Pages: [1]
Gopherboy76
 
BAM!ID: 27158
Joined: 2007-05-22
Posts: 50
Credits: 3,520,846
World-rank: 99,692

2007-06-10 01:12:31

Is there a general formula for awarding credits according to wu's/cpu time/complexity or wu/results, etc or does each project have their own credit scoring system for work units?

It just seems that some projects seem to churn out more credit than others even though wu's are set to have an equal share of processor time.

My example is QMC, Einstein and Predictor all seem to be climbing twice as fast as other projects (On user total graph in BOINC Client)
Gopherboy76
 
BAM!ID: 27158
Joined: 2007-05-22
Posts: 50
Credits: 3,520,846
World-rank: 99,692

2007-06-10 01:28:18

Thanks to AlphaLaser who just let me know a few details that projects reward credit on an individual basis.


"Each project determines credit differently. The default is to base it on BOINC benchmarks and CPU time, but to name a few exceptions: SETI has a FLOP-counting scheme and CPDN uses a fixed credit per trickle." - AlphaLaser


I'm still curious if anyone knows any rough or exact math for any projects credit calculator?
Trog Dog
 
Volunteer tester
BAM!ID: 206
Joined: 2006-05-11
Posts: 119
Credits: 66,172,318
World-rank: 12,522

2007-06-10 01:59:11
last modified: 2007-06-10 02:04:15

Thanks to AlphaLaser who just let me know a few details that projects reward credit on an individual basis.


"Each project determines credit differently. The default is to base it on BOINC benchmarks and CPU time, but to name a few exceptions: SETI has a FLOP-counting scheme and CPDN uses a fixed credit per trickle." - AlphaLaser


I'm still curious if anyone knows any rough or exact math for any projects credit calculator?


http://en.wikipedia.org/wiki/BOINC_Credit_System is a good place to start.

For the benchmark based projects - Claimed credit is a function of time taken to complete wu and host benchmark score.

For non-benchmark based projects you're going to have to dig around in each project for the answer.

EDIT: and here's a page from the BOINC wiki (which is what I initially thought I was posting) http://boinc-wiki.ath.cx/Credit
Gopherboy76
 
BAM!ID: 27158
Joined: 2007-05-22
Posts: 50
Credits: 3,520,846
World-rank: 99,692

2007-06-10 02:58:15

Many thanks, it's more or less what I've been looking for, Great!
Gopherboy76
 
BAM!ID: 27158
Joined: 2007-05-22
Posts: 50
Credits: 3,520,846
World-rank: 99,692

2007-06-10 03:42:28

I'm a little suprised that that there isn't a standardised credit system.
I understand that each project has different variables, larger or smaller wu's, etc. but I'm finding in my personal stats that projects such as ufluids don't seem to reward as much as others such as Einstein for the same amount of CPU time.

The main thinking here is that alot of people who are stat orientated and aiming to climb the stats tables may drop some projects in favour of others for better credit rewards which may leave some projects lacking in CPU time.

My stats show if I dropped ufluids and put that processor time into Einstein or QMC, I'd end up gaining more credit for the same amount of CPU time.

But anyway, like I said, it's just a personal thought/observasion.
AlphaLaser
BAM!ID: 4118
Joined: 2006-08-16
Posts: 98
Credits: 1,209,990,414
World-rank: 1,561

2007-06-10 05:26:31
last modified: 2007-06-10 05:30:58

What you are finding is in fact correct. Some projects award more or less credit for the same work and time. This has caused credit-oriented participants to float between projects just for that reason. The fact also weakens the meaning of combined BOINC credit, because your 10,000 combined credits and my 10,000 combined credits don't really mean the same if we run a different set of projects.

A nice page is this credit comparision table. What you are seeing are the ratio in credit granted between the project in the row to the project in the column. For example, if you go to the row labeled NanoHive@home, you see that most of the numbers going across are close to 2, so it is granting very high credit compared to the other projects. As you found Einstein also gives rather high credit.

There is a buzzword, "cross-project parity," where there is an attempt to keep credit even between all projects. The differences exist in the first place for a couple reasons. One is what I already mentioned: different methods for granting credit, which can make it hard to ensure the different methods work out the same in terms of credit/cpu time. In addition you have optimized science applications which do more processing in the same amount of time, so you might be getting more credit if you run projects with those apps than ones without. SETI@home and RieselSieve are two projects I know of where optimized apps are found.

For "cross-project parity" to work requires cooperation between all the project admins, and with so many projects and with credit not always top priority for projects, it would be very difficult to achieve.
Saenger
 
Tester - Translator
BAM!ID: 5
Joined: 2006-01-10
Posts: 1735
Credits: 140,538,026
World-rank: 7,375

2007-06-10 09:26:22

Another reason for non-parity between the projects on one special machine may be that it's simply better suited for on project then another.

Some OS/CPU combinations are better for Einstein, some for Rosetta, some for CPDN...

There are three possible fair methods of granting:

1) Take a measure for the possibilities of a puter (benchmark) and count the donated CPU-time, multiplied with some factor to make this number manageble, and you have more or less the original system.

But the benchmarks were easy to manipulate, so a serverside solution was looked for, say #2:

2) If you know beforehand how much effort a WU will take, you can grant a certain amount per WU, consistent with the amount generated via 1), and grant it regardless of benchmarks.

This will as well show how good your puter is for this particular project, as a well suited puter will get more c/h then a not so well suited.

let's not forget the othe clientside solution:

3) You can count the actual mathematical operations (flop-count) within the WU. AFAIK it's the Seti-method.
I'm no puter wizzard, but I think this will have to take some time away from actual science, and it will probably show as well different c/h amounts for different projects on the same puter.

1) is fair as it grants the same for the donated puter power, if the projects don't utilize it, it's not the participants fault. On the other hand will nobody know whether his/her puter will be better on another project.

2) and 3) are fair, as they grant the same credit for the same science accomplished, and it's easy to see where my puter is good at.
Grüße vom Sänger
Saenger
 
Tester - Translator
BAM!ID: 5
Joined: 2006-01-10
Posts: 1735
Credits: 140,538,026
World-rank: 7,375

2007-06-10 10:44:25

Too late for an edit, so it's gotta be a new post.

Here's an old discussion about this topic in this forum.
Grüße vom Sänger
Pages: [1]

Index :: BOINCstats general :: Basis for rewarding credits per wu
Reason: