Message boards : Number crunching : A Challenge
Previous · 1 . . . 4 · 5 · 6 · 7
Author | Message |
---|---|
Tony DeBari Send message Joined: 13 Apr 06 Posts: 12 Credit: 2,944,090 RAC: 0 |
Cross project credits is also broken like with seti 'optimised' apps since the optimised apps are not really doing more work (they are just more efficient, they do less work to get teh same result so why should they get more than me here for the same of flip-flops ;-) First, cross-project credit calibration has to be based on the official science apps for the respective projects. Not all projects have third-party optimized apps and even in the ones that do, they are used by a small (but very productive) minority of participants. Second, the optimized Seti Enhanced apps are neither doing more work nor are they doing less work than the official app. They are doing the same work - as measured in FPOps - and so are entitled to the same credit for a given WU. The advantage is that they do the work faster, thereby earning a higher credit/hour than the official app. This is not inconsistent with a desire to equalize credit/hour across projects because again, cross-project calibration must be based on official science apps (and IMHO official core clients) only. Now, if a time should come when Seti@Home incorporates some or all of the processor-specific optimizations that are currently in the third-party apps into their official app - as happened in Einstein@Home - then they should lower their credit/WU to offset the increase in crunching speed. To not do so would unbalance the cross-project calibration they are trying so hard to maintain. [edit]As the "flagship" of the BOINC fleet, this would set a poor example for other projects dealing with the same issue. Regards, -- Tony |
carl.h Send message Joined: 28 Dec 05 Posts: 555 Credit: 183,449 RAC: 0 |
Tony in projects where a quorum exists and people are running 5.5 even those running std will gain from the 5.5 users such as at WCG where a quorum of 3 exists. If two are using 5.5 the higher points are given. It`s interesting to note that those from Seti who were legitamely using 5.5 also are a lot of x project users therefore would bring the 5.5 to play in all projects they use. If such cross project calibration exist`s it isn`t working and I doubt it will. As for those who dislike the idea of quorums and see them as a waste of resources, I disagree, science has to be accurate therefore each experiment needs repeating as is done in the physical field. Just imagine that someones machine misses the one piece that holds the key because of OC`ing or something else ! Not all Czech`s bounce but I`d like to try with Barbar ;-) Make no mistake This IS the TEDDIES TEAM. |
suguruhirahara Send message Joined: 7 Mar 06 Posts: 27 Credit: 173,220 RAC: 529 |
Could someone explain why the amount of credits granted per day has been obviously decreased in this week? It's strange... |
Astro Send message Joined: 2 Oct 05 Posts: 987 Credit: 500,253 RAC: 0 |
Could someone explain why the amount of credits granted per day has been obviously decreased in this week? It's strange... I'm not seeing it being lower this week. [edit]chart updated to include return dates and legends |
tralala Send message Joined: 8 Apr 06 Posts: 376 Credit: 581,806 RAC: 0 |
Could someone explain why the amount of credits granted per day has been obviously decreased in this week? It's strange... It could be due to the fact, that a fraction of 5.5.0 users have left and more stock clients have joined, together with the fact that old overcredits WU from the RALPH runs are no longer in the system. My subjective "felt decrease" is about a few % though. |
Christoph Jansen Send message Joined: 6 Jun 06 Posts: 248 Credit: 267,153 RAC: 0 |
Could someone explain why the amount of credits granted per day has been obviously decreased in this week? It's strange... I'd rather guess you allude to the fact that the total credits/day of the project have decreased from roughly 3.9 mio to 3.5 mio. and that the TeraFlops have slided accordingly? That number is now incomparable to the credits/day of last week due to the simple fact that now credits are counted in a different way. At the same time some of the crunchers with pretty high daily outputs have left and, as you can see here at BOINCstats, a number of new users have joined the project and some inactive ones seem to have returned. As a whole the turnover in users was significant. Its numeric consequences cannot be exactly assessed and one cannot really say whether the amount of work really done per day has increased or decreased. So just forget all numbers till last week and watch how the new ones evolve. As I said, it is incomparable and all else is pure speculation unless you have a source telling you what the current credits would be in old units - which nobody of us has. Regards, Christoph |
Tony DeBari Send message Joined: 13 Apr 06 Posts: 12 Credit: 2,944,090 RAC: 0 |
Quoting Carl.h: Tony in projects where a quorum exists and people are running 5.5 even those running std will gain from the 5.5 users such as at WCG where a quorum of 3 exists. If two are using 5.5 the higher points are given. This is true. I freely admit to occasionally having benefited from this phenomenon myself. However, I submit that the likelihood of a credit quorum being formed with two optimized core clients (I choose not to single out Crunch3r's 5.5.0 as they all claim higher-than-standard credit) is exceedingly small and so can be safely ignored from a cross-project calibration standpoint. It`s interesting to note that those from Seti who were legitamely using 5.5 also are a lot of x project users therefore would bring the 5.5 to play in all projects they use. Also true, plus don't forget those from Einstein as well during the S4 run. In theory they were supposed to revert back to the official client when the need for an optimized client was removed; i.e. the arrival of Seti Enhanced and Einstein S5. However, as we all know, this didn't happen in a significant number of cases and thus complicates the calibration effort. If such cross project calibration exist`s it isn`t working and I doubt it will. The desire to keep the projects relatively equal is certainly there among the project staffs and I dare say the majority of users. It may not be working now, and perhaps you are right that it will never be fully achieved. But I think persuing it is worth the effort, considering that it is supposed to be a 'feature' of the BOINC platform. Regards, -- Tony [edit - added quote ID for clarity and corrected typos] |
Angus Send message Joined: 17 Sep 05 Posts: 412 Credit: 321,053 RAC: 0 |
Could someone explain why the amount of credits granted per day has been obviously decreased in this week? It's strange... Excellent analysis of the situation. Clearly points out that there can be NO before/after "the change" comparision of credits - the true numbers are not available. Proudly Banned from Predictator@Home and now Cosmology@home as well. Added SETI to the list today. Temporary ban only - so need to work harder :) "You can't fix stupid" (Ron White) |
David E K Volunteer moderator Project administrator Project developer Project scientist Send message Joined: 1 Jul 05 Posts: 1018 Credit: 4,334,829 RAC: 0 |
Here are the numbers since the change: claimed_credit: 32466190.9148111 claimed_work_credit: 27367705.102651 granted_credit: 27320061.827033 (valid results) granted_credit/claimed_credit: 0.841493 So you can approximate what the TFLOPS would be with the older system by just dividing the estimate reported on the home page by 0.84 which is ~42 currently. To be more accurate, I could just consider the results returned in the last 24 hours but I think it's okay for a rough approximation. This is actually an underestimate because I extracted the information from our logs and the claimed_credit is sometimes not reported in the logs for invalid results (which was granted under the old system). |
Message boards :
Number crunching :
A Challenge
©2024 University of Washington
https://www.bakerlab.org