Parasite
    What they did is basically level the playing field so that everyone gets the appropriate credit for the actual amount of work they have done.
    That still has to be proven that they have leveled the playing field, I did a lot of checking on Computers @ the Project last night. I was only looking for Computers like mine, some of them where getting about the same as me & some of them where getting 40-50 Credits Per hour.

    Even if these computers where Overclocked I hardly think that they where putting out 10 times more work to deserve 10 times more Credit ...

  9. #9
    Quote Originally Posted by PoorBoy
    That still has to be proven that they have leveled the playing field, I did a lot of checking on Computers @ the Project last night. I was only looking for Computers like mine, some of them where getting about the same as me & some of them where getting 40-50 Credits Per hour.

    Even if these computers where Overclocked I hardly think that they where putting out 10 times more work to deserve 10 times more Credit ...
    Yes of course they will have to make sure the method they are using is actually fair, and I'm sure it will be tweaked over time. Note that each protein is different so the same machine working on two different proteins will generate a different amount of credits. So you need to compare systems that are crunching the same protein. Even then it won't be exactly the same but shouldn't be 10x different.

  10. #10
    Senior Member
    Join Date
    Feb 2004
    Location
    You can always tell a Brummie, but you can't tell him much.
    Posts
    189
    So, Rosetta has joined the growing number of BOINC projects abandoning the BOINC benchmark for their own scheme. There seems alot of emotion over at the boards at the moment, some folk seem to be crowing about an "anti-cheat credit system". Some pharmers are even pulling their boxen for the time being in protest.

    The project admins are very quiet that makes me think they aren't so sure that this new credit system is as good as claimed.

    Restating credits? Been here before on FAD. I got lucky, my stats were sorted within a month. Lauren will be able to tell you exactly how long it can take.

  11. #11
    Ancient Haggis Hound Angus's Avatar
    Join Date
    Jan 2002
    Location
    Seattle/Norfolk Island
    Posts
    828
    The new scoring method must still be broken. It would be a crime to use a broken algorithm to apply retroactively back to Feb.

    Supposedly each 'decoy' with a certain WU type will get the same credit. I returned 4 WU last night, all from the same WU type, sequential WU numbers, and got drastically different credit for each one, and the time crunched was virtually the same.

  12. #12
    Quote Originally Posted by Digital Parasite
    I thought the new credit system was just on RALPH but the news page makes it sounds like it will be on the regular project as well. I guess I will have to watch my credits to see if they change much. For the stats I guess you don't need to update the client at all, it does everyone on the server end to figure out the new credits?
    You won't see any change in the actual stats as they are still calculated exactly the same way as they were before. There has just a new column been added to the "Results" pages in your Account named "granted work credit" which reflects the new system. Only when the system works properly for the different machines out there it will become the system to feed the stats pages. And I can see from my results that it is not really exact. The same WU type of the same protein gets between 350 and 120 credits for the same number of decoys in 12 hours.

  13. #13
    =>Team Joker<= LAURENU2's Avatar
    Join Date
    Dec 2004
    Location
    Chicago IL USA
    Posts
    5,478
    Blog Entries
    1
    Quote Originally Posted by Petey
    So, Rosetta has joined the growing number of BOINC projects abandoning the BOINC benchmark for their own scheme. There seems alot of emotion over at the boards at the moment, some folk seem to be crowing about an "anti-cheat credit system". Some pharmers are even pulling their boxen for the time being in protest.

    The project admins are very quiet that makes me think they aren't so sure that this new credit system is as good as claimed.

    Restating credits? Been here before on FAD. I got lucky, my stats were sorted within a month. Lauren will be able to tell you exactly how long it can take.
    A Long Time It took about a year + to reclaim the (I Think) 60,000 missing Jobs

  14. #14
    Senior Member
    Join Date
    Feb 2004
    Location
    Maryland, USA
    Posts
    771
    Quote Originally Posted by Christoph_Jansen
    You won't see any change in the actual stats as they are still calculated exactly the same way as they were before. There has just a new column been added to the "Results" pages in your Account named "granted work credit" which reflects the new system. Only when the system works properly for the different machines out there it will become the system to feed the stats pages. And I can see from my results that it is not really exact. The same WU type of the same protein gets between 350 and 120 credits for the same number of decoys in 12 hours.
    Here's one of my credits which resulted in 0 "granted work credit". Seems like a normal one -- 6K cputime, outcome success -- when I compare with other WUs, makes me wonder about that new system of theirs...


    Result ID 33278505
    Name t383__CASP7_JUMP14_JUMPRELAX_SAVE_ALL_OUT_BARCODE_hom003__1035_7089_2
    Workunit 26983484
    Created 18 Aug 2006 4:51:20 UTC
    Sent 18 Aug 2006 4:55:08 UTC
    Received 18 Aug 2006 11:34:24 UTC
    Server state Over
    Outcome Success
    Client state Done
    Exit status 0 (0x0)
    Computer ID 57352
    Report deadline 25 Aug 2006 4:55:08 UTC
    CPU time 6038.28604
    stderr out

    <core_client_version>5.2.14</core_client_version>
    <stderr_txt>
    Graphics are disabled due to configuration...
    # random seed: 2612382
    # cpu_run_time_pref: 7200
    # DONE :: 1 starting structures built 4 (nstruct) times
    # This process generated 4 decoys from 4 attempts


    BOINC :: Watchdog shutting down...
    BOINC :: BOINC support services shutting down...

    </stderr_txt>

    Validate state Valid
    Granted credit 20.8765330929665
    Granted work credit 0
    application version 5.25

  15. #15
    Please post these on the R@H message board. . it's just the type of data they are looking for.

  16. #16
    Senior Member
    Join Date
    Feb 2004
    Location
    Maryland, USA
    Posts
    771
    Quote Originally Posted by Exci
    Please post these on the R@H message board. . it's just the type of data they are looking for.
    {Chuckle} that's a good one. I'll consider it, not, whilst I lower my Rosetta allocation. I hear the party over at WCG is warming up! They welcome Linux and SSE2 crowds there.

  17. #17
    Rosetta is native to linux and doesn't use SSE2, the new work credits should reflect that. . . given bugs are reported.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •