Strange Credit Allocation System (SCAS)
6 Jan 2010 1:37:42 UTC
Topic 194705
(moderation:
In looking over the results for my last two WUs I've noticed some things I can not explain. First, the data:
WU ID 64804744
Type Hierarchical S-5 all-sky GW search #6
Elapsed Time 34.80 hr
CPU Time 35.34 hr
Claimed 59.57
Granted 199.57
Cr/hr 5.73
WU ID 65300886
Type ABP1
Elapsed Time 32.54 hr
CPU Time 30.59 hr
Claimed 298.30
Granted 250.00
Cr/hr 7.68
The first thing that caught my eye was that for the GW search both the claimed credit and the granted credit are lower than they are for the ABP1, even though the GW search took longer to process, in terms of both elapsed time and CPU time. The second thing that caught my eye is that for the GW search the granted credit is enormously greater than the claimed credit, while for the ABP1 the granted credit is lower than the claimed credit. The third thing that caught my eye is that for the GW search the reported elapsed time is less than the reported CPU time. And the forth thing that caught my eye is that the credit per hour is different for the two WUs. I'm reasonably certain that granted credit bears no relationship to elapsed time (or CPU time), so the forth item, while interesting, is included only to round out the SCAS. And since my old computer can't compete in the credit world, and I crunch for science, I bring the first three items to your attention merely because I think things should be done for logical reasons. Call me Spock if you want, but even the revered Albert said, "God doesn't play dice." Of course he wasn't talking about credits, he was talking, I believe, about his disdain for quantum mechanics. Maybe some of you software developers/programers know of reasons, in terms of the code for the two different applications, why these apparent inconsistancies exist. But I do not know. I'm just asking for explanations. Admins, the microphone is yours.
Cheers, Steve
Strange Credit Allocation System (SCAS)
)
Hi!
ABP1 is almost over now, to be replaced by ABP2. It is assumed that an ABP2 result will take only 15th the processing time of an ABP1 result, so the credit per result was reduced to 50. It will be interesting to see whether the 1/5 runtime assumption holds across different hardware.
The credit for the ABP1/2 searches might be a bit on the generous end of the spectrum, see it as an incentive not to opt out of the search :-).
CU
Bikeman
RE: Hi! ABP1 is almost
)
Hi Bikeman,
You didn't answer all my questions, but I appreciate the fact that you did reply. I crunch for science, and I look forward to ABP2.
Cheers, Steve
So I'll try to answer some
)
So I'll try to answer some more stuff :-)
Claimed credit: This was never (as far as I can remember) used by Einstein@Home to calculate the granted credits. It is simply ignored and there is no real reason to include it in the web stats. Anyway this mechanisms is regarded by many as mostly obsolete now, and it's also quite open to manipulation and "cheating".
The only reason to use claimed credit based credit granting is IMHO when the computational effort for the results cannot be estimated very well in advance (e.g. if there is some unpredictable data dependency of the runtime). This is not the case for ABP1/2. For S5R6, the predictability of runtime is not as good as it was for S5R5 but still in the long run (over several results) it all averages out nicely.
Elapsed time < CPU time
Hmmm...where did you see the elapsed time? S5R6 tasks use only one core at a time, so you are right, this looks strange. Where did you see the elapsed time? In the Boinc Manager GUI?
CU
Bikeman
Elapsed time is less than CPU
)
Elapsed time is less than CPU time in AQUA, which is multithreading.
Tullio
RE: So I'll try to answer
)
Back in s4 granted credit was the average of the first two WU's submitted.
RE: Claimed credit: This
)
There was shortly after I joined The Time Of Optimised BOINC Clients .... :-) :-)
These home-brewed variants had a number of, well, contentious features such that one could in effect dial-your-own credit. Contentious is an understatement actually. Civil war? Anyway server side awarding was instituted here at E@H, and elsewhere, to repel the runaway credit currency rot ( and other issues ) that was risked by that.
As for credit generally there is more or less an agreement amongst the BOINC project community of the idea of a standard loaf of bread ( which like most standards probably doesn't suit all ) - so a sort of mechanism to equate value/effort/contribution/whatever the resources committed to the various projects. I'm not by any means up on the latest as regards this .... as I crunch for E@H solely, my only interest is from a mod's perspective .... I'd certainly welcome a refresh from anyone who has better current knowledge of this aspect.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
RE: Elapsed time < CPU time
)
Yes. The elapsed time was from the BOINC Manager GUI, and my CPU only has one core.
Cheers, Steve