Some toughts about the credit system


(from here: completed WU = successfully completed, similar, valid WU)
1. My contribution to the project are the completed WUs.

2. The more completed WUs the more contribution.

3. The contribution is directly proportional to the completed WUs in a given period.

It is evident: if a machine completes 6 WUs a day and another 9 a day, then the contribution of the latter is 1.5 times bigger.

In other way the contribution must be directly proportional to the computation power (in a given period).
This is the case, because the processing time is inversely proportional to the power of the machine.

4. The contribution must be independent of opsys.

This is the case, because the processing time is independent from the opsys it depends only the power.

5. The reward of my contribution is the credit I received, so contribution should be proportional to the received credit.

6. THE RECEIVED CREDIT IS ONLY THE PROPERTY OF THE WU NOT OF THE MACHINE!!

7. THE (claimed and received) CREDIT FOR A GIVEN WU MUST BE THE SAME FOR EVERYBODY!! (Independently of power and opsys.) These two are the strongest constraints of the credit system.

Therefore:
- You shouldn't bother with the benchmarks. There is no need for benchmarks! Seti is the bechmark!
- You shouldn't bother with claimed and granted credits. Granted credit is equal with claimed credit if result is valid and same as others.

(from here: WU = WU)
8. There shuld be a credit system because there are different WUs and BOINC is a multiproject environment.

9. The more "complicated WU" the more received credit for it.

If I receive 50 credits for a WU with 3 hours completion time, then I expect 100 credits for a WU with 6 hours completion time.

10. I would like to receive the same credit for a given amount of work for different projects.

So, if I work 2 hours for Seti and 2 for Predictor I expect (eg.) 40-40 credits.

That's all.

---

You Guys think way to much, I take what it gives me when it gives it to me and don't bother to worry about it ... Life's to short to worry about something that don't mean Diddle Squat in the Grand Scheme of things and is absolutely worthless other than to bolster ones Over Inflated Ego ... ;)

---

> You Guys think way to much,
Yea they do it hurts my head just reading it

> I take what it gives me when it gives it to me and
> don't bother to worry about it ... Life's to short to worry about something
> that don't mean Diddle Squat in the Grand Scheme of things and is absolutely
> worthless other than to bolster ones Over Inflated Ego ... ;)
Yea, and plus what the chance of you winning to change a credit or even the whole credit system?

---

Yea, and plus what the chance of you winning to change a credit or even the whole credit system?
=========

0 to None by my figuring...Right now a change to the Credit system is listed as Low on the Priority List...

There may be a few minor changes to it as there already has been but I highly doubt that there is going to be any wholesale changes made just because somebody doesn't like the way it is or somebody has some grand scheme on how to change it...IMO

---

Please please please someone make a credit message board...

---

> Please please please someone make a credit message board...
>

Seems to me that 'credit issues' would be a relevant topic on the 'number crunching' board. Looks like the problem is more that this board has been used for lots of other things because boards for those topics either don't exist or, as is the case quite often, are 'question and answer' topics that arrive here due to the separation of those boards from the forums.

> > Please please please someone make a credit message board...
> >
>
> Seems to me that 'credit issues' would be a relevant topic on the 'number
> crunching' board. Looks like the problem is more that this board has been
> used for lots of other things because boards for those topics either don't
> exist or, as is the case quite often, are 'question and answer' topics that
> arrive here due to the separation of those boards from the forums.
>

It doesn't matter if it's on topic or not. It's so much of the same tripe over and over again that it should have its own board.

---

And what about to grant some credit also for posts in forum?

---

ALL GLORY TO THE HYPNOTOAD!
Potrebujete pomoc?
My Stats

Did anyone else get the impression that the first half of the description described the SETI Classic Credit system but the second half described the BOINC system?

---

<a href="http://www.boincstats.com/stats/boinc_user_graph.php?id=877f93559fda9f7c5a65f974a8763090"><img src="http://www.boincstats.com/stats/banner.php?cpid=877f93559fda9f7c5a65f974a8763090"></a>

Most of the (seemingly endless) threads on credit miss one critical point.

BOINC is not SETI. SETI is not BOINC.

For example, it is pretty obvious that the SETI project's credit is skewed by benchmarks. So we could have a wonderful new credit system which said:

Assign credit to WU
Scrap benchmark code
Send out WU
Process returned WU
Once validated as scientifically correct against 2 other WUs, grant credit.

Lovely. Simple. Efficient. Slower machines get less credit, faster ones more. Forget which processor family, Opsys, HT enabled or not........
No cheating possible.

For SETI it would work.

But it wouldn't necessarily work for CPDN, LHC, Predictor or some other projects yet to come, and BOINC is/will be the common platform for them all.

So we can't do it. Endex. Changing the credit system is not an option.

What we MUST do is sort the benchmark code out so that it is more realistic and fair- but perfect it will never be.

---

> And what about to grant some credit also for posts in forum?

Perhaps grant negative-credit for posts in the forums??? [ducking]

---

> > And what about to grant some credit also for posts in forum?
>
> Perhaps grant negative-credit for posts in the forums??? [ducking]
>
Perhaps grant credit for posts in the forums based on rating??

---

BOINC WIKI

> And what about to grant some credit also for posts in forum?
>

Yea, that would be something! Some people have more posts then crunced WUs.
Happy crunching!

---

Happy crunching

If you look here, there is a discussion about average cobblers/WU, started by Petit Soleil.

The theory is that if the system is working, we should all be getting about the same average credit. You take a whole bunch of WUs and add up all the once that were granted more than 0.0 cobblers. Divide. So far, everyone who has reported has got between 32 and 38 cobblers/WU.

