Message boards :
Number crunching :
Calculation of BOINC Credits ???
Message board moderation
Author | Message |
---|---|
Lionel Send message Joined: 25 Mar 00 Posts: 680 Credit: 563,640,304 RAC: 597 |
On the website it states the following in relation to the calculation of credits: Credit The project's server keeps track of how much work your computer has done; this is called credit. To ensure that credit is granted fairly, most BOINC projects work as follows: a) Each work unit may be sent to several computers. b) When a computer reports a result, it claims a certain amount of credit, based on how much CPU time was used. c) When at least two results have been returned, the server compares them. If the results agree, then users are granted the smaller of the claimed credits. Each project gives you credit for the computations your computers perform for it. BOINC's unit of credit, the Cobblestone 1, is 1/100 day of CPU time on a reference computer that does a) 1,000 double-precision MIPS based on the Whetstone benchmark. b) 1,000 VAX MIPS based on the Dhrystone benchmark. Eventually, credit may reflect network transfer and disk storage as well as computation. How credit is determined When your computer completes a result, BOINC determines an amount of claimed credit in one of two ways: a) In general, the claimed credit is the result's CPU time multiplied by the CPU benchmarks as measured by the BOINC software. NOTE: the BOINC software is not optimized for specific processors. Its benchmark numbers may be lower than those produced by other programs. b) Some applications determine claimed credit themselves, and report it to BOINC. This would be the case, for example, with applications that use graphics coprocessors or other non-CPU hardware. Claimed credit is reported to a project when your computer communicates with its server. The granted credit that you receive may be different from the claimed credit, and there may be a delay of a few hours or days before it is granted. This is because some BOINC projects grant credit only after results have been validated. For SETI at Home: Question 1: What does this really mean and how are the credits actually calculated. Question 2. Given your explanation, how would it apply to 3 hypothetical systems, for example: System A: PII 450MHz, 100Mflops, 12 hours to complete a WU System B: PIII 1GHz, 225Mflops, 6 hours to complete a WU System c: PIV 3.2 GHZ, 450Mflops, 3 hours to complete a WU |
Petit Soleil Send message Joined: 17 Feb 03 Posts: 1497 Credit: 70,934 RAC: 0 |
> Question 1: What does this really mean and how are the credits actually > calculated. Here's how I see it. It takes an algebraic formula to figure it out. But It's tricky ! I've tried but on the formula I've made slower machine had higher credit ! There's got to be a 1/X somewhere but where ? On the benchmarks or on the CPU time ? I really don't know. It's not clear. There must be a math guy among us who could help... Regards -.-. --.- -.. -..- . - .-.-. -.- --... ...-- |
STE\/E Send message Joined: 29 Mar 03 Posts: 1137 Credit: 5,334,063 RAC: 0 |
System c: PIV 3.2 GHZ, 450Mflops, 3 hours to complete a WU ============================== First off a P4 3.2 will score around 2000 MFlops & run a WU in about 1:30 to 1:45 time frame in single CPU mode. It will run 2 Wu's in about anywheres from 3:00 hrs to 3:45 hrs & mins in HT mode (ie running 2 at once, 1 on each side of the CPU). At least thats what my P4 3.2 will run them in. My P4 3.06 is just a little slower than the P4 3.2 but for some reason always seems to gain more Credit than the P4 3.2...??? I really don't have a clue as to how the Credits are calculated nor personally care to give myself a sevear migraine headache trying to figure all that out. I'll leave all that up to the Rocket Scientist's in the forum, because when it all comes down to it you can pretty much throw all that out anyway with all the variables involved with how people benchmark their systems. I've turned in Wu's that Claimed say 45 Credits and I have received as little as .01 Credits and as high as 55 Credits...Go figure, like I said, your at the mercy on how the other 2 people that turned in the result for that WU on how much your going to get. I don't care how often the Server has you Benchmark your PC, if your getting low Benchmark Scores continually & don't do anything to increase them then thats all your ever going to get is low Benchmark scores. The sad thing about it is most people don't even realize that they are getting Low Scores in the first place, so they never do anything about it unless somebody points it out to them & then if they are not Computer Savvy enough to know what to do about it they never will increase their scores, so they will not only short change themselves on Credits but also shortchange the other 2 people that turned in the same WU. That in a nutshell is the whole pitfall to a credit system like this, where 3 people are reliant on each other to properly benchmark their systems so everybody's getting a fair shake with the credit system. |
Rattledagger Send message Joined: 22 Jan 00 Posts: 23 Credit: 1,010,202 RAC: 0 |
> For SETI at Home: > > Question 1: What does this really mean and how are the credits actually > calculated. > > Question 2. Given your explanation, how would it apply to 3 hypothetical > systems, for example: > System A: PII 450MHz, 100Mflops, 12 hours to complete a WU > System B: PIII 1GHz, 225Mflops, 6 hours to complete a WU > System c: PIV 3.2 GHZ, 450Mflops, 3 hours to complete a WU > > cobblestone_factor is currently 100 host.credit_per_cpu_sec = (fabs(host.p_fpops)/1e9 + fabs(host.p_iops)/1e9) * cobblestone_factor / (2 * SECONDS_PER_DAY) claimed_credit = cpu_time * host.credit_per_cpu_sec You haven't actually give enough flops-info, but if takes fpops & iops to be the same as mentioned this gives: a = 5 b = c = 5.625 Seti currently needs 3 for validation, and granted in this case is remove highest & lowest claimed and average the rest, so in most case the middle. ;) If a, b & c is the 3 first returned and all passes validation, all will therefore get 5.625 as credit. Poorboy, you're talking about beta there many was running debug-versions and for most time the wu had min_quorum 2, and in that case the granted credit is the lowest claimed. The current system isn't perfect, but it's a lot better than "classic". ;) Oh, and hopefully v4 has fixed the bug there it's only on client-restart the benchmark is re-run then 5 days is expired... |
Petit Soleil Send message Joined: 17 Feb 03 Posts: 1497 Credit: 70,934 RAC: 0 |
> cobblestone_factor is currently 100 > > host.credit_per_cpu_sec = (fabs(host.p_fpops)/1e9 + fabs(host.p_iops)/1e9) > * cobblestone_factor / (2 * SECONDS_PER_DAY) > > claimed_credit = cpu_time * host.credit_per_cpu_sec > We found the math guy I was talking about ! Thanks According to this a 2 times faster machine would not necessarily have a 2 times higher credit. Can we assume here that slowest PC has some "advantage" over faster machine as for credits ? What do you think ? Regards Marc -.-. --.- -.. -..- . - .-.-. -.- --... ...-- |
ric Send message Joined: 16 Jun 03 Posts: 482 Credit: 666,047 RAC: 0 |
|
STE\/E Send message Joined: 29 Mar 03 Posts: 1137 Credit: 5,334,063 RAC: 0 |
Poorboy, you're talking about beta there many was running debug-versions and for most time the wu had min_quorum 2, and in that case the granted credit is the lowest claimed. The current system isn't perfect, but it's a lot better than "classic". ;) Oh, and hopefully v4 has fixed the bug there it's only on client-restart the benchmark is re-run then 5 days is expired... ----------- Yes, I probably was thinking of beta & with 3 rusults needed now it should be somewhat better. But since we have never been able to get into our results or pending credits almost from day 1 this site opened we don't really know what we are getting for credit now do we...We could be still only getting very little credit on some and a lot of credits on others... |
WiKyD Send message Joined: 12 Oct 00 Posts: 22 Credit: 453,192 RAC: 4 |
Ok...call me stupid...but can this be explained in English ? (ok I really mean in laymen's terms for us slower non-rocket science guys) I mean Set@H1 waws much easier to figure out...you analyze 1 unit, you see one unit on your "total". why this change for Set@H2 and please help me understand how to figure it out... thanks... |
Paul D. Buck Send message Joined: 19 Jul 00 Posts: 3898 Credit: 1,158,042 RAC: 0 |
> Ok...call me stupid...but can this be explained in English ? (ok I really mean > in laymen's terms for us slower non-rocket science guys) > > I mean Set@H1 waws much easier to figure out...you analyze 1 unit, you see one > unit on your "total". why this change for Set@H2 and please help me > understand how to figure it out... The probem is that it is not that simple to explain. I have quite a bit of words on the subject and I am not sure that *I* understand it still. I know that I have not been able to follow the arguments about the faster / slower machines and the benchmark is higher / lower and therefor does "x" to the credit. ANYWAY, you can look up what I do know here: Documentation Menu (pdb) Release Notes = |
Ingleside Send message Joined: 4 Feb 03 Posts: 1546 Credit: 15,832,022 RAC: 13 |
A 486-33 MHz using 1000 hours on a wu will get the same claimed credit as a 486-33 GHz using 1 hour on the same wu. Since 2 different 2 GHz-machines can have huge differences in crunch-times, due to core-differences, memory-differences, multiplier-differences, OS-differences and other processes running at the same time. Since the BOINC-benchmark is synthetic it doesn't catch all of these differences, so the 2GHz using 4 h on a wu and another 2GHz using 3 h on the same wu isn't giving the exact same claimed-credit as it should. This is fixable by BOINC having the ability to supply project-specific benchmarks, or make the synthetic benchmark better. But by letting the middle of 3 results decide the credit they're filtering out much of the variance, and over time more/less than claimed will for most average out so there isn't a huge problem. The bottom line is, just like in "classic", the 2 GHz crunching 6 wu/day will get less "cobblestones" than the 2 GHz crunching 8 wu/day. BTW, the cobblestone-factor has always been the same for different cpu's, but was changed a long time ago in beta to 300 to among other things give more comparable numbers to folding@home, and decreased to 100 then the debug-code was removed from the windows-clients since this gave a huge benchmark-jump. ;) |
ric Send message Joined: 16 Jun 03 Posts: 482 Credit: 666,047 RAC: 0 |
