I know the subjects of Credits have been discussed endlessly, but I do feel that the present
system is not good enough. I will not go down the route of many and suggest we follow the
Seti Classic system of one Work Unit (WU) one credit, that has so many flaws in it that its
hardly worth mentioning.

The present system basically works of the Dhystone/Whetstone benchmarks and devided by
the number of processors. This is first basic flaw, as most studies show that the advantage
gained by each extra processor is appoximately 1.8.

The second problem is that the benchmarks chosen only measure the CPU and the ALU and
not the computer system. This is shown in the comparison of basic HT P4'S, 512KByte Cache
and Xeon's with 2MByte of cache memory with the same clock speed.
The Xeon's are much faster because they rely much less on main memory and HDD performance.

Another problem is the casting aside of the bottom and top Claimed Credits and averageing
the remainder, usually its only one result. This ensures that those fortunate enough to be
use fast computers and claim low credits get inflated Awarded Credit (there is a least one
64 processor machine out there claiming less than one credit/WU but been awarded many times that) and those with slow computers, those claiming 50 or more Credits counting themselves lucky if they are awarded 40 credits.

As one of those with a slow computer I could put my WU cache up to 10 days and then choose
to process only those WU's where the awarded, or likely, credit is say greater than 35, and
abort those with low awarded credit. But I don't, I have sons with much faster machines who
also contribute, with their own accounts, but with much faster machines, most of my
computer, no make that all, is made from their cast off's.

If the present system of calculating Credits is to continue then I think that one should get the Claimed credit score. In my mind the penniless student processing units on a P2 slot1 300Mhz machine deserves more credit that someone who has access to many machines, often fast, at his/her place of work and claims it as his/her own farm. Yes Science needs a many

fast machine as possible but don't penalise the enthusiastic participant with one slow machine running limited hours per day.

This method would only require minor adjustments to the program and therefore could be implemented quickly.

I look forward to anyone else's sensible suggestions.

> If the present system of calculating Credits is to continue then I think that
> one should get the Claimed credit score. In my mind the penniless student
> processing units on a P2 slot1 300Mhz machine deserves more credit that
> someone who has access to many machines, often fast, at his/her place of work
> and claims it as his/her own farm. Yes Science needs a many

I think we've already got an "optimized" client that returns a set of configured answers instead of running benchmarks -- it always returns the values the owner set.

Granting the claimed credit instead of the granted credit would open BOINC to general kind of cheating that happened under the original SETI.

As it is, it takes a couple of people working together to really influence the granted score, and that only works if the two people get the same work unit.

Any individual trying to artificially raise their score will generally get tossed out before the scores are averaged.

The solution may be a better benchmark, but benchmarks are incredibly difficult to design.

---

By using the present system and getting the claimed credit instaed of the 'average' uses the present benchmark etc. Therefore the time to complete a unit is approx. known, so presumably the only way to cheat would be to artificially increase the processing time. This would surely be spotted because:

they would return too many units in a given period,

or their score would decrease because they were submitting a reduced number of results

and they could be spotted because the average times to complete a 'normal' units would be either varying wildly or be way outside the normal for their system.

Do you know when the next benchmark run is on your computer? If this interval between benchmark runs is known, would it be very difficult to adjust the program to make the interval random.

> By using the present system and getting the claimed credit instaed of the
> 'average' uses the present benchmark etc. Therefore the time to complete a
> unit is approx. known, so presumably the only way to cheat would be to
> artificially increase the processing time.

Claimed credit is basically processing time * benchmarked speed. Doubling the benchmark speed doubles the claimed credit.

... but the scoring system is based on the idea that any given WU is worth the same on a 386 as it is on a high-end multiple CPU mega-cruncher.

It's the same number of operations on the fastest machines as it is on the slowest.

---

> Yes Science needs a many
> fast machine as possible but don't penalise the enthusiastic participant with
> one slow machine running limited hours per day.
>

I'm not quite sure where you are going with this argument, since most folks feel that if anything, it's the faster machines that are 'penalized' slightly on BOINC. On Classic, every work unit was worth one credit. If a fast machine could crunch more units per day, then it accumulated more credit. Nobody had a problem with this on Classic. Over here on BOINC though, your slower machine gets an advantage in that you receive more credit per work unit than my faster one, for doing the same amount of work:

Your Computer
Average Credit Granted (last 12 results) = 33.43

My Computer
Average Credit Granted (last 12 results) = 27.53

I'm not complaining... i like BOINC a whole lot. I'm just trying to understand how you feel that your machine should get more credit for doing the same work mine does. And by the way, i am a penniless college student, with a low-end HP machine that i picked up at Walmart. :)

Happy Crunching!

Dig

---

> Your
> Computer
> Average Credit Granted (last 12 results) = 33.43
>
> My
> Computer
> Average Credit Granted (last 12 results) = 27.53
>
But looking at your results you are rewarded approx about the sams as you claim, I have always got less claimed 50 - 55 approx. and presumably you process more units per processor than I do.

