BOINC credit/points explained

[sierra_bound]

For those of you who are new to WCG and use of the BOINC client, here's an explanation of how credit/points are awarded.

Quorum system

BOINC uses what is referred to as the quorum system. For WCG projects, all of members of the quorum have to complete the same work unit before that WU is validated and credit is awarded. This is to ensure that the WU has undergone enough processing to satisfy the scientists. The size of the quorum varies, depending on the project.

Below is an example of a validated WU. The "granted credit" is calculated by taking the average of the claimed credit results. This explains why your granted credit may be higher or lower than your claimed credit. Your result is the one that's highlighted (see below).

Update: Now that FAAH uses a quorum of 2, granted credit is awarded two ways. The standard BOINC practice is to award the lower of the two claimed credits. If the two claimed credits are similar, then the granted credit will be an average of the two, which was the way credit was awarded when the FAAH quorum was made up of 3 hosts.

[sierra_bound]

Validation time

There are various reasons why your finished WU may be taking a long time (days) to be validated. Even if, for example, it took you only three hours to complete a work unit, it may take another member of the quorum 10 hours or more to finish the same WU. Also, if one of the other quorum members has an error in his/her WU or does not finish it, then another member is recruited to crunch that work unit.

[sierra_bound]

Calculating points

The credit system is based on the power of your computer and the amount of time it takes to finish a WU. Your machine is measured against a so-called "reference computer" that scores 1,000 in both the Whetstone and Dhrystone CPU benchmarks. As most of you know, you can run those benchmarks using the BOINC manager. Just go to Advanced, and then Run CPU Benchmark. The results will appear under the Messages tab. The CPU benchmark is run automatically and periodically on your BOINC installation to determine your machine's processing power at a given time.

The formula used for calculating claimed credit is:

([whetstone]+ [dhrystone])/1000 * 100 / (2 X secs_per_day) * wu_cpu_time

Or:

([whetstone]+ [drystone]) * wu_cpu_time / 1728000

wu_cpu_time is the number of seconds taken to process a work unit.

If you want to learn more about how BOINC points are calculated, check out the BOINC Wiki.

http://boinc-wiki.ath.cx/index.php?t...NC_FAQ:_Credit

[Bootup05]

Nice thread

[Reinvented]

thanks Sierra. that explains a lot for those new people.

Sticky it!!!

[serialk11r]

Now I get it thanks!

[Movieman]

thanks Sierra. that explains a lot for those new people.

Sticky it!!!

Done and agreed, nice thread. Thanks Sierra_bound!

[[XC] serlv]

What I'm curious about/tend to feel about the benchmarking process is that it no longer seems to matter what our benches are. High or Low, or right in the middle. Of course if everyone had higher benches; if the quorum, for any given WU, was higher due, to all participants in the quorum scoring higher then it would matter.

So I just try to get the machine to run as fast as possible, stabile, so as to go through as many WUs as possible. And benchmark be damned. I barely even look at it anymore.

Am I wrong, or am I missing something here. In regards to the benchmark.

edit - If you'd like me to delete this, so as to unclutter a good and simple explanation/info post I will. Just let me know.

[sierra_bound]

Thanks guys

What I'm curious about/tend to feel about the benchmarking process is that it no longer seems to matter what our benches are. High or Low, or right in the middle. Of course if everyone had higher benches; if the quorum, for any given WU, was higher due, to all participants in the quorum then it would matter.

So I just try to get the machine to run as fast as possible, stabile, so as to go through as many WUs as possible. And benchmark be damned. I barely even look at it anymore.

Am I wrong, or am I missing something here. In regards to the benchmark.

I think you're right to a certain extent. You are somewhat at the mercy of the other participants in the quorum. But those other members can work in your favor too. I recently had claimed credit of around 70 on one WU, but ended up with 100 granted points.

[[XC] serlv]

That is nice when that does happen.

I'm just leary of ever having a WU points beng declared invalid, getting half credit, due to being an "outlier". I don't ever want to go back to that. So I just tend to not even look at the benches anymore, or try to rebench.

Thanks for the explanation, though. I was curious, too. Just never bothered to find out. Good info!

[sierra_bound]

Did WCG ever define what qualifies as an outlier? I remember there was a lengthy discussion by members here about this.

For those not familiar with what serlv is referring to, toward the end of last year, WCG implemented a new scoring system aimed at cutting down on cheating and the use of optimized clients. Basically any scores that appeared to be statistical outliers would be thrown out. There were fears that those with really fast machines would be punished as well. To be honest, I have not seen any widespread evidence of this.

