Message boards :
Number crunching :
Seti Enhanced Credit Fair?
Message board moderation
Previous · 1 . . . 7 · 8 · 9 · 10 · 11 · 12 · 13 . . . 23 · Next
Author | Message |
---|---|
SargeD@SETI.USA Send message Joined: 24 Nov 02 Posts: 957 Credit: 3,848,754 RAC: 0 |
So he said he believes optimized clients are cheating.Yes, CLIENTS. Thats the BOINC bit. Either way, he still called us cheaters...... |
Rjmdubois Send message Joined: 27 Sep 99 Posts: 12 Credit: 111,608 RAC: 0 |
[/quote] I don't see competition stopping after the transition is complete, as everybody's RAC will drop by the same percentage. [/quote] I check my ranking at Boincstats, so it's not only a S@H game. At my benchmark computer, a Prescott 3.2, I got: 75 credits/hour running Crunc3r 4.11 65 credits/hour running E@H U41.04 35 credits/hour runnind CPDN standard 25 credits/hour running Rosetta 25 credits/hour running S@H 5.12 So, if I change the share of S@H from 50% to 5%, and increase the share of E@H, my RAC will not drop dramatically. People that care about science, but also like the competition DO have a choice. |
Lazy2 Send message Joined: 4 Sep 00 Posts: 14 Credit: 23,552,278 RAC: 0 |
It looks like Enhanced has the value of my dollar dropping to 33 cents. That kills the fun of the competition. I hope this is resolved so that I don't have to take my measely 130 machines elsewhere to get my money's worth. Optimizing helps one get more bang for the buck without corrupting the science. Credit is important to a lot more people than a few here may think. I believe that I may have to follow Daniel if there is no improvement. Electricity isn't cheap and about 26 of these machines eat up power at home. Individuals such as Crunch3r and Trux should be commended for all the time and effort that they have put into improving this project. So what if people are credit whores. That gives more incentive to crunch and is a cheap way to reward them for doing it. By all these people going after credits and adding machines the science has been helped. To not realize that, one would have to have a short circuit above the neck. This is only a test... |
Daniel Schaalma Send message Joined: 28 May 99 Posts: 297 Credit: 16,953,703 RAC: 0 |
Well, it looks like between me, SargeD, and Lazy, this project could be loosing at least 175 hosts alone. And this is just 3 people. There are hundreds more who feel the same way. But, like someone said, none of this will make any difference, because we "credit mongers" don't count. Our contribution to this project and our crunching ability is worthless, because we also care about credits. P.S. In regards to the comments about us cheating, the last person suspected of ACTUALLY cheating had his account deleted not too long ago, after an entire thread was devoted to that investigation. Regards, Daniel. |
The Gas Giant Send message Joined: 22 Nov 01 Posts: 1904 Credit: 2,646,654 RAC: 0 |
Acknowledgement of people’s contribution is required to keep a majority of people involved with distributed computing. The conundrum that the developers of BOINC have been working on is how best to do this fairly. The benchmark idea was great at the time when only 1 or 2 projects were on BOINC and only a few individuals were optimising, but as more projects came online and optimised apps and clients became available en-masse, its limitations were discovered pretty quickly. Resulting in peoples preference for projects that gave good credit. The idea of counting the number of operations performed came out as a good replacement, but now very early in its public infancy we are seeing that it has problems as well. So how do the developers ensure that the same computer claims the same amount of credit no matter which project is crunched? Do we just want to rely on having faster or more computers to boost our overall RAC? Do we, the volunteers, want to see a fair credit system? Are we for a socialist credit system or a free-market capitalist style of credit system? Do we want a credit system that is transportable across all BOINC based projects? Do we want to see a credit system that is only good on a specific project where we end up only counting the number of wu’s completed and validated per project? Live long and crunch. Paul (S@H1 8888) And proud of it! |
W-K 666 Send message Joined: 18 May 99 Posts: 19356 Credit: 40,757,560 RAC: 67 |