Andy

P.S. I've just taken a look at my sons account, his average claim, 19 results is 23.04 but he was granted an average of 27.00 for the same 19 units, He has a P4 HT 3.0GHz 1Gig Mem and 880GByte HDD space SATA RAID

If you go to this one

http://setiweb.ssl.berkeley.edu/show_host_detail.php?hostid=639525

You will see a fast computer, For 12 results he claimed a total of 2.82 credits and recieved 283.91 i.e. about 100* more whereas you get approx 1: i.e. you are "average" and I get abot 60% thus slow. And he probably doesn't own that machine.

Someone ought to contact the host's owner.

---

> If you go to this one
>
> http://setiweb.ssl.berkeley.edu/show_host_detail.php?hostid=639525
>
> You will see a fast computer, For 12 results he claimed a total of 2.82
> credits and recieved 283.91 i.e. about 100* more whereas you get approx 1:
> i.e. you are "average" and I get abot 60% thus slow. And he probably doesn't
> own that machine.
>

As long as we can keep this a friendly discussion i guess i'll still participate. LoL. I don't mind a little healthy debate, but i don't want this to turn ugly. :)

I think we should probably leave Mr. 64-CPU's out of our discussion since he's not here to defend himself. Whether or not he actually owns the machine is not really our business. Looking back through his results though, i think i'd be pretty PO'd if each of my processors worked approx. 6.5 hours on a WU and only ended up receiving an average credit of 23.65. He only makes up for it in quantity, which is not what we're talking about here.

I'm still trying to understand where you're coming from, so bear with me here. Let's say that you get what you want, and everybody is actually granted what they claim. If i understand you correctly then, you feel that it would be entirely fair if your machine gets 55 credits per work unit, and mine gets 27 for doing exactly the same amount of work?

The idea behind the benchmarking, etc. if i understand it correctly is to try to end up with everyone getting about the same credit per unit, while at the same time reducing the amount of cheating. In my mind, it isn't a factor of your receiving less credit than you claim. Rather, i feel that your machine shouldn't be claiming so much to begin with, if the benchmarking process were entirely accurate.

Maybe Tony will jump in here and give me the right number, but the average credit per work unit overall has been calculated to be about 32 i believe. You actually receive a little bit above that, and i am currently well below that. That doesn't bother me at all, i'm just still trying to figure out where you are coming from here. :)

Happy Crunching

Dig

---

> I'm still trying to understand where you're coming from, so bear with me here.
> Let's say that you get what you want, and everybody is actually granted what
> they claim. If i understand you correctly then, you feel that it would be
> entirely fair if your machine gets 55 credits per work unit, and mine gets 27
> for doing exactly the same amount of work?
>
> The idea behind the benchmarking, etc. if i understand it correctly is to try
> to end up with everyone getting about the same credit per unit, while at the
> same time reducing the amount of cheating. In my mind, it isn't a factor of
> your receiving less credit than you claim. Rather, i feel that your machine
> shouldn't be claiming so much to begin with, if the benchmarking process were
> entirely accurate.
>
> Maybe Tony will jump in here and give me the right number, but the average
> credit per work unit overall has been calculated to be about 32 i believe.
> You actually receive a little bit above that, and i am currently well below
> that. That doesn't bother me at all, i'm just still trying to figure out
> where you are coming from here. :)
>
> Happy Crunching
>
> Dig
>
To be honest the credit thing doesn't bother me that much, but I recently met a schoolboy who is crunching seti classic and I asked why he hadn't changed over to Boinc. He said his friend had and he felt judging from what he had seen he didn't think he would because:

1. he was doubtful he complete a unit in time (seti classic - 57 hrs average using screen saver - thats how i spotted it) and his Mother only lets him have computer on in the evenings and at weekends. Probably about 40hrs max.

2. His friend said "he was getting much lower credit than he was claiming."

I don't know what proc his friend had but he's got P2 300MHZ and now has 512Mbyte memory and told him that if he ran it without screen saver he would probably be ok. He still hasn't converted he's trying to get to 100 credits on classic, presently in the 80's.

Therefore I do think the credit system has to be fairer if the classic participants are convert and not jump off when it closes.

Andy

> I don't know what proc his friend had but he's got P2 300MHZ and now has
> 512Mbyte memory and told him that if he ran it without screen saver he would
> probably be ok. He still hasn't converted he's trying to get to 100 credits on
> classic, presently in the 80's.
>
> Therefore I do think the credit system has to be fairer if the classic
> participants are convert and not jump off when it closes.
>
> Andy
>

I guess that's what i don't understand... the system IS fairer to slow computers. They are rewarded in fact. He'll get more credit per work unit over here on BOINC than the equivalent he is getting on Classic. He'll move ahead faster than on Classic because he'll get more credit per work unit than the faster machines. :)

Dig

---

>
> I guess that's what i don't understand... the system IS fairer to slow
> computers. They are rewarded in fact. He'll get more credit per work unit
> over here on BOINC than the equivalent he is getting on Classic. He'll move
> ahead faster than on Classic because he'll get more credit per work unit than
> the faster machines. :)
>
> Dig
>
I am afraid I don't understand. I get less credit per unit done than I claim because I have a slower that average processor and can only do a max of 4 per day even if its on 24/7, You get about the same as you claim, my son gets more than he claims (faster than average) and he can do about 8-10 units per day 24/7.