Isn't this what you guys want?? isn't it what we're getting? No, every unit is NOT worth exactly the same amount, but the fact that they're all worth ABOUT the same amount is pretty good.

I remember when Predictor was up, I knew that for a "normal" unit I got about 18-20 cobblers. LHC gives (me) about 5-6 for the short units and ~100 for the big ones.

What you guys want is for an estimate BEFORE we crunch the unit... but they don't KNOW before the unit is crunched if it's going to be canned after 2 minutes because of so much interference... but it averages out, which is really the best you can expect in life.
[/rant]

---

I say let's rid ourselves of the credit system entirely and see who sticks around purely for the science. I'd bet Seti would keep 80% of it's users, and those users would be the ones who supply the highest percentage of returned WU's anyway, And with the fewest complaints.

---

>
> What we MUST do is sort the benchmark code out so that it is more realistic
> and fair- but perfect it will never be.
>

Amen.


Regards Hans

---

> If you look <a> href="http://setiweb.ssl.berkeley.edu/forum_thread.php?id=5889">here[/url],
> there is a discussion about average cobblers/WU, started by Petit Soleil.
>
> The theory is that if the system is working, we should all be getting about
> the same average credit. You take a whole bunch of WUs and add up all the
> once that were granted more than 0.0 cobblers. Divide. So far, everyone who
> has reported has got between 32 and 38 cobblers/WU.
>
> Isn't this what you guys want?? isn't it what we're getting? No, every unit
> is NOT worth exactly the same amount, but the fact that they're all worth
> ABOUT the same amount is pretty good.
>
> I remember when Predictor was up, I knew that for a "normal" unit I got about
> 18-20 cobblers. LHC gives (me) about 5-6 for the short units and ~100 for the
> big ones.
>
> What you guys want is for an estimate BEFORE we crunch the unit... but they
> don't KNOW before the unit is crunched if it's going to be canned after 2
> minutes because of so much interference... but it averages out, which is
> really the best you can expect in life.
> [/rant]
>

Yes, the theory of "avarage credit" is fine.
The distribution of claimed credits (defined by any measurement system) on the computer population defines an average claimed credit. This fact and the validation system mean that the more processed WUs the more precise fit to this average. (I could explain this in a more detailed way but it is not important.) This average is moving very slowly by some reasons.

So, finally the two systems (the realized and "the simple" one) are really identical.

Therefore I have NO COMPLAINTS, since I know this.

My SUGGESTION is: if I had two identical system I would use the simpler one.

---

> Yes, the theory of "avarage credit" is fine.
> The distribution of claimed credits (defined by any measurement system) on the
> computer population defines an average claimed credit. This fact and the
> validation system mean that the more processed WUs the more precise fit to
> this average. (I could explain this in a more detailed way but it is not
> important.) This average is moving very slowly by some reasons.
>
Sorry, but what do you mean by "This average is moving very slowly"?? Are you referring to the time taken to validate a unit, or the recent average credit (RAC)?

> So, finally the two systems (the realized and "simple" one) are really
> identical.
>
> Therefore I have NO COMPLAINTS, since I know this.
>
> My SUGGESTION is: if I had two identical system I would use the simpler one.

I agree, they should use the simpler one. But the simpler one of the USER end might not be the simpler one on the PROGRAMMING end.

I guess the question I have for you is about why you chose SETI to be the benchmark? What if someone doesn't ever DO a SETI unit? Why should the other projects be dependant on SETI for their benchmarks?? The whole purpose of BOINC is to make the credits non-project specific.

If all the other systems were to take SETI as the basis, and then the amount of work in a SETI unit doubled, the other systems would have to re-calibrate against the new SETI units... That's not really fair. Or even better, what if SETI (G-d forbid) ran out of money and died? What would be the new benchmark?

I could just as easily say that Predictor or LHC or Pirates should be the basis. it's all sort of aribtrary. Much more aritraty than basing it on your computer's processing speed...

---

> Sorry, but what do you mean by "This average is moving very slowly"?? Are you
> referring to the time taken to validate a unit, or the recent average credit
> (RAC)?

Almost RAC. RAC is for a given user but the mentioned average is (almost) the average of RACs on each users.

> I guess the question I have for you is about why you chose SETI to be the
> benchmark? What if someone doesn't ever DO a SETI unit? Why should the other
> projects be dependant on SETI for their benchmarks?? The whole purpose of
> BOINC is to make the credits non-project specific.
>
> If all the other systems were to take SETI as the basis, and then the amount
> of work in a SETI unit doubled, the other systems would have to re-calibrate
> against the new SETI units... That's not really fair. Or even better, what if
> SETI (G-d forbid) ran out of money and died? What would be the new
> benchmark?
>
> I could just as easily say that Predictor or LHC or Pirates should be the
> basis. it's all sort of aribtrary. Much more aritraty than basing it on
> your computer's processing speed...
>

I'm afraid you misunderstood me.
"- You shouldn't bother with the benchmarks. There is no need for benchmarks! Seti is the bechmark!"
"Seti is the benchmark" doesn't mean that I want Seti to be the benchmark for each project.
I suggets no benchmarks at all!
As I suggested "THE RECEIVED CREDIT IS ONLY THE PROPERTY OF THE WU NOT OF THE MACHINE".

> I remember when Predictor was up, I knew that for a "normal" unit I got
> about 18-20 cobblers. LHC gives (me) about 5-6 for the short units and
> ~100 for the big ones.

You helped me and gave the "average credits" for Predictor and LHC.
Let's give 36 credits for a Seti WU, 20 for a Predictor...
Big crunchers :) know the average credits very well.

---