I've noticed that Rosetta grants credit very poorly, compared to some other projects and I don't understand why.

For example, this is a Rosetta work unit:

CPU time,sec claimed credit granted credit
10,282.01 51.30 76.79

This is from another project:

CPU time,sec claimed credit granted credit
10,297.95 48.87 143.37

These results are very consistent, it is not only one instance.
Can somebody please explain this to me, as I have already started to move away from Rosetta in favor for other similar projects, which give better credit.

I've noticed that Rosetta grants credit very poorly, compared to some other projects and I don't understand why.

For example, this is a Rosetta work unit:

CPU time,sec claimed credit granted credit
10,282.01 51.30 76.79

This is from another project:

CPU time,sec claimed credit granted credit
10,297.95 48.87 143.37

These results are very consistent, it is not only one instance.
Can somebody please explain this to me, as I have already started to move away from Rosetta in favor for other similar projects, which give better credit.

Dr. David Anderson of Seti and the creator and still main programmer of Boinc thinks all projects should give exactly the same amount of credit to for each chunk of time spent crunching for all projects and has pre-programmed that amount into the Boinc Server side of the software. Some projects disagree with that idea and change that, some don't. The reasons a project may change the numbers vary by project and are as diverse as the Boinc projects themselves. The best thing is to consider credits as comparable within a specific project and not worry so much about how many you have but rather what you are accomplishing with your cpu time. Credits after all are worthless in the real World, whereas our crunching time can mean alot to a project.

Rosetta@Home credits are more valuable than the other projects' credits.

---

Very interesting statement Chilean.

Personally I don't know how ones points can be better than other...

---

WWW of Polish National Team - Join! Crunch! Win!

Very interesting statement Chilean.

Personally I don't know how ones points can be better than other...

I think his opinion is in the 'eyes of the beholder'! I respect it too!

Very interesting statement Chilean.

Personally I don't know how ones points can be better than other...

I didn't mean more valuable as in, the project is better than any other project. But due to the mere fact that it is harder to get the same amount of credits than another project, then they are more valuable. At least that's how I see it.

---

R@H credits are based on the number of decoys generated - not specifically based on the CPU time required to generate these decoys. Some work units yield more decoys than others. Generally speaking, everyone should get a similar distribution of work units creating a consistent benchmark for everyone crunching R@H. These credits are not easily compared to other projects.

R@H historically yields lower credit than other projects. We appreciate your help as we are not chasing credits, we are looking for ways to improve the lives of people by better understanding how proteins work.

---

Thx!

Paul

My understanding is that's all correct except for this bit:

David Anderson, earlier referred to by Mikey, is the one who determined how credits would be claimed. So ultimately he determined how Rosetta would grant credits. :)

I think the standard BOINC way to calculate credits is (benchmark score * cpu time) whereas Rosetta only uses that for the claimed credit. Here, the granted is, as you quoted, the average credit claimed in all currently submitted jobs of that work-unit * number of decoys processed.

---

What this means, I think, is:
Yes, the number of credits to grant is not directly related to the CPU time.
But, is related because claimed credits are used to determine granted credits. And claimed credits are calculated using CPU time.

David Anderson, earlier referred to by Mikey, is the one who determined how credits would be claimed. So ultimately he determined how Rosetta would grant credits. :)

Dr. Anderson wants every Project to give the same amount of credits for the same amount of work and even wrote that into the server side of the Boinc software, it is now distributed with EVERY update. However each Project is run autonomously and therefore can change that if they chose too. DNETC, Androinc, Moo, Collatz there are MANY others that chose to change that amount of instead use their own numbers. For those projects it just means that EVERY time they update the server software, there are LOTS of little tweaks to do and that is but one of them. I saw a chart the other day someone made up showing the different projects and how the credit granting compares between them, Rosetta is NOT among the highest!

Credits though have always been a 'sticky wicket' though, if you give a ton of credits for almost no work, people will come that have no long term plans and force your Project to give them work they may not even finish when some other project comes along that pays even more. To me credits are needed but if you give too few people will show absolutely no loyalty and leave at the drop of a hat. You need to give enough credits to get people to come and stay, but not so many credits as to attract those that will not stay for a long time. I also think Dr. Anderson's idea of cross credit comparisons are ridiculous and ill fated!! IMO credits should be compared WITHIN a Project ONLY! I will let my signature show for illustration purposes. I have over 300 million cobblestones total, but that means nothing to Rosetta where I do not even have 200 thousand! A week and a half ago I put some gpu's onto a new for me project, I now have almost 10 million credits there, that is just amazing to me and in no way is it comparable to Rosetta!!!

---

Yes, that is what I said, didn't I? At what point did I go wrong?