[meshmesh]

No. The admins of WCG have not officially disclosed what the rules of their system are. But as you may recall, have been looking into it last December. And AFAIK, they have not changed the rules. From my understanding, this is how it looks so far from examples of quorums in my AMD results:



Notes:

Underclaiming: (examples 1-2) Occurs when Boinc occasionally under-benches. My understanding is that this problem is limited to C2D at the moment. Happened to me multiple times. In this case, the system comes to your rescue. The WUs are excluded from quorum and awarded the average of the other two. So no harm done to you.

normal: (examples 3-7) All WUs are claiming within a tight range. All awarded average. After the initial settling phase in late November, this is currently typical of the vast majority of the WUs.

Overclaiming: (examples 8-13) May occur if both other quorum members are running under-benched systems. Specifically their average claim must be below 73% of normal for this rule to incorrectly trigger againest your claim. Although it is possible, the probability of this is very low indead. So far, in all the cases I saw, it was because one other machine was definitly claiming too high, as seen above. From what is seen so far, if your machine is claiming normal, it will always be included in quorum calculations.

Outlier: (examples 14-16) The WU is flagged as "Invalid". There are three reasons this may occur for Wus returned by a machine: (a) the machine is running a non standard Boinc client. Solution, install standard client. (b) the machine swaps a lot to the HD and thus clocks a lot of wall CPU time without actually doing work. Solution would be to to add more ram and to set preference to "keep in memory". (c) both other quorum members are running way under-benched systems. Specifically their average claim must be below 53% of normal for this rule to incorrectly trigger againest your claim. So far, haven't seen it and I imagine it would not happen.

Edit: These rules do not apply to HFP2 which uses another system all together.

[sierra_bound]

I don't think there is much of a controversy, despite all the fears that were raised when the new system was first announced. The points system is not perfect. I don't think anyone denies that. But the way I look at it, everyone is stuck with it. Personally I worry more about their servers crashing, as was the case in late December.

[meshmesh]

I don't think there is much of a controversy, despite all the fears that were raised when the new system was first announced. The points system is not perfect. I don't think anyone denies that. But the way I look at it, everyone is stuck with it. Personally I worry more about their servers crashing, as was the case in late December.

Agree. Once the vast majority switched to running standard installations, all the claims fell within a tight margin very quickly. Also the fact that WCG separates the linux/Win/mac OS quorums solves a big problem.

So far, the system works surprisingly well.

[[XC] gomeler]

I just hate how slower machines get more credit per WU just due to the fact that it takes them decades to get the job done. There will probably never be a fair system that'll cater to everyone. Excellent analysis/thread, provides a nice insight into the system.

[meshmesh]

I just hate how slower machines get more credit per WU just due to the fact that it takes them decades to get the job done. There will probably never be a fair system that'll cater to everyone. Excellent analysis/thread, provides a nice insight into the system.

Actually they don't. They may "claim" more, but every one gets the same. The system is fair.

[[XC] Adywebb]

I just hate how slower machines get more credit per WU just due to the fact that it takes them decades to get the job done.

They don't exactly get more credit - they get the same as everyone else crunching that particular unit - faster machines get more credit overall due to crunching more units during the same amount of time.

The system isn't perfect by any means, and it never will be, but the principle of everyone getting the same credit for the same work completed is a valid one.


Edit: Sorry meshmesh you beat me to it!

[sierra_bound]

People with slower computers who are part of your quorum actually will help boost your granted credit. Below is an example. The downside is that it may take longer before that WU is finally validated.

[Vapor]

And there's no reason to not tweak as well as you can for the benchmark as well....if you can manage an extra 5% in the benchmark, that will average out to an extra 1.66% in points....and every bit counts

[[XC] gomeler]

I meant precisely that, say I do a WU along with a P3-600 and a P2-266. Both of them will claim much more than I would therefore boosting the value of the WU and giving them more credit than should be given. Always thought it was funny that turn-around time wasn't a factor, that's why I love number crunching projects, get points based on sheer performance, no evening out of the playing field.

[[XC] Adywebb]

I meant precisely that, say I do a WU along with a P3-600 and a P2-266. Both of them will claim much more than I would therefore boosting the value of the WU and giving them more credit than should be given.

Thats the thing they aren't getting more than they should be given - everyone crunching that unit gets the same credit as they've all done the same amount of work (1 WU) regardless of how long it took.

Edit: My mistake, I think I read what your trying to say now, that ALL the machines (including yours) are getting more, not just the two slow ones meaning the overall WU value is higher than say another similar WU crunched by 3 fast machines?

If thats the case then yes, but it evens out over time I suspect

[Okda]

first time to see this thread and glad i did

thanks for the nice explanation

[StyM]

1 boinc pt = 6.999 wcg pts.

[road-runner]

How many times a day do the stats update? From what I can tell only one time?

[[XC] junglemaster]

The WCG stats (from the WCG stats site) update 2 times.

Boinc stats - 1 time

Stats.free-dc.org - not sure...I think 1 time.