> A 486-33 MHz using 1000 hours on a wu will get the same claimed credit as a > 486-33 GHz using 1 hour on the same wu. tks for posting it's only on theory, 1000h (/24)= 41 day. there is the 14 day delay.. the boinc benchmark could be better, on the same machine, running the benchmark several times, can give different results, you can se this on your on machines. inviting to take a think: due the boinc doku, the reference machine makes 300 cobblestones in 24 hours ric update 100! u'right. well, the above pictured tool is running with the new (100) factor. |
Ingleside Send message Joined: 4 Feb 03 Posts: 1546 Credit: 15,832,022 RAC: 13 |
> > A 486-33 MHz using 1000 hours on a wu will get the same claimed credit as > a > > 486-33 GHz using 1 hour on the same wu. > > tks for posting > > it's only on theory, 1000h (/24)= 41 day. there is the 14 day delay.. Uhm, being running CPDN the last days, so seeing 1000 h to finish is normal. ;) > > the boinc benchmark could be better, on the same machine, running the > benchmark several times, can give different results, > you can se this on your on machines. > Yes, but fpops atleast is unexpectedly stable, with less than 1% difference between normal runs. iops on the other hand is all over the place, but AFAIK this seems to be due to dual-machine. If affinity-locks to only one cpu these numbers is also very stable. Don't remember the exact difference then fooled around with priorities, but it was atleast less than the single difference of going from v3.03 to v3.08 of 10%, so then throwing AR, v3.00 & wu terminating after 1 minute into the mix the new credit-system is much better than the old "1wu = 1 credit" even the benchmark isn't completely stable between runs. If v4 re-benchmarks every 5 days regardless of re-starting client or not, should normally not have the problem of benchmark running on boot either. ;) > inviting to take a think: > > due the boinc doku, the reference machine makes 300 cobblestones in 24 hours > Look again, it's 100 cobblestones. ;) The text was updates after changing benchmarks, and they've removed their small "billion" from the documentation. ;) http://boinc.berkeley.edu/credit.php |
not in use Send message Joined: 22 Jun 01 Posts: 32 Credit: 1,554 RAC: 0 |
From http://boinc.berkeley.edu/credit.php >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>> Each project gives you credit for the computations your computers perform for it. BOINC's unit of credit, the Cobblestone 1, is 1/100 day of CPU time on a reference computer that does 1,000 double-precision MIPS based on the Whetstone benchmark. 1,000 VAX MIPS based on the Dhrystone benchmark. Eventually, credit may reflect network transfer and disk storage as well as computation. >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>> |
Paul D. Buck Send message Joined: 19 Jul 00 Posts: 3898 Credit: 1,158,042 RAC: 0 |
> the boinc benchmark could be better, on the same machine, running the > benchmark several times, can give different results, > you can se this on your on machines. During the Beta we suggested a "running-average" for the benchmark results and this may come in time when the I/O problems become less of a factor. Then with repeated runs you will converge on a number that is more accurate for your system. I agree that the use of the select the middle also will remove much of the variance from the equation. As others have stated, this is fundamentally a better system that now allows for cross-project totalization, which I have noticed that the statistic sites are now doing ... |
ric Send message Joined: 16 Jun 03 Posts: 482 Credit: 666,047 RAC: 0 |
|
STE\/E Send message Joined: 29 Mar 03 Posts: 1137 Credit: 5,334,063 RAC: 0 |
The only Calculation you need to know about BOINC Seti is this: You run 1000 Wu's=0 Credit...You run 2000 Wu's & it still =0 Credits...Simple isn't it ;) Switch to CPDN, you actually get real live Credits there... :) |
belgix Send message Joined: 7 Apr 00 Posts: 40 Credit: 1,951,701 RAC: 0 |
> The only Calculation you need to know about BOINC Seti is this: You run 1000 > Wu's=0 Credit...You run 2000 Wu's & it still =0 Credits...Simple isn't it > ;) Yep, basic rule in maths : 0 x anything = 0 |
Petit Soleil Send message Joined: 17 Feb 03 Posts: 1497 Credit: 70,934 RAC: 0 |
> Switch to CPDN, you actually get real live Credits there... :) > I am trying to find some informations on how CP credits works. I know a WU takes weeks to process. I suppose you get credits at completion of it. How it works ? is it 1 point per units ? Thanks Marc |
ric Send message Joined: 16 Jun 03 Posts: 482 Credit: 666,047 RAC: 0 |
oh yes your probaly right both, helped to sort out. Still trying to understand, if 4 clients making 1000 x 0 WUs, are they faster as 8 clients doing 500 x 0 WUs?, the only winner is the provider of power/electricity Poorboy! the way to CPDN is comming closer and closer... ;-) if there is a forum for signature test > Yep, basic rule in maths : 0 x anything = 0 > > |
STE\/E Send message Joined: 29 Mar 03 Posts: 1137 Credit: 5,334,063 RAC: 0 |
The Credits at CPDN are Granted on a Trickle basis. Depending on your CPU speed will determine how often you recieve Credits. My P4 3.2 CPU running in HT Mode gets about 76 Credits (76 x 2) for each WU every 8 hr's or so... A slower CPU will take longer to get Credits, I thought I seen a post by somebody with a 1.7 Ghz Cpu saying they got credits every 14 hr's... |
©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.