David Anderson's method (and therefore the BOINC method) is to use the BOINC benchmark * time, and rosetta doesn't use that for the granted credit so DA didn't determine how rosetta grants credits - the credit granting method here is unique to rosetta. ;)

---

So dcdc is talking about the method... transient is talking about the basis... both apply. Dr. Anderson may not have come up with the method, but the method uses the benchmarks as the basis for establishing how much credit a decoy of a given work unit it worth.

---

Rosetta Moderator: Mod.Sense

Why shouldn't we get the same number of points from each project? Why should Rosetta change it, what is the point?

---

Why shouldn't we get the same number of points from each project? Why should Rosetta change it, what is the point?

Because each project has its own goals and aspirations, meaning if a project is looking to get a result by the end of the month they just crank up the credits and all the folks looking for lots of credits crunch there, while those projects looking more long term lose those folks, and may never get them back. There are around 50 or more Boinc projects right now, in total, and each wants YOU AND I crunching for them! The way they 'convince' us to come to them is with the credits, otherwise they are just another Science or Math or whatever project. There are TONS of specialty projects whether you want to look for ET, the cure for a disease and there are LOTS of those projects, solve a Math problem, map the Milky Way or even search the Universe for other things. How does one project get YOU AND I to crunch for THEM instead of any other one, credits is the easiest way! If people would just stop trying to compare credits among projects it would be fine, here at Rosetta I have x number of credits, over at DNETC I have y number of credits, they were earned differently by different pc's why should the credits be comparable, is my thought process! It is kind of like comparing the work of you and I, you are you and I am me, we are different but both working towards the same goals here. I have 12 pc's crunching for Rosie right now while you have 3, how is that comparable? Answer is...that we are both trying to help Rosie solve what they are looking for!

Why shouldn't we get the same number of points from each project? Why should Rosetta change it, what is the point?

Inter-project credit parity is nice in theory but in reality it's completely flawed. Different projects have different requirements and different computers have different relative strengths and weaknesses. The BOINC benchmark is far too simplistic to account for any of this variation (or at least it used to be - they might have upgraded it, not that it changes the principle?).

Basic example: a CPU with a large L2/L3 cache will do much better at Rosetta than a comparable CPU without, but on another project the size of the cache might make no difference. Do they get the same amount of credit per hour or different? The upcoming Bulldozer CPUs might be a good example - they might be excellent at Rosetta due to the balance of integer/FPU work required, but might be terrible at Einstein for the same reason.

How much credit does a GPU get vs a CPU, when the CPU is slower but is doing tasks that the GPUs can't do?

That's why Rosetta gives credit based on how good the computer is at crunching rosetta tasks.

---

Well, after reading through all the opinions, I have come to a conclusion that the system of calculating and granting of credit in R@H at least is somewhat inefficient. In my opinion, what really should matter is the total actual accumulative and may be normalized amount of CPU time in general spent for processing a unit and accumulated within the project or across the projects for all processed units.
Think about this, running a computer at 100% load is equal to a 100 Wt bulb at least, and is something you need pay for in your electricity bill! Well, at least getting a higher credit can be considered a "reward" sort of, especially when you participate in one of the meaningful projects, like folding proteins or similar, aimed to find cures for diseases. I would never spend a second of CPU time for projects like SETI( as I personally do not believe in such things as extraterrestrial intelligence ) or pursuing the mathematical challenges on the other hand.
In any case, I am out of Rosetta for good, in favor of other similar projects, which give better credits for the very similar science behind.

Well, after reading through all the opinions, I have come to a conclusion that the system of calculating and granting of credit in R@H at least is somewhat inefficient. In my opinion, what really should matter is the total actual accumulative and may be normalized amount of CPU time in general spent for processing a unit and accumulated within the project or across the projects for all processed units.
Think about this, running a computer at 100% load is equal to a 100 Wt bulb at least, and is something you need pay for in your electricity bill! Well, at least getting a higher credit can be considered a "reward" sort of, especially when you participate in one of the meaningful projects, like folding proteins or similar, aimed to find cures for diseases. I would never spend a second of CPU time for projects like SETI( as I personally do not believe in such things as extraterrestrial intelligence ) or pursuing the mathematical challenges on the other hand.

So in your opinion an i5/i7 CPU @ 4GHz should get the same credit per second as a celeron 400 from 1999? You might want a higher average credit, but CPU time is irrelevant to the project - the important thing is work done and that's what credit is granted against.

In any case, I am out of Rosetta for good, in favor of other similar projects, which give better credits for the very similar science behind.

What other project has very similar science?

---

We all have our own goals. I only look at credit within project so see how I am progressing. I may look at my postion compared to other to see if I am holding position. I use this as a personal benchmark to see that things are runing as it should.

If Joe (???) has been getting 50 RAC and I typically get 47, does that reamin about constant? If so, then things are running about right. If Joe is running 200 suddenly, no new machine, ad I am down to 25 ... something changed.