Good Morning, I'm still here. To clear up a few points, I was refering to the BOINC client, not the science application. I did say if you had read all of the post carefully, I use the optimised Seti apps and my complaint and accusation was about the over-inflated credit claims brought about by using 'optimised' BOINC clients. (The BOINC manager etc. on our computers are clients of the main BOINC programs on the Project servers) no one asked for Fpops Maybe nobody did, but why were these or similar questions asked frequently; Why am I only getting 25 credits when I claim 43. Why does this hostid claim 6 credits and granted 23. Why do I only download 10 units for a one day cache when my computer can do 16 units per day. Why has my computer gone into edf mode. On my dual OS machine do I claim 10 credits/unit using OS1 and 26 credits/unit using OS2. etc. etc. At the beginning of last year these questions/complaints were almost daily, and they are all caused by the benchmarks. John McLeod VII (JM7) helped by proposing and then programming the scheduler, which helps in downloading the correct amount of cache units and ensures that work is returned before the deadline, if its left alone. There were many dicussions about the claimed credits problem and the ways to fix it, quite often led, even if not started, by Paul D. Buck. Most of the proposals were non-starters and the systems similar to the present Fpops method were rejected because it was thought it would lengthen the time to crunch a unit. Other methods discussed were for the projects to send every x days a reference unit, and base future claims on how long that took to process. A system of calibrating against standard computers, where your computer benchmark was adjusted every time you met a 'standard host' doing the same unit. etc etc. So we may not have asked for Fpops, but we certainly asked for a system where each computer claimed approximately the same credits for the same unit. For those credit hogs thinking of jumping ship to Einstein; 1. Look at the sever status page 2. Read the S4 search will take more than 200 days to complete I predict you have less than 30 days to fill your credit basket. and in reply to RJM Bubois, my figures are; CPND 16/hr Einstein 12/hr using standaard app Einstein 36/hr using optimisation Seti 4.18 14/hr standard app (June 2005) Seti 4.11 45/hr using Crunch3r's optimised app Seti Enhanced 16/hr all of these figures are without using an 'optimised client' And to all those using enhanced 5.11, have you read this forum_thread.php?id=30709#305178 Andy |
Steve Cressman Send message Joined: 6 Jun 02 Posts: 583 Credit: 65,644 RAC: 0 |
I'll notice if you leave, that's for sure :o) I don't know where you think the competition has gone. As long as we are crunching and getting credits, the competition still exists. And as for you and seti.germany, it looks like you think that germany is going to continue with the old system while you have to use the new system. Talk about a persecution complex. In case you haven't figured it out seti.germany is using the new credit system also, not just you. 98SE XP2500+ @ 2.1 GHz Boinc v5.8.8 And God said"Let there be light."But then the program crashed because he was trying to access the 'light' property of a NULL universe pointer. |
zoom3+1=4 Send message Joined: 30 Nov 03 Posts: 66279 Credit: 55,293,173 RAC: 49 |
It's official, Einstein allows and encourages the use of optimized apps/client and I asked a mod there at this place: http://einstein.phys.uwm.edu/forum_thread.php?id=4050#33718 Savoir-Faire is everywhere! The T1 Trust, T1 Class 4-4-4-4 #5550, America's First HST |
Odysseus Send message Joined: 26 Jul 99 Posts: 1808 Credit: 6,701,347 RAC: 6 |
And as for you [Daniel / SETI.USA] and seti.germany, it looks like you think that germany is going to continue with the old system while you have to use the new system. Talk about a persecution complex. In case you haven't figured it out seti.germany is using the new credit system also, not just you. Well, in fairness I think you're missing his point: the existing lead was established under the old system, so it's proportionally greater under the new. For a greatly exaggerated analogy, suppose you're in a rally race and the car ahead of you has a 20-km lead and averages 100 km/hr; you're doing 120 km/hr so you expect to catch up in one hour. Suddenly the roads become impassable and you’re both forced to proceed on foot, the leader at 5 km/hr and you at 6 km/hr (magically maintaining the same ratio), so that 20-km lead will now take you 20 hours to make up. Looking at the Top Teams page, I see that SETI.USA’s RAC is about 100,000 greater than SETI.Germany’s, and their total is about 80 million less. So if the status quo were maintained they could expect to catch up in 800 days, somewhat over two years. If we accept, for the sake of argument, the premise that both teams’ production will now drop by a factor of three, the time required to overtake will increase to 2400 days or six and