Does one credit represent one completed workunit, or one hour of crunching or something else? I am thrilled to see the credits rolling in, but (and I'm not complaining here) I seem to be getting more credits than I expected.
</p>

Credit Calculation

Credit Calculation

Actually the credit system is not easy to understand. It takes
a minimum of algebric knowledge to really understand it.

Documentation Menu
Release Notes
BOINC FAQ
BOINC Owner's Manual
SETI@Home Web Site Owner's Manual
SETI FAQ

This 'SPACE' Rented.

The anonymity of the Internet, brings
forth, yet another EXPERT.

M7 Seti@h Berkeley's Staff Friends Club ©

Thanks to Paul D. Buck for the information pages Pascal just posted.
I had a look at it for the first time and it's very well done.

The base principles of the credit system are simple but one who
wants to calculate precisely how much CB will be granted for his
work will need to do some algebric math.

A coputer that bench 567 double-precision MIPS
1339 VAX MIPS and takes 8 Hours and 29 minutes
to complete one work units will be granted
???? CB

The formula been posted some time ago here but I don't
remember it.....


-.-. --.- -.. -..- . - .-.-. -.- --... ...--

Thank you Pascal for posting the links to the manual. A lot of my questions are answered as I browse this info. Obviously, credits are more complex than I orginally thought. It looks like I have some reading to do.

Thanks again!


</p>

> I was under the impression that the granted credit was the average of the
> claimed credit; not the lowest credit of the bunch. Please dont tell me that
> all it takes is someone sluggin' on a 486 to lower the score (and take all 14
> days to do so)?

The claimed credit is the benchmark times the time. This calculation is supposed to be equal no matter which computer does the work. In any case for S@H, three are required, the high value and the low value are thrown out, and the middle is taken as the amount of credit granted.

For Predictor, two values are required, the high is thrown out, and the low is taken.

CPDN does not do validation.

> The claimed credit is the benchmark times the time. This calculation is
> supposed to be equal no matter which computer does the work. In any case for
> S@H, three are required, the high value and the low value are thrown out, and
> the middle is taken as the amount of credit granted.
>
> For Predictor, two values are required, the high is thrown out, and the low is
> taken.
>
> CPDN does not do validation.

But this calculation penalizes fast computers while rewarding slow computers. I may be preaching to the choir here but for the sake of fairness wouldnt it be better if we were granted our own credit?

> But this calculation penalizes fast computers while rewarding slow computers.
> I may be preaching to the choir here but for the sake of fairness wouldnt it
> be better if we were granted our own credit?
>

Misfit,


Uhh... Awarding everyone the amount of credit they themselves claimed would open up a credit-boosting exploit. Having the credit granted be the 'middle value' of the three (on S@H) or the low value of the two (predictor) prevents people from taking this sort of advantage. Due to the sheer number of people participating, the likelyhood of being grouped (in that work unit) with someone else also running an exploit is very small indeed unless virtually everyone did it. And if everyone did it, then it would no longer be an exploit. Besides, back when we could check it, I found myself 'gaining' credits (from being the low person) about as often as I lose credits (from being the high person). It all evens out.


------------
KWSN-MajorKong
KWSN Forum Admin (retired)
http://www.kwsnforum.com

BOINC Beta tester
Member of the 'Magnificent 7'

>
> But this calculation penalizes fast computers while rewarding slow computers.
> I may be preaching to the choir here but for the sake of fairness wouldnt it
> be better if we were granted our own credit?
>

A 486-33 MHz using 1000 hours on a wu and a 486-33 GHz using 1 hour on the same wu should "claim" the same credit. But the 33 GHz will have 1000x more "claimed" credit after 1000 hours. :)
Of course, you'll need longer deadline than currently in seti to expect getting any credit on the slow machine. ;)

The BOINC-benchmark isn't perfect, and even seti uses cpu-time other processes you're running alongside influences the cpu-time a little. Due to this some machines will normally always "claim" a little higher than they really should, while some always a little lower.

But crediting in seti is the middle claimed of the 3 passing validation, so it's already some averaged out. Over time this will be even better, so a computer that crunches 10 wu/day will have roughly 2x the credit as another crunching 5 wu/day.


AFAIK the only project there slow computers has an advantage is in UD's (or whatever they've called now) "think", there the cpu-benchmark is 35% or something of the total, meaning a 1 GHz-machine crunching a wu in 3 hours will get much higher credit than a 3 GHz-machine using 1 hour on the same wu.

BTW, the credit-calculation in BOINC is this:

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