So, we don't get credits for the time we crunch but for the decoys?

right! That was introduced with the new credit system to avoid cheaters.

So its luck if you get 4,5 odr 9 credits?
Well, that reduces the fun of races a lot...
Aren't there other possibilities to avoid cheaters?

Yes, it's a mixture of luck and the CPU/OS your using (some CPUs work more efficient on some proteins), the credits/decoy is averaged over the other users crunching the same protein.

There are other ways, like fixed credit/unit, but since every protein is different, the rosetta staff would have to fix credits for every proetin they ever run individually, which is a lot of wasted work.

This credit system builds a good bridge between avoiding cheaters and granting the right amount of credit for the science done (!!!) not the time spent.

The downside is that if a lot of people with high specs machines crunch the first units of a batch, the slower computers that follow will get less credits/decoy because they're not able to create as many decoys/hour as the others.

---

...
The downside is that if a lot of people with high specs machines crunch the first units of a batch, the slower computers that follow will get less credits/decoy because they're not able to create as many decoys/hour as the others.

It may be that the system is biased to do that. My four-core mega-cruncher will pull down four times as many WUs as my one-core wimpy cruncher.

---

dag
--Finding aliens is cool, but understanding the structure of proteins is useful.

PLEASE NOTE: High spec machines, in and of themselves, do NOT have an impact on credit per model granted to other users. What WOULD have an impact is a machine type where the performance predicted by the BOINC benchmarks (which require no use of L2 cache for example) differs significantly from the actual Rosetta work that machine is able to produce (Rosetta work runs well on machines with a nice sized L2 cache).

---

Rosetta Moderator: Mod.Sense

I'm sorry to disagree with you, but in the real world I have seen my credits for my 4 AMD Mobile 512k barton core crunchers drop 50 ppd each since Dec. I was told it would even out. It did not. Something changed, and did not return to the way it was before.

3 of them have been pulled, parted out and sold, and the other is just crunching until the new owner picks it up.

These where all water cooled NF motherboards running at 2585 to 2700mhz. These benched out between a FX 53-55. Before Dec. they where getting over 320 ppd each. As of late they where getting around 275 ppd each. These same 4 rigs where getting 375 ppd average on Predictor before it shut down.

I don't care what anyone says, I saw my overall RAC drop by 200 ppd in the last 4-5 months. It did not even out. There was a thread about this and ASTRO's spreadsheets proved that the credit had indeed fallen. This is now one of the lowest average credit projects in Boinc.

They are now replaced by 3 x2 Operton rigs. And if you go by benchmark scores, they are still getting short changed. Is it a coincidence that this happed at the same time as the Core2 Duo's came out? Highly unlikely. I believe that the new credit system is slanted to Intel. Thats fine, there are more Intel's out there. But if you want to keep the AMD crowd around, I think someone should look at the way credit is awarded.

Why not be one of the higher average credit projects? A small tweak of the credit system, and bingo. Lots of new users that are looking for nothing but credit. The project benefits as a whole. And cross project pairity is still kept intact.

And to all that reply and say nothing has changed for them, good for you. But I saw my credits fall by 100 ppd per rig from Predictor to this project. 50 when I moved here and 50 in the last 4-5 months.

Now these are my observations of my 4 rigs that ran 24/7. If you were not affected, good for you!

---

BETA = Bahhh

Way too many errors, killing both the credit & RAC.

And I still think the (New and Improved) credit system is not ready for prime time...

I also have a Turion x64 mobile CPU which isn't competitive anymore, it's getting far less credits granted than claimed. (other projects running on that computer don't show that drop in credits.)

To get a normal claimed/granted ratio it seems like you need to follow the trend of buying new CPUs every year or every few months.

---

i'm running a "old" amd 2800xp and my average takes a wild roller coaster ride all the time. before the fold and dock work came there were some work units where i took a hit of -2 points per work unit. now with the fold and dock i am climbing back up. so i don't need a mega cruncher to get my average to climb. just need work units that fit my cpu better than others.

---

Stevea, I cannot dispute your experience with RAC over time. But I point out that you cannot have enough information to assert what caused that RAC change. Overall, according to these cross project stats comparing projects on the same machines, Rosetta credit per CPU second is comparable to other projects.

And it's not that Rosetta's Project Team has intentionally written any code to favor one system or CPU over another. In fact significant portions of the code have been around longer then any of these CPUs. What you are observing is that the advanced architectures of newer CPUs do actually have some performance advantages in doing work more efficiently.

