it's really good for amassing boinc credits

Well, if anyone here wants to discuss the credits, there's a new thread started here at the BOINC forums by someone from Gridcoin, who wants to know what the next gen credit should do and look like.

Good luck.

---

Well, if anyone here wants to discuss the credits, there's a new thread started here at the BOINC forums by someone from Gridcoin, who wants to know what the next gen credit should do and look like.

Good luck.

Better get started on the Essay response a bit later...

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

That will make two essays they have to read, I've just started mine...

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Aller guten Dinge sind drei? All good things come in threes?

But I'll wait for you both ;)

---

A person who won't read has no advantage over one who can't read. (Mark Twain)

Well, if anyone here wants to discuss the credits, there's a new thread started here at the BOINC forums by someone from Gridcoin, who wants to know what the next gen credit should do and look like.

Good luck.

Why not just fix the existing one?
From what I remember, what we had prior to Credit New was pretty good; payment for MB work was roughly on par with AP.

---

Grant
Darwin NT

Well, if anyone here wants to discuss the credits, there's a new thread started here at the BOINC forums by someone from Gridcoin, who wants to know what the next gen credit should do and look like.

Good luck.

Why not just fix the existing one?
From what I remember, what we had prior to Credit New was pretty good; payment for MB work was roughly on par with AP.

"Fixing the Existing one", for the purposes here, is roughly on par with fine tuning and complete re-engineering. It's a question of how much you refine or change completely, depending on Whether functional is good enough, or excellence is warranted. [I'd be pretty happy with mediocre...]

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

"Fixing the Existing one", for the purposes here, is roughly on par with fine tuning and complete re-engineering. It's a question of how much you refine or change completely, depending on Whether functional is good enough, or excellence is warranted. [I'd be pretty happy with mediocre...]

I suspect that would a whole lot easier than redefining what credit is & how it's determined & granted.

If you re-define the what & how of credit then that would be a whole new series of issues to be found & resolved, just to get back to where we are now.

---

Grant
Darwin NT

"Fixing the Existing one", for the purposes here, is roughly on par with fine tuning and complete re-engineering. It's a question of how much you refine or change completely, depending on Whether functional is good enough, or excellence is warranted. [I'd be pretty happy with mediocre...]

I suspect that would a whole lot easier than redefining what credit is & how it's determined & granted.

If you re-define the what & how of credit then that would be a whole new series of issues to be found & resolved, just to get back to where we are now.

I agree. No need to redefine what a cobblestone is, change its value etc, just restore the stated intent. I'm more concerned with the intertwined estimate mechanism, which has several classic engineering instabilities.

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

Just from my perspective my 2 rigs (though change 3570K with Q6600@3GHz, same GPU's) use to pull around 90K RAC under MB V6 only and that dropped to 64K RAC under MB V7. I'm still doing personal tests with MB V8 to see what impact it has.

It would just be nice to have a standard baseline that could be comparable across all MB versions.

I also remember when doing AP's was a penalty. ;-)

Cheers.

... If you re-define the what & how of credit then that would be a whole new series of issues to be found & resolved, just to get back to where we are now.

I agree. No need to redefine what a cobblestone is, or change its value etc, just restore the stated intent. I'm more concerned with the intertwined estimate mechanism, which has several classic engineering instabilities.

(In summary, ramble... ramble... no easy fix. Against my engineering perfectionism, I also agree easiest is to stabilize the present system...)



Rambling:

There are a few good puns defining the cobblestone...

Which are a reminder to check whether we are measuring what is intended...


Some notes that would make for a long essay:

• The cobblestone is a unit of floating-point + integer operations done. Yet, there are other compute aspects that incur a resource cost such as network and storage utilisation;
  
  
• Should other aspects be rewarded such as turn-around time and reliability and availability?
  
  
• There are other projects other than s@h, some of which have minimal computation and for which the cobblestone is a poor measure if only arithmetic operations are counted;
  
  
• There appears to be arbitrary 'other interpreted value' imposed upon the cobblestone unit used by other projects to give a level of 'feel-good' fiddle factor to the cobblestone counts given.
  
  
• And there is cobblestone deflation in CreditNew dependent on what is seen to be the average cobblestone measured machine. What happens to CPU users when that machine is eventually seen to be a vastly more numerical capable GPU?...
  

Hence, should we set up a sort of exchange market of the 'cobblestone' to balance what 'cobblestone reward' should be given for other values outside of merely counting the arithmetic operations executed?...


My own engineering view is that we should directly measure in cobblestones for all aspects that we feel should be somehow rewarded. However, with our present structure that would be a huge diversion of effort from more immediate science.

Hence, the CreditNew automatically rationalising/levelling between projects is a good enough fix... Even if a fragile fix...


And then further, we have great disparity of capability between CPUs, GPUs, GPGPUs, and new hardware soon to come...

What should we really be 'rewarding'?

The electrical power people consume for the project, regardless of efficiency/effectiveness? Or should we reward for providing results?...


OK... That's my lunchtime ramble. No easy fix.

Is there a student online willing to pick up the idea of making the cobbleston NIST-based?... ;-) Having a reliable measure based on reality would be useful to be able to more easily compare hardware performance... :-)


Happy crunchin',
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

Since the introduction of Green Bank data, RAC has fallen circa 20-25%. Anyone else noticed a similar effect ... its almost been a gradual roll down

---

I have seen 20% or more decrease, but that can also be because with all this GBT data to crunch, the AP splitting is far longer between runs.

I don't think it is due to the lack of AP. When they are up I tend to not get many of them anyway so I don't think it has a large bearing on the issue at hand ... could be wrong though ...

---

Somewhere around the 20% fall, while some of that can be attributed to the lack of APs a quick look at the returns for guppi units suggests that they are bout 25% down per hour - The CreditScrew algorithm is working as implemented and NOT coping with new types of work correctly. Even as designed it would be hard pressed to cope it is such a poor algorithm...

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Has this been run up the flag pole to others ????

---

Is the FFTW library distributed with the Stock CPU application updated over the v7 similar variant ? [increased] Transition to AVX support in that reference app alone can account for a ~20%+ drop, since CreditNew will claim >100% efficiency through a design flaw (downscaling everyone else)

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

Is the FFTW library distributed with the Stock CPU application updated over the v7 similar variant ? [increased] Transition to AVX support in that reference app alone can account for a ~20%+ drop, since CreditNew will claim >100% efficiency through a design flaw (downscaling everyone else)

There's been no change in the stock app since January, and the data distribution didn't start until April. If people are being precise in their statements that the credit drop happened when the data arrived (April not January), that can't be the explanation.

Is the FFTW library distributed with the Stock CPU application updated over the v7 similar variant ? [increased] Transition to AVX support in that reference app alone can account for a ~20%+ drop, since CreditNew will claim >100% efficiency through a design flaw (downscaling everyone else)

There's been no change in the stock app since January, and the data distribution didn't start until April. If people are being precise in their statements that the credit drop happened when the data arrived (April not January), that can't be the explanation.

Makes sense. Then a remaining possible explanation is the shift to VLAR and associated parameters, For which the reference CPU applications are already relatively efficient. Added weight in numbers, of returned reference app results can swing the normalisation reference also, just as Arecibo Shorty storms used to swing it the other way.

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

I have a GTX 750 Ti which I run as closely as possible under lab conditions - constantly active with SETI MB tasks. Its RAC today is 12,815.88.

Looking back at the extended stats graph for December (v7), its RAC for the month fluctuated between 12,500 and 13,000 - I don't see a downward trend there.

That's been a constant problem with unravelling the credit conundrum: people notice and report the dips, but remain silent about the humps.