a half years. Note, however, that this is also based on the assumption that all the members of both teams are running optimized v4.x apps. A more realistic assessment would take account of the actual prevalence of optimized workers, considering that the earnings per hour of the standard app and the Enhanced are pretty similar. If half the members of each team are running stock apps, for example, still accepting Daniel’s claim of a 67% earnings drop from optimized 4.x to 5.x, the overall RAC ‘devaluation’ will be only about 35%, yielding a time-to-overtake of three years or so. IMO even two years is a very long time in the computing world. I’m reminded of the story about a layman attending an astronomy lecture, and becoming quite upset on hearing that the Sun will become a red giant, vaporizing the Earth, and then burn out. The lecturer points out that five billion years is an unimaginably long time in the future, with no relevance to anyone alive today, or even their great-great-…-grandchildren. Much relieved, the visitor says, “Oh, that’s all right, then. I thought you said five million years.†|
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13841 Credit: 208,696,464 RAC: 304 |
If there are some chunchers that want more credit, give them. Then other projects up the credits to get them back, so Seti ups the credits to get them back, then the other projects up the creditys to get them back. Doesn't sound like a particulalry good idea to me. It's much easier than raising money thru donnations. True, but i expect the money would be appreciated much more than the "donated" time on hardware. For me donated implies giving their time. From the looks of it it's more a case of whoring for credits, not donating. Grant Darwin NT |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13841 Credit: 208,696,464 RAC: 304 |
How long have you been here??? Credits & the claiming & allocation of them have been the single biggest bitching point since BOINC came out. Like it or not, fpop counting addresses most of those issues. Grant Darwin NT |
Jack Gulley Send message Joined: 4 Mar 03 Posts: 423 Credit: 526,566 RAC: 0 |
Lets settle this the hard way. Lets compare: Apples(4.18) to Apples(5.12) and Oranges(4.11) to Oranges(5.12) And stop letting some developers call Oranges(5.12) an Apple(5.12), as Apples(5.12) does not exist. Step 1: Remove all the optimization code (trig cache, etc.) from Seti Enhanced 5.12 so that we have Apples(5.12) instead of Oranges(5.12) and call it 5.13. Step 2: Now with Seti Enhanced (Apples(5.13) less Oranges(5.12) optimization) determine its worth in Credits based on the current fpop method and time it takes, and calibrate it against the stock 4.18 version as has been explained in this thread. {Hum.. bet that number comes out to be something a lot more than 63 credits, but at least you are comparing Apples(4.18) to Apples(5.12/5.13)}. Step 3: Release Apples(5.13) as .. oops, Seti Enhanced 5.13, as the new "fixed" stock Seti application. Step 4: Optimizers release their new improved Oranges(5.13) that claim the same credit based on the fpop method. Step 5: Allow everyone to now compare Apples(4.18) to Apples(5.13) and to compare Oranges(4.11) to Oranges(5.13) and decide which application they wish to run. This way, this whole debate on Seti Enhanced Credit Fairness would not exist. (Well, except among those who can not find the optimized versions and among the developers.) |
Ingleside Send message Joined: 4 Feb 03 Posts: 1546 Credit: 15,832,022 RAC: 13 |
And noone asked for fpop counting... Well, let's for arguments sake say Seti_enhanced does not use "flops-counting", but still relies on benchmark/cpu-time, if so, what will happen with granted credit? Computer, uses 11400s with "normal" seti-v4.18, and 3600s with "optimized", and average granted is 24.2 CS/result. Let's also for easy's sake say this computer has the same claimed as granted. Seti_Enhanced, ar=0.427219 used 28457.59375s with "normal", claimed/granted 62.239310296005 CS if uses "flops-counting". Let's also say optimized enhanced slashes 20% off cpu-time, giving 22766s for this result. Meaning, optimized is 25% faster, and should give 25% more credit/day than "standard" Seti_Enhanced-application. If relies on the standard benchmark and runs the "standard" application for both v4.18 & v5.12, the claims under Seti_Enhanced is: 24.2 CS / 11400s * 28457.59375s = 60.41 CS If runs optimized Seti_Enhanced but "normal" BOINC-client: 24.2 CS / 11400s * 22766s = 47.09 CS. If runs both optimized Seti_Enhanced and "optimized" BOINC-client: 24.2 CS / 3600s * 22766s = 153.04 CS. If relies on "calibrated" BOINC-client: "old" v4.18 -> fpops_est=2.79248e13 -> 2.79248e13/8.64e11 = 32.32 CS ar=0.427219 -> fpops_est=6.42747e13 -> 6.42747e13/8.64e11 = 74.39 CS. Just to sum it up, the claimed credits under Seti_Enhanced are: Optimized SETI, normal BOINC : 47.09 CS Standard SETI/BOINC : 60.41 CS Flops-counting : 62.24 CS Calibrated BOINC : 74.39 CS Optimized SETI/BOINC : 153.04 CS But, remember while the claims can be largely different, the granted credit is decided at same time wu is validated, and SETI@Home uses these rules: 1; If only 2 passed validation, lowest claimed to all. 2; If 3 or more, remove highest & lowest claimed and average the rest. Any results returned later that also passes validation, gets the same granted credit as the rest, no re-calculation of granted credit is done. SETI@Home is currently running with min_quorum = 3 & target_nresults = 4, hopefully they'll decrease these "soon" but anyway, with these rules the granted credit is normally decided when 3 results is in. Meaning, if relies on benchmark, only if more than 1 of these 3 results is run with optimized seti-application will the granted credit be changed. So, yes, if Seti_Enhanced had choosen to continue to rely on the BOINC-benchmark, someone running optimized seti-application will ocassionally be paired-off with someone else running optimized seti-application, and therefore get 150 CS for a result. But, on next wu can be paired-off with a win9x with "timer-bug" and another there benchmark timed-out, so gets only 1 CS or something... Over time, these spikes to either way will average-out, meaning you'll land very close on 60.41 Cobblestones. But hang on, "flops-counting" gives immediately 62.24 Cobblestones, and this is within 5% of average granted credit you'll get by relying on benchmark/cpu-time, but you're not getting all the spikes. So bottom line is, Seti_Enhanced still relying on benchmark/cpu-time would not have changed the average granted credit for someone running optimized Seti_Enhanced more than maybe 5% compared to it's now instead relying on "flops-counting". This difference is so small, that in my opinion it's insignificant. BTW, while "flops-counting" actually gave higher credit in this example doesn't mean this is true for other angle-ranges, but would expect majority of wu is within 5%-10% of whatever granted credit you'll get from relying on cpu/benchmark instead. |
SargeD@SETI.USA Send message Joined: 24 Nov 02 Posts: 957 Credit: 3,848,754 RAC: 0 |
There is one big hole in your argument. It is currently true only for a very small minority. The numbers of people who have optimized far exceeds those who have not. Case in point: I get at or very near what I claim about 85-90% of the time. The times that I get a very low (or even significantly lower) GC from what I claim are so far and few between that they have a negligible affect on my RAC. So it would appear that the majority do indeed use the optimized client/apps. |
Daniel Schaalma Send message Joined: 28 May 99 Posts: 297 Credit: 16,953,703 RAC: 0 |
Daniel, I forgot to address this point in my previous posts. YES, I thought it would be different when enhanced went live. Before I had to stop testing because I needed to devote more time to my terminally ill father, the enhanced workunits were being GRANTED credits in the range of between 160-230 credits per workunit, depending on angle range. Then, all of a sudden, just as the enhanced app was released for production, the version 5.12 came along, DRASTICALLY cutting granted credit per workunit. So, yes, this is what I was expecting when enhanced went into production. The granted credits remained decent, all the way through version 5.11. Now v5.12 comes along, and BANG. Slap in the face. Only 3 out of 23 machines are now working on enhanced, and already my RAC is dropping like a rock. Eric stated that he was trying to ensure that our GRANTED credits would remain the same as before. I was replying that after the credit CHOP of v5.12, this is simply NOT POSSIBLE. When the time to process goes up by a factor of SIX, but the credit per workunit goes up by a factor of TWO, the result is that your net granted credit goes DOWN by a factor of THREE. And that is exactly what is happening on my machines. It is very disconcerting to have built up some outstanding stats, and for all my time, effort and money spent, now seeing it all crumble before my eyes in a short amount of time. It is also interesting that no one with an opposing view of this has an RAC of over 5000! Regards, Daniel. |
W-K 666 Send message Joined: 18 May 99 Posts: 19356 Credit: 40,757,560 RAC: 67 |