The BOINC benchmarks are not complex enough to reflect all of these architectural advantages. This is why there is often a disparity between claimed and granted credit. And it is often a very consistent ratio for a given machine. Yes, there are exceptions, and specific tasks that behave differently, and specific models you stike upon from time to time that take significantly longer then the average for a task type. But, you can see the chart, it does even out.

---

Rosetta Moderator: Mod.Sense

it could be argued that the variations in the credit granted actually show how effective it is. The variation appears to be due to how well WUs fit certain architectures - possibly largely cache dependant, although it seems there are other factors involved.

I guess rather than changing the benchmark, the logical thing to do would be to allocate the WUs to CPUs more effectively, which would increase the project's productivity.

It isn't a difficult thing to do in theory - initially run WUs on all types of machine and then compare how effective the different CPUs have been. Then rank the WU types against each CPU type and allocate future jobs based on that.

how easy that would be to implement is another matter! It could tie in with allocating by amount of RAM too though. I guess the ideal setup is to have WUs allocated dynamically dependant upon amount of available RAM and CPU type...

---

This will probably be my last post on this subject. I moved to QMC less than two weeks ago and my RAC is already at 267 and clinbing. Here my RAC fell from 235+ to barely 200 with the same machine.

If there is a larger L2 cache requirement for many of the new WU's, then that would explain why the 1MB L2 cache machines have experienced this drop.

As to Intel vs. AMD, my machine is an Intel P4 running at 2.8GHZ, 1MB L2 Cache with 1GB 333MHz Dual Channel DDR ram.

Stevea, I cannot dispute your experience with RAC over time. But I point out that you cannot have enough information to assert what caused that RAC change. Overall, according to these cross project stats comparing projects on the same machines, Rosetta credit per CPU second is comparable to other projects.

It was not my intention to assert that I had enough information as to what caused this recent drop of credit. Only that I have seen my RAC drop, as well as others.

I stated it was around the time that the core2 duo's and quads came out that the credit dropped. Is it a factor? I do not know. I know that the time frames match up, and offered it as a possible cause. I remember Who? joking that it was all his fault. I never said it was, just that the time frames matched.

My point was if you do not want people like Keith and I to feel short changed, than something needs to change. It takes a long time for someone getting 235 ppd to accumulate 70,000 + credits. He has proved what I said, that this project is low on the average credit given. Even your link states the same thing. There are many other projects out there that offer a lot more credit per second than here.

And I ask again. Instead of being in the bottom of the pack as far as credit awarded, why not adjust the system and be one of the top awarding projects. Is there a Boinc rule that says you cannot be in the top 10%? Or even the top project for credit allowed?

You and the project managers may not care that a single Keith left, because he saw his credit drop without explanation. You just had 5000 new members join in a large team. But if nothing changes and 5000 other Keith's leave, what did you gain?

I was just trying to say that it would be nice if 5000 new members joined, without another 5000 leaving. If it takes a look at how credit is awarded, and adjusted to be able to keep the 5000, now would be a good time to do it. And how would that be a bad thing?

I will only crunch disease related projects for personal reasons, but many others may not be so tunnel visioned in their reasons. That being said, it does not mean that I will crunch here for 0 credits, or if I feel that I am being short changed. Basic human nature. There are other disease related project out there. If I was getting the same 375 ppd on the boxes that where pulled from here, 2 of them would have stayed running, and I would have had 5 boxes crunching instead of 3.

If the system remains the same, and no logical explanation is given as to why this has happened, or no changes will be made. Who knows, I could be the next Keith...


And Keith, I in no way intended to offend you. Your post just happened to be here at the right time. If I did offened you in any way I oppologise.

---

BETA = Bahhh

Way too many errors, killing both the credit & RAC.

And I still think the (New and Improved) credit system is not ready for prime time...

... according to these cross project stats comparing projects on the same machines, Rosetta credit per CPU second is comparable to other projects.

According to my stats Rosetta gives pretty low credits. At least with low-performance CPUs (Duron, Sempron, Athlon XP). Granted credits are about 50 to 80% of the claimed credits. Modern CPUs (P4, A64, X2) give at least equal credits, compared to claimed (in fact it's slightly more). Rosetta claims to not depend on SSE2, but interestingly all SSE2 CPUs perform pretty well, SSE1 CPUs don't. Atleast with my CPUs :)

I guess it's as you suggest, a matter of personal choice.

I am happy with what Rosetta is trying to accomplish and I, personally - not speaking for anyone else, would crunch for 0 credits per day.

The "value" I derive from credits is to help me determine if there are any "problems" with the machine I am crunching on (i.e. something stealing cpu cycles, etc).

