Message boards :
Number crunching :
Seti Enhanced Credit Fair?
Message board moderation
Previous · 1 . . . 15 · 16 · 17 · 18 · 19 · 20 · 21 . . . 23 · Next
Author | Message |
---|---|
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
I believe that the curve is a myth. If we're actually counting floating point operations, then the count is the count. We then multiply the count by some scaling factor, and that's the score. If a WU is ten times as long, there will be ten times the number of floating point ops, and we'll multiply a bigger number times the scaling factor. |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
It is an issue because an hour of SETI is supposed to be about the same as an hour of Einstein and about the same as an hour of Rosetta, or CPDN, or Hash Clash. If SETI suddenly starts giving out "double credit" then those BOINC users who are "shopping" for credits will move to BOINC at the expense of the other projects. So, to compensate, Einstein might artifically increase their credits, and now we've got a shift from SETI to Einstein. Then Rosetta starts giving out 5x credits, and those who like credits over everything start crunching Rosetta. ... and if that continues, someday 1 minute of crunching will be worth about a million cobblestones. SETI is trying to avoid credit inflation. |
Keck_Komputers Send message Joined: 4 Jul 99 Posts: 1575 Credit: 4,152,111 RAC: 1 |
I believe that the curve is a myth. It is possible that you are correct, however I think there is a curve to be considered. We talk about a FLOP count almost exculsively, but that is not complete there is an IOP count as well. There is a logical basis for mostly ignoring the IOP count, SETI is very heavily weighted towards FLOPs. I suspect however that the ratio between IOPs and FLOPs does change depending on the angle range. The question is if this change in the ratio is enough to make a difference in the final numbers. BOINC WIKI BOINCing since 2002/12/8 |
sideband@seti.usa Send message Joined: 19 Jun 99 Posts: 25 Credit: 2,774,864 RAC: 0 |
Hrm.. An hour and 18 minutes on Einstein on my AMD A64 3000+ Win2KProSP4 machine (on a "Romeo" WU" "Zed"'s take a little longer, but not much)netted around 24 CC and 13.6 GC... SETI Enhanced 5.12 WU's take, on average, FOUR HOURS AND FOURTY-ONE MINUTES on the same machine, on average, and are being CC'ed around 37 minutes and GC'ed around 18 credits, IF I'm lucky Now, you're talking about "fair across the board"... DO THE MATH. Is it any wonder why I've started migrating over to Einstein? Now, it seems to me, that if it's supposed to grant across the board, then either Einstein is broken (which it's not, there ARE no optimized clients for Einstein, that I know of), or SETI Enhanced is.. and I think, from the postings in this thread, it's obvious which is which... 73 de AI8W, Chris Abdico Concussio Fidens Servo Libertas Semper! |
ML1 Send message Joined: 25 Nov 01 Posts: 21118 Credit: 7,508,002 RAC: 20 |
I believe that the curve is a myth.It is possible that you are correct, however I think there is a curve to be considered. ... the ratio between IOPs and FLOPs does change depending on the angle range. The question is if this change in the ratio is enough to make a difference in the final numbers. I doubt that the IOP counts are significant. What likely is far more significant is the difference in the mix of the sizes of FFTs run for each range of AR. (Phew, hope you followed that.) In other words, rather than just FLOPs, you are seeing the effects of the FLOPs count and memory bandwidth usage for the FFTs. Some computer architectures will be more sensitive to the mix of big/small FFTs run than others and the FFT mix changes depending on the AR... In summary: Your hardware will likely have a sweet AR for maximum FLOPs gained. If your system is truly CPU limited (and has infinite memory bandwidth), then the FLOPs should always remain constant and rate of credit should hence remain constant. (Ofcourse, most PC systems suffer a memory bandwidth bottleneck. Certain systems far more so than others...) Happy crunchin', Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13842 Credit: 208,696,464 RAC: 304 |
SETI Enhanced 5.12 WU's take, on average, FOUR HOURS AND FOURTY-ONE MINUTES on the same machine, on average, and are being CC'ed around 37 minutes and GC'ed around 18 credits, IF I'm lucky What is cc'd? My machines are getting around 28.7 Credits for doing around 2.8 hours worth of work. Grant Darwin NT |