Therefore by donating one computer/day I would be awarded about 133 but claimed 220, and my son would get over 220 but claimed 185. Therefore the system used which is supposed to even it out by the benchmark * time method in fact works in the opposite way.

Andy

> Therefore I do think the credit system has to be fairer if the classic
> participants are convert and not jump off when it closes.

So, if three different machines do exactly the same work, they should each be granted different credit?

---

> I am afraid I don't understand. I get less credit per unit done than I claim
> because I have a slower that average processor and can only do a max of 4 per
> day even if its on 24/7, You get about the same as you claim, my son gets more
> than he claims (faster than average) and he can do about 8-10 units per day
> 24/7.

It isn't faster processors vs. slower processors -- it is just a difference between the predicted speed and the actual speed.

If one computer "benchmarks" slow, then it's going to get more credit than it claims. If another appears faster in the benchmark than it can crunch, then it's going to appear to lose points.

---

> > Therefore I do think the credit system has to be fairer if the classic
> > participants are convert and not jump off when it closes.
>
> So, if three different machines do exactly the same work, they should each be
> granted different credit?
>
I can see where you are coming from i.e. you pay 3 or 4 people the same for eaactly the same work, but I say pay same for using one proceesor/time period * benchmark.

Let me put in another way, the credits = benchmark * time/constanttherefore

if benchmark = 800 and time = 1000 then credit = 800,000/constant
therefore if another computer processes the same unit with the same result but has a benchmark of 1000 then it should do the unit in a time = 800. Unfortunately the present system doesn'y do this it squews the results in favour of the faster machine.

Andy

> I am afraid I don't understand. I get less credit per unit done than I claim
> because I have a slower that average processor and can only do a max of 4 per
> day even if its on 24/7, You get about the same as you claim, my son gets more
> than he claims (faster than average) and he can do about 8-10 units per day
> 24/7.
>
> Therefore by donating one computer/day I would be awarded about 133 but
> claimed 220, and my son would get over 220 but claimed 185. Therefore the
> system used which is supposed to even it out by the benchmark * time method in
> fact works in the opposite way.
>
> Andy
>

Well first of all, you're getting way too hung up on this 'claimed' vs. 'granted' credit thing. You feel that you are losing out here because you are not getting the credit you claim. In actuality, if the system were really unbiased, we'd all be 'claiming' the same amount of credit per workunit anyway. So the first inequity in my mind is that your computer even tries to claim 55 credits to begin with. But that's another story altogether.

In order to make a fair comparison, you can only look at credit granted... the credit per work unit that our two machines are actually earning and getting credited to our accounts. On Classic, you would crunch a WU in so many hours and get one credit. I would crunch the same work unit and also get one credit. That was fair, right? My machine crunched faster so i got more total WU's done, but that has no bearing on the issue. We each got our one credit per work unit, regardless of how long it took us to crunch it.

On BOINC, your machine is currently getting 1.21 credits to every one of mine... or six credits more per work unit. This has nothing to do with claimed vs. granted credits... this is simply looking at our results histories and comparing what your machine is actually earning by comparison to mine. That's what i meant when i said that slower machines tend to do better here than they do on Classic. Whereas on classic it was always a 1:1 ratio, over here you get more credit per work unit than a faster machine.

Dig

---

In actuality, if the system were really
> unbiased, we'd all be 'claiming' the same amount of credit per workunit
> anyway.

There you said it the system is meant to award the same credits for doing exact the same piece of work and it dosen't.

Andy

> There you said it the system is meant to award the same credits for doing
> exact the same piece of work and it dosen't.
>
> Andy
>

You're right ... it doesn't. It gives you more credits than it gives me for doing the same work. :)

Dig

---

The current system is biased in favor of slower and HT computers. The reason for this is if that computer is claiming very high credit (anything over 45 CS) then they are dropped and a higher but more realistic amount is granted. So therefore they might average 35 CS granted per workunit when everyone else is averageing 32 CS granted per workunit.

It is known that the benchmarking/credit system has some problems. However it works reasonably well and the averageing done at most projects smooths out the bumps even more. Don't look for any improvements in this area until other more pressing problems are fixed.

---

BOINC WIKI

BOINCing since 2002/12/8

> > There you said it the system is meant to award the same credits for
> doing
> > exact the same piece of work and it dosen't.
> >
> > Andy
> >
>
> You're right ... it doesn't. It gives you more credits than it gives me for
> doing the same work. :)
>
> Dig
>
Maybe you're right, but the benchmark * time formula should produce the same answer answer no matter what machine it is processed on, and I know it dosen't and until it does I think they should award us what their formula works out for each machine not squew it one way ot the other.

Andy