I don't understand how these roll up to some global BOINC credit bank. But if accumulating the most credit was what I cared about then I would run only those projects that give me the most credit per CPU cycle.

We all have our own goals for running these projects. Mine is to make some small but useful contribution to research that might someday save the life of my kids or their kids.

So in your opinion an i5/i7 CPU @ 4GHz should get the same credit per second as a celeron 400 from 1999? You might want a higher average credit, but CPU time is irrelevant to the project - the important thing is work done and that's what credit is granted against.

Why does it have to be "per second" ?
I have to admit that I may not fully understand all the nuances, but there is concrete quantitative factor, called number of floating-point operations.
In fact, on my account page in the Rosetta project there is a link called "Certificate" which reads when clicked, that I have contributed 588.81 quadrillion floating-point operations to the project. Why can't this number be a simple and straight-forward cross-project factor to represent my total credit? Obviously, if you have a faster CPU( more floating operations per second ) you will get a bigger number - bigger credit over the same elapsed period of time.

What other project has very similar science?

I meant to say, similar projects in the area of biology and medicine, such as POEM, SI MAP, DOCKING and others.

Docking project, for example also ".. aims to help cure diseases such as Human Immunodeficiency Virus (HIV)."

So in your opinion an i5/i7 CPU @ 4GHz should get the same credit per second as a celeron 400 from 1999? You might want a higher average credit, but CPU time is irrelevant to the project - the important thing is work done and that's what credit is granted against.

Why does it have to be "per second" ?
I have to admit that I may not fully understand all the nuances, but there is concrete quantitative factor, called number of floating-point operations.
In fact, on my account page in the Rosetta project there is a link called "Certificate" which reads when clicked, that I have contributed 588.81 quadrillion floating-point operations to the project. Why can't this number be a simple and straight-forward cross-project factor to represent my total credit? Obviously, if you have a faster CPU( more floating operations per second ) you will get a bigger number - bigger credit over the same elapsed period of time.

Apologies - I misunderstood your point there. The first post I ever made when I started running United Devices back in '99 or 2000 was exactly that. I'm not sure I ever got an answer then. My understanding now is that I don't think counting flops directly is possible, or if it is, it takes a large overhead of CPU activity, but I might be wrong. Half of BOINC's benchmark attempts to approximate that through the whetstone benchmark and the integer dhrystone benchmark makes up the other half. My understanding is that Rosetta is somewhere around 50-50 on floating point and integer work (rather than being FPU heavy as I originally thought it was). Counting flops would therefore only account for half of the work done, so the BOINC benchmark might give a better approximation, but that can be faked and doesn't take into account cache size, architecture efficiency, RAM speed etc over and above the benchmark which is far too small to capture much of that. Also, some people cheated the benchmark, which annoyed other people and the place was full of arguments over credit.
The current system gives you the average claimed credit for each of the models you've completed, and IMHO works really well, although not perfect. Faster computers complete more models within a work unit and therefore get more credit.

Some tasks will be better suited to some architectures than others - for example a larger L2/L3 cache might make larger models much faster than smaller models. If you could count each instruction calculated then you'd take that into account but I'm not sure that's practical.

What other project has very similar science?

I meant to say, similar projects in the area of biology and medicine, such as POEM, SI MAP, DOCKING and others.

Docking project, for example also ".. aims to help cure diseases such as Human Immunodeficiency Virus (HIV)."
[/quote]
Fair enough - not necessarily similar in science but similar in purpose ;)

---

Not that it matters but one area where the credit granting system does skew things horribly is in the printing of Certificates of Participation or whatever they are called.

Not that I have ever known anyone who actually prints and frames them.

It is easy to accept that a Rosetta credit is "worth more" in terms of expended cycles than those granted by other projects and if you are of a competitive nature comparing Rosetta credits to those granted by another project will likely put you at a distinct disadvantage.

For me the only real use of the credit system is to act as a "red flag" that something is wrong when your RAC takes a significant hit.

However, even when you accept that the amount of processing cycles required to earn a credit varies between projects there is still another real area of inconsistency.

When generating the "Certificate of Participation" one credit is deemed to be equal to one cobblestone of work. A standard conversion is then made between cobblestones and floating point operations.

This conversion appears to be consistent across all projects - it is likely embedded in the "template" project website. After looking at a number of projects it appears that 1 cobblestone of work translates to 11,574 million floating point operations.

Thus you have to crank out a more hardware cycles with Rosetta to be credited with contributing a floating point operation than you do with other projects.

Since a floating point operation is an easily measured function of the hardware you would expect the number of floating point operations reported on the Certificate of Participation to closely reflect the actual number of hardware cycles donated, regardless of the project involved.

Not that it matters that much, but it was a good exercise for one of those sleepless nights.