In the "real world", 100k of Boinc credits and $2 gets you on the New York City subway.

Just my $0.02

That being said, it does not mean that I will crunch here for 0 credits, of if I feel that I am being short changed. Basic human nature.

Well in the real world it just cost them 2 rigs. 3 if you include Keith's.

Right around 800 ppd.


800 x 365 = 292,000

---

BETA = Bahhh

Way too many errors, killing both the credit & RAC.

And I still think the (New and Improved) credit system is not ready for prime time...

my barton-core Athlon-XPs (512Kb cache) get about the same as they claim, which is about the same as a Venice core Athlon-64 (also 512Kb cache):

Barton
Barton
Barton
Venice

---

It dosen't "cost" Rosetta anything. Rosetta is a volunteer project. You are free to dontate to this, or any other project you choose, or not.

Everyone's contribution is welcome. I would like you to stay.

But if for your own personal reasons being granted more credits at another project is more important, you certainly are free to do so.

Well in the real world it just cost them 2 rigs. 3 if you include Keith's.

What would your suggestions be to handle "credit inflation"? If a project started offering credits at the "top of the pack" rate, how soon do you think it would be before other projects did the same in an attempt to re-gain lost crunchers?

And I ask again. Instead of being in the bottom of the pack as far as credit awarded, why not adjust the system and be one of the top awarding projects.

If Rosetta became "top of the pack" and then a hypothetical (disease-related) project started offering you 1,000,000 credits per day for your rig, would immediately leave Rosetta?

Where on the slippery-slope does it end? 1M credits would so distort the concept of credit-parity as to make any comparisons meaningless.

I will only crunch disease related projects for personal reasons, but many others may not be so tunnel visioned in their reasons.

Again, I suppose it comes down to personal choice. One cruncher might be happy even with 0 credits a day, and another may not be happy unless they're crunching for a "top of the pack" project, regardless of credit inflation and credit-parity.

Implying that users are somehow being pety or expecting eveness of credits over time is somehow expecting too much I think might be interpreted the wrong way by both individuals and teams who come to any project to compete.

SETI@Home established the standard for all BOINC projects by being the first to conceive and implement DC/Grid Computing. SETI saw fit to establish a fair credit system to reflect as honestly as possible a person's/machine's contribution to the project.

I think the second statement really reflects what the credit system concept is really all about.

I still beleive that Rosie is a worthy project, but it's not the only worthy project. I didn't go to QMC out of haste. I carefully researched the projects before joining QMC.

Due to the different nature of projects, I don't believe there will ever be true credit parity.

Certainly there is an element of competiviness among crunchers. And I did get a "warm fuzzy feeling" when I broke 100k.

But a project's efforts and resource allocation, IMHO, should be focused on the project's goals. To engage in a round-robin of credit inflation, with potentially masses of users shifting back and forth to the "top of the pack of the day" does more harm than good.

IMO, near parity is not too much to expect from a project, but exact parity is.

Implying that users are somehow being pety or expecting eveness of credits over time is somehow expecting too much I think might be interpreted the wrong way by both individuals and teams who come to any project to compete.

And I'm not certain that Rosetta has an "unfair" credit system. My understanding is that Rosetta has choosen, as a definition to what a fair credit system should be, scientific work done.

Different projects, with different work to be done, may define a fair credit system differently.

While Rosetta may be at the "bottom of the pack", I don't believe it fair to say that it is not near-parity with other projects.

My understanding is that in terms of all Boinc projects, Rosetta has the fourth highest cumulative total of credits. If Rosetta is so far away from parity (on the under-granting end of the spectrum), would such achievement be possible?

SETI@Home established the standard for all BOINC projects by being the first to conceive and implement DC/Grid Computing. SETI saw fit to establish a fair credit system to reflect as honestly as possible a person's/machine's contribution to the project.

I think the second statement really reflects what the credit system concept is really all about.

And I crunch for other projects as well. But I do not do so out of concern for whether their credit system is at the "top of the pack" or "bottom of the pack". Personal choice.

I still beleive that Rosie is a worthy project, but it's not the only worthy project. I didn't go to QMC out of haste. I carefully researched the projects before joining QMC.

Rosetta just got closer to being "top of the pack" - Einstein has further reduced it's granted credits for the next series of wu's.

---

Due to the different nature of projects, I don't believe there will ever be true credit parity.

Certainly there is an element of competiviness among crunchers. And I did get a "warm fuzzy feeling" when I broke 100k.

