How is Recent Average Credit calculated?

Don
Don
Joined: 16 Sep 06
Posts: 1
Credit: 1,266,621
RAC: 0
Topic 191907

Sorry if I missed this in the documentation, but what exactly is RAC and what time frame does it use to do the calculation?

MattDavis
MattDavis
Joined: 18 Jan 05
Posts: 68
Credit: 10,565,341
RAC: 0

How is Recent Average Credit calculated?

I hope you have a PhD in mathematics.

http://boinc-wiki.ath.cx/index.php?title=RAC

tullio
tullio
Joined: 22 Jan 05
Posts: 2,118
Credit: 61,407,735
RAC: 0

RE: I hope you have a PhD

Message 47524 in response to message 47523

Quote:

I hope you have a PhD in mathematics.

http://boinc-wiki.ath.cx/index.php?title=RAC


Lie algebras are easier.
Tullio

Alinator
Alinator
Joined: 8 May 05
Posts: 927
Credit: 9,352,143
RAC: 0

RE: Sorry if I missed this

Quote:
Sorry if I missed this in the documentation, but what exactly is RAC and what time frame does it use to do the calculation?

It is essentially a "rolling" average, incorporating an exponential decay factor (half life of 1 week). It is re-calculated on every credit granting event for every host you have attached, any team you're a member of, and your account overall.

Formula here:

http://boinc.berkeley.edu/credit.php

HTH,

Alinator

neubaum
neubaum
Joined: 11 Sep 06
Posts: 23
Credit: 61,959
RAC: 0

RE: RE: Sorry if I missed

Message 47526 in response to message 47525

Quote:
Quote:
Sorry if I missed this in the documentation, but what exactly is RAC and what time frame does it use to do the calculation?

It is essentially a "rolling" average, incorporating an exponential decay factor (half life of 1 week). It is re-calculated on every credit granting event for every host you have attached, any team you're a member of, and your account overall.

Formula here:

http://boinc.berkeley.edu/credit.php

HTH,

Alinator

I'd like to add some related questions here:
1) Is RAC the best oveall mesure of how efficiently your computer/-s contibute to the attached projects within a given timespan?
2) Is RAC= the "Host/User Average" mesures you find in BOINC's Statistic pane, or? What about the difference between "Host" and "User"?. My stats-curves are identical when I compare Host with User.
3) I've found some indication that the Average is sensitive to the "Connect to network about every xx.xx days" entry in General Prefernces. May it be so?
4) If the answer to (1) is yes, is there some simple rules of thumb for optimizing the RAC, e. g. by fine-tuning your BOINC Preferences, and where are they to be found. I've tried to find out browsing the Message Boards and Wiki but the info there is so split up, sometimes incongruous and often far to technical for me to understand.
5) By the way what does these Rank% refer to?

I'm grateful for any suggestion.

Annika
Annika
Joined: 8 Aug 06
Posts: 720
Credit: 494,410
RAC: 0

2. The difference between

2. The difference between host and user is that "host" means only one single computer whereas "user" refers to your whole account. If you have only one box attached the two will indeed be identical, but with two or more boxes there will be a difference, so that you can also see which computer contributes how much to your total credit.
3. Maybe if you connect less often the RAC will be a bit less constant because your credit is upgraded less frequently.
5. The % rank means that you are better than x % of all users in a given project. For example in Seti you are among the top 20 % because you are at 81 %. So, higher is obviously better ;-)

Alinator
Alinator
Joined: 8 May 05
Posts: 927
Credit: 9,352,143
RAC: 0

A couple of other

A couple of other comments:

1.) The only catch with using RAC as an efficiency metric is that it is not solely dependant on your machine. IOW, a credit event which causes a RAC recomputation is dependant on when the quorum is formed and credit granted, which is not necessarily when your machine reports.

3.) You are correct with your analysis here. If you set your CI to a longer interval, you increase the probability your machine will be granted credited as soon as it reports. Therefore this would tend to make RAC a better metric for gauging overall performance for that host.

Alinator

neubaum
neubaum
Joined: 11 Sep 06
Posts: 23
Credit: 61,959
RAC: 0

RE: 3.) You are correct

Message 47529 in response to message 47528


Quote:
3.) You are correct with your analysis here. If you set your CI to a longer interval, you increase the probability your machine will be granted credited as soon as it reports. Therefore this would tend to make RAC a better metric for gauging overall performance for that host.

My RAC-curves (S@h and E@h) turned downwards a couple of weeks ago, after constantly been raising since I joined BOINC. I think it was due to my change of "CI" from 0.01/0.05 days to 0.25 days in BOINCs "General Preferences" and after I went back to 0.05 a couple of days ago the curves seem to turn upwards again. It may be a coincidence of course. However it doesn't seem to match what you say about "longer intervals". Going from "Conect to network about" 0.25 days to 0.05 days should diminish "the interval" and, according to your analysis, lower the RAC or ...? I'd be happy for further comments on this matter- ;-)

Alinator
Alinator
Joined: 8 May 05
Posts: 927
Credit: 9,352,143
RAC: 0

RE: RE: 3.) You are

Message 47530 in response to message 47529

Quote:

Quote:
3.) You are correct with your analysis here. If you set your CI to a longer interval, you increase the probability your machine will be granted credited as soon as it reports. Therefore this would tend to make RAC a better metric for gauging overall performance for that host.

My RAC-curves (S@h and E@h) turned downwards a couple of weeks ago, after constantly been raising since I joined BOINC. I think it was due to my change of "CI" from 0.01/0.05 days to 0.25 days in BOINCs "General Preferences" and after I went back to 0.05 a couple of days ago the curves seem to turn upwards again. It may be a coincidence of course. However it doesn't seem to match what you say about "longer intervals". Going from "Conect to network about" 0.25 days to 0.05 days should diminish "the interval" and, according to your analysis, lower the RAC or ...? I'd be happy for further comments on this matter- ;-)

Keep in mind that there is always a "Luck of the Draw" factor when it comes to RAC. As I said before RAC is recomputed for your account overall, every host host you have attached, and every team you're a member of on every credit granting event. Pendings do not count towards RAC until they become granted.

That being said, 0.25 days or less is still a pretty short CI if you get teamed up with users running a CI of 7 days or more for example. In this case the odds are your host will finish and report long before the others do, therefore you will have a tendency to build a fair amount of pending credit which isn't being counted towards your RAC. Now lets suppose your host completes one which gets granted right away. Since there hasn't been a credit granting event for awhile, when the RAC computation is performed the