The new credit system gives far too high values for the WUs
11 Jul 2006 18:16:39 UTC
Topic 191560
(moderation:
Per definition in BOINC everything that comes straight from the projects should get the same amount of credits for the same amount of CPU-time on the same setup.
Here in Einstein not even the WUs from the project itself (S4 and S5) follow the rules: I claimed (and got) 4 times the credit that my puter deserved for S5, while S4 fits smugly in the whole system.
I think it's a glitch in the new credit calculating algorithm that only needs to be (and easily can be) adjusted.
Grüße vom Sänger
The new credit system gives far too high values for the WUs
)
Saenger,
you might want to read at least some of the topics posted throughout the last weeks. The short answer is: S5 WUs are 4 times bigger than S4 WUs, require 4 times or more of the time to get calculated and therefore also get 4 times the credit.
Well, compared to other
)
Well, compared to other projects, the FpOps counting here results in quite high credits per hour. It isn't 4 times as much as it should be, it's more like maybe 25% too high.
Of course, if the S5 client runs on a CPU that can use special commands (SSE), the per-hour-value is higher.
Comparing the SETI stock client to Einstein is not valid as SETI claimes different per-hour-values dependant on the angle range. With the optimized(!) SETI Enhanced client, some SETI WUs claim as much as Einstein WUs (per hour), others claim only about half of that per hour.
Considering that the Einstein application is optimized for SSE now, I think the claimed values are not so far off.
RE: Saenger, you might
)
Wrong!!!
Every WU, regardless of it's name, should claim (and get) the same amount of credits per hour of CPU-use un a given puter-setup. A bigger (i.e. longer) WU should get more in the same proportion as it takes longer to crunch.
Credit per WU usually should be something, that's a bit floating, depending on the kind of science (and it's predictability), but credit per hour should stay forever the same (until Moores law puts the machine to rest;)
Grüße vom Sänger
RE: Wrong!!! Every WU,
)
I don't think so!
Lets do a look into BOINC-Wiki (Topic 'Computation of Credit'):
The new Einstein applications does 4 times the FpOps per hour than the old one. I also think it is fair to grant 4 times the credits even it is no longer in agreement with the Wiki definition.
But E@H is in full agreement with the topic 'Credit' in the BOINC-Wiki:
4 times the FpOps therefore 4 times the credit.
Udo
RE: Well, compared to other
)
What you're saying is (as I read it), that the old benchmarking system didn't care about special commands, and thus generated "wrong" benchmarks for systems with it and WUs capable of using them. So a puter could get the same benchmark as some other although he could crunch less powerful (and I don't talk about "optimised" benchmarks). That seems to be fair, and if so, the BOINC benchmarking probably has to be adjusted.
But regarding different angel ranges: That's imho not fair. Same project, same science, same puter, same app, same client, same OS, same SSE, same HD,... this should lead to the same amount of credit claimed per hour (and in an ideal world as well granted;). I "donate" my CPU time, and whether the projects have a good or not that good programmer, whether the app uses more additions or more sinuses, whether the WUs last 10min or 100h, my donation is always the same from my pov.
If some factor for the quality of the developers is to be introduced, who will decide, what's better and deserves more? Why it should project A get more per hour than project B? What kind of operation should get more credits, and again: why?
Grüße vom Sänger
RE: RE: General A term
)
I donate CPU-time.
I get some programms from the projects that do some operations with bits and bytes.
This operations differ from project to project, as does the projects interpretation of this operations in the real world.
From my pov as a donator of CPU-time, it's all just 1s and 0s, that use my CPU (and my electricity;) and I have to be "rewarded" for that.
If I have a faster puter, I'll get more work done per hour, so I get more reward per hour, but I don't have to and don't want to care type of programming of the app, it's all just the same for me.
Grüße vom Sänger
RE: ... But regarding
)
It isn't fair but it is a fact.
RE: RE: ... But regarding
)
So it has to be changed!
But that's OT here and belongs on the Seti-boards.
Grüße vom Sänger
The mystery has been solved
)
The mystery has been solved :
Saenger is talking about Linux - and there the S5 client is one of the very few project applications that gives Linux about the same credits per hour as windows.
So he is right saying that S5 gives nearly 4 times as many credits, but it's not a bug, it's just fair treatment of the Linux platform.
who on earth really cares
)
who on earth really cares about any credits - it will not count for anyrhing concrete like uni credits - the object is surely only about helping the Project number crunch their data.
May I suggest that all users load BOINC as a service and walk.