ML1 Send message Joined: 25 Nov 01 Posts: 21118 Credit: 7,508,002 RAC: 20 |
What is cc'd? cc: Claimed Credit gc: Granted Credit ? Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13842 Credit: 208,696,464 RAC: 304 |
(Ofcourse, most PC systems suffer a memory bandwidth bottleneck. Certain systems far more so than others...) Now, now. Don't be nasty towards Intel. Grant Darwin NT |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13842 Credit: 208,696,464 RAC: 304 |
cc: Claimed Credit Initally i thought so too, but then he mentions minutes. Grant Darwin NT |
ML1 Send message Joined: 25 Nov 01 Posts: 21118 Credit: 7,508,002 RAC: 20 |
(Ofcourse, most PC systems suffer a memory bandwidth bottleneck. Certain systems far more so than others...)Now, now. Don't be nasty towards Intel. Now how did you guess that? ;-) (Not to confuse the Intel "Hyperthread" cludge with the AMD "Hypertransport" (fast) CPU-integrated memory management unit.) Happy crunchin', Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
W-K 666 Send message Joined: 18 May 99 Posts: 19365 Credit: 40,757,560 RAC: 67 |
(Ofcourse, most PC systems suffer a memory bandwidth bottleneck. Certain systems far more so than others...)Now, now. Don't be nasty towards Intel. Now, now, if you use an Intel with2 MB L2 cache you hardly need memory bandwidth. Andy |
Idefix Send message Joined: 7 Sep 99 Posts: 154 Credit: 482,193 RAC: 0 |
Hi, Now, you're talking about "fair across the board"... DO THE MATH. As long as an optimized application is envolved in the math it will become wrong. You must not compare optimized applications which are only used by a minority. The majority uses the standard software. And this software must result in equal granted credits. Regards, Carsten |
W-K 666 Send message Joined: 18 May 99 Posts: 19365 Credit: 40,757,560 RAC: 67 |
Hrm.. An hour and 18 minutes on Einstein on my AMD A64 3000+ Win2KProSP4 machine (on a "Romeo" WU" "Zed"'s take a little longer, but not much)netted around 24 CC and 13.6 GC... SETI Enhanced 5.12 WU's take, on average, FOUR HOURS AND FOURTY-ONE MINUTES on the same machine, on average, and are being CC'ed around 37 minutes and GC'ed around 18 credits, IF I'm lucky Don't know what you're doing on Einstein I claim 10 and average granted over 30 for the longer unit which take 45 mins. And I suggest you look more closely, for Akosf, Read this New Scientist article first posted by ML1. Andy |
ML1 Send message Joined: 25 Nov 01 Posts: 21118 Credit: 7,508,002 RAC: 20 |
(Not to confuse the Intel "Hyperthread" cludge with the AMD "Hypertransport" (fast) CPU-integrated memory management unit.)Now, now, if you use an Intel with2 MB L2 cache you hardly need memory bandwidth. Very true for s@h crunching at present. I'll agree that Intel's "Pentium M" and its follow-on processors ("Enhanced P3" architecture) are a good design as opposed to the Marketing driven "P4 Netburst" silliness... Meanwhile, AMD is still doing neat and cool stuff with their designs... But we digress... This comes back to whether the whole Boinc-s@h & host computer system should be "calibrated" to award "accurate" credit? Or whether the FLOPs counting is "accurate enough". For myself, the credits are too inaccurate to be useful for any scientific analysis into the credits themselves or what they might show. They may be 'good enough' for trends analysis. Otherwise, they are a very good ego counter for certain 'competitors'. Myself, I think Berkeley have hit a very good balance. It's just a shame that a very few hot-heads are screaming over the small fractional numbers just because they can see a momentary change when multiplied up by their 10 machines. Note: It's all relative! And meanwhile, if we all learn a little about science and computer architectures and Marketing scams, then all the better! :-) Happy crunchin', Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
sideband@seti.usa Send message Joined: 19 Jun 99 Posts: 25 Credit: 2,774,864 RAC: 0 |
SETI Enhanced 5.12 WU's take, on average, FOUR HOURS AND FOURTY-ONE MINUTES on the same machine, on average, and are being CC'ed around 37 minutes and GC'ed around 18 credits, IF I'm lucky CC = Claimed Credit GC = Granted Credit 73 de AI8W, Chris Abdico Concussio Fidens Servo Libertas Semper! |
sideband@seti.usa Send message Joined: 19 Jun 99 Posts: 25 Credit: 2,774,864 RAC: 0 |
The big complaint is this: Let's take my AMD A64 3000+ (FSB 240), 512M DDR400, FC4 machine as an example. Under SETI BOINC, 42 minutes netted around 32 credits.. Under SETI Enhanced, 2 hours and 45 minutes net around.. any guesses? That's right.. 32 credits, plus or minus three credits. There is a big disparity in the amount of work being done and the credits being both claimed and granted. This disparity needs to be fixed. THAT is the big complaint. The sooner we realize the resistance/complaint here is not: 1. SETI Enhanced itself. 2. "cheating" 3. AMD vs Intel 4. Windows vs. Linux vs. NetBSD vs. FreeBSD vs. OS X vs. PickYourFavoriteOS ... the sooner we can get down to the problem and FIX IT. 73 de AI8W, Chris Abdico Concussio Fidens Servo Libertas Semper! |
Femue Send message Joined: 6 Feb 01 Posts: 132 Credit: 4,673,738 RAC: 1 |
The big complaint is this: Hi, just that i can follow your arguments, can you please post a link to some sample results that prove this? I just found results like this 320116254 76859834 7 May 2006 23:03:31 UTC 12 May 2006 21:07:49 UTC Over Success Done 3,034.82 6.96 7.90 320116208 76859799 7 May 2006 23:03:31 UTC 12 May 2006 20:12:18 UTC Over Success Done 3,151.71 7.23 12.60 where neither you claimed 32 credits for 42 minutes nor got them granted. But maybe i've just looked in the wrong place, so i just want to understand. Thanks, Femue |
ML1 Send message Joined: 25 Nov 01 Posts: 21118 Credit: 7,508,002 RAC: 20 |
The big complaint is this... Phew! Scorchin' hot blast of superheated gas also!! Your "s@h boinc" claims must be for the "Optimised Apps" (see my sig!). And there is the source of a few spoilt runaway expectations. It also comes to whether we 'reward' people's efforts or whether we 'reward' people's scientific contribution... With the way that the original s@h application was optimised by some very hard work by a small group of dedicated people, the increase in credit claims were getting to be a little silly. Yes, more "science" was being done, but the contributer's efforts remained the same. I agree with Berkeley that the fairest solution is to as accurately as is reasonable to reward contributer's EXPENDED EFFORT. It's up to Berkeley as to how 'valuable' the science is from that. The release of s@h-enhanced deliberately levels the credits for EVERYONE oncemore. And a good thing too. The squealers complaining about losing their x6 advantage is a good example of why the credits needed realigning (that is rationalising). And yes, much more science is now being done due to s@h-enahnced being built on the optimisations put into the previous client version. Very good too and very good for the science. Sorry, you still get only your 32 credits (or whatever, and not x6 that value). All very fair for EVERYONE. Now, can we get back to some quietly efficient crunchin'? Happy crunchin', Martin (Mmmm, perhaps I'd better update my sig! :-) ) See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
W-K 666 Send message Joined: 18 May 99 Posts: 19365 Credit: 40,757,560 RAC: 67 |
The big complaint is this: Right lets get things right Claimed has absolutely nothing to your performance, Granted is what conunts. The average granted on Seti 4.18 was/is about 24 credits. So on my computer ID 688149 at the moment I am using standard app because I also crunch for Beta the app_info file and BOINC get confused. But my performance is; 4.18 6300s granted 24.29 credits = 13.88/hr wuid=78045650 5.12 14,287s granted 56.46 = 14.23/hr wuid=78922770 That is the real calculation to make. I also note you are still pushing the edges of the envelope a bit, didn't you read and take note of Inglesides Note message305280 Andy |
SargeD@SETI.USA Send message Joined: 24 Nov 02 Posts: 957 Credit: 3,848,754 RAC: 0 |
I find this extremely funny, considering the posts by you and your teammates on the BBR board. This is 180 degrees from what you and most of them were saying when you thought Seti.USA was threatening your position. Oh, I forgot, the reduction in credits granted will extend our time to overtake you by a factor of about 3 (or more). Hmmm, that may be a very good reason to flop sides. |
©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.