There is one big hole in your argument. It is currently true only for a very small minority. The numbers of people who have optimized far exceeds those who have not. ..... And were do you get this information from, this is not somehing I see when looking at the other hosts crunching the same umits as my computers. The average person out there loads the software and runs it, they almost never visit these pages. It is one of the reasons it takes so long to get the BOINC client updated because until the Project insists on a minium version the majority never update it. Most of the people who post on the Q & P pages, who have credit, do so not realise they would get a faster response, and frequently a more knowledgable one, on this board to the problem they have. Andy |
Geek@Play Send message Joined: 31 Jul 01 Posts: 2467 Credit: 86,146,931 RAC: 0 |
BIG snip.......... Sorry to inform you that my RAC was at 6500 and has taken a hit as everyone's has. I alway's felt that since I was using an "optimized" application that it's use could be withdrawn at any time and then be forced to use the normal distributed app. I alway's felt the RAC near 6500 was artificially enhanced. Now it's back to normal. [edit]WHY did I use the "optimized" client then? Cause my personality demands the newest, best, biggest of anything.[/edit] Boinc....Boinc....Boinc....Boinc.... |
Bob Bihari Send message Joined: 21 May 99 Posts: 6 Credit: 1,124,128 RAC: 0 |
Well, It's fascinating reading all the justifications being espoused by folks about how this new credit system is "really just the same because everyone is in the same boat" and various formulae and convoluted logic to justify how things "really aren't any different..." (if you have a quart of pineapples and exchange them for a stack of brass slugs....Yada Yada......) My Seti Stable is pretty small in comparison to many, but I just got my electric bill on Saturday..... my power usage for my home systems and associated hardware (on a separate circuit) was over $230 for the month.....and this doesn't include the strain on my air conditioning (this is Florida, and it's getting hotter by the day). I know the standard response will be "well, nobody asked you to do this.....etc etc" ....well ,you're right. It's just a hobby. I guess I'm competitive by nature, but without the competition and camaraderie of belonging to a competitive team, it has no appeal to me. Obviously, your mileage may differ. I figure with what I can get by auctioning off some systems, what I typically spend on new/upgrading crunchers, monthly electric/air-conditioning bills, (let alone going deaf from listening to fan noise)etc, I can buy myself another Harley, or a new car. A NICE new car...... Sounds like a no-Brainer. ~I like my beer dark, cigars strong, coffee black, bourbon straight and politicians on the end of a rope~ Mark Twain |
Ingleside Send message Joined: 4 Feb 03 Posts: 1546 Credit: 15,832,022 RAC: 13 |
There is one big hole in your argument. It is currently true only for a very small minority. The numbers of people who have optimized far exceeds those who have not. Case in point: I get at or very near what I claim about 85-90% of the time. The times that I get a very low (or even significantly lower) GC from what I claim are so far and few between that they have a negligible affect on my RAC. So it would appear that the majority do indeed use the optimized client/apps. Well... one of your computers has currently 71 with granted credit, there your average claimed is... 30.87 CS/result while average granted is... 24.48 CS/result... If you starts digging deeper into individual wu, you do have some wu there all has used optimized seti-application, but on many others you're the only one. Also, even some is using optimized seti-application, some is not using optimized BOINC-client, meaning they're claiming around 5-10 CS/result... There's also another point, let's say before Seti_Enhanced was released, seti got 1 M result/day, thereof optimized was responsible for 500k of these results. If cpu-times goes up 2.5x, it means afterwards the "normal" users crunches... 500k / 2.5 = 200k results/day. The optimized on the other hand... 500k * 1.25 / (3 * 2.5) = 83k. Meaning, the "50%" suddenly dropped to 30% under Seti_Enhanced. Remember, the optimizers isn't 3x faster any longer, they're maybe only 1.25x faster, and will therefore be responsible for a smaller part of all results, and has therefore also less chance to influence granted credit. And no, have no idea how many was really running optimized seti-applications. |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
Actually, I remember quite a few threads where people said "why can't they just count the actual number of operations?" |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.