IMO, near parity is not too much to expect from a project, but exact parity is.

While Rosetta may be at the "bottom of the pack", I don't believe it fair to say that it is not near-parity with other projects.

Well lets see... 375 ppd at predictor, I said in an earlier post that if the credits fell to 100 ppd less than predictor, the boxes would be pulled. Thats over 25% less credit. Guess what, there gone...

My understanding is that in terms of all Boinc projects, Rosetta has the fourth highest cumulative total of credits. If Rosetta is so far away from parity (on the under-granting end of the spectrum), would such achievement be possible?

Because up until last Nov. you received whatever credit you claimed. And inflated client software was allowed. And my 4 boxes that were getting 375 ppd ea on Predictor with the quorum of 3 system, were getting over 825 ppd ea here with the inflated client software. My 5 highest daily totals are still from the 2 weeks before the credit system changed, topping out with a whopping 3,825, the five days I ran the inflated client software. It was that way from the beginning. I was only here for the last 2 weeks, then the new credit system was implemented. Thats right more than half of this projects credits came from inflated credit allowance. Its only been 6 months since the new credit system has been implemented. Thats how.

SETI@Home established the standard for all BOINC projects by being the first to conceive and implement DC/Grid Computing. SETI saw fit to establish a fair credit system to reflect as honestly as possible a person's/machine's contribution to the project.

I think the second statement really reflects what the credit system concept is really all about.

If all the projects were forced into being within 10% of each other, their would be no reason to be having this conversation. I think that would satisfy everyone .

Personal choice.

Yep my personal choice is to not let my 4 rigs suffer the indignity of 25% less credit. If nothing is done to correct this, 3 more rigs will be leaving soon. If it was your goal to get me miffed and sidetracked you have done a wonderful job.

And you are not helping the project by trying to be the Rosetta cop. You just want to make me leave sooner... Sometimes its best to know what you are talking about before stating facts that are untrue. Go ahead and make my day by stating some other facts that are not backed up by the project... and see me gone.

I believe this conversation was between me and Mod Sense.

I really did not want to get into a fight with a Bad Penquin. But you just had to interject yourshelf into the middle of it. Please stay out of it. I dont care what you think is fair or not, I am stating facts. And trying to get the project admins to see they have a problem and to do something about it before more damage is done.

---

BETA = Bahhh

Way too many errors, killing both the credit & RAC.

And I still think the (New and Improved) credit system is not ready for prime time...

Sorry, didn't realize that private conversations were being held on public message boards instead of pm's.

I believe this conversation was between me and Mod Sense.

Nor do I with you. Must be that darn Socratic method from school. I am still willing to be convinced that there is a "problem".

I really did not want to get into a fight with a Bad Penquin... And trying to get the project admins to see they have a problem and to do something about it before more damage is done.

OK, then if that is true, and the primary concern is that the majority of crunchers want to obtain the absolute maximum number of credits they can, and Rosetta is at the "bottom of the pack", why for the most 24 hour reporting period, out of all the Boinc projects did Rosetta obtain the third highest number of credits? The most recent results should have nothing to do with the pre-Nov. credit system.

Because up until last Nov. you received whatever credit you claimed. And inflated client software was allowed. And my 4 boxes that were getting 375 ppd ea on Predictor with the quorum of 3 system, were getting over 825 ppd ea here with the inflated client software. My 5 highest daily totals are still from the 2 weeks before the credit system changed, topping out with a whopping 3,825, the five days I ran the inflated client software. It was that way from the beginning. I was only here for the last 2 weeks, then the new credit system was implemented. Thats right more than half of this projects credits came from inflated credit allowance. Its only been 6 months since the new credit system has been implemented. Thats how.

My understanding is that in terms of all Boinc projects, Rosetta has the fourth highest cumulative total of credits. If Rosetta is so far away from parity (on the under-granting end of the spectrum), would such achievement be possible?

Interesting choice of words: "If all the projects were forced..."

Again, please anyone correct me if I am wrong, but hasn't David Anderson stated just the opposite?

I don't know this answer, so what is your estimation of the amount of effort and resources that projects would have to divert from their primary goal, in order to coordinate with each other and implement this within their own projects?

And what happens if a single project refuses, and decides to award 25% more credit just to drain away users from all the other projects who invested all the effort and resources?

If all the projects were forced into being within 10% of each other, their would be no reason to be having this conversation. I think that would satisfy everyone.

I'm open minded, and I don't believe I have been rude. If I am wrong about something, please point it out.

Sometimes its best to know what you are talking about before stating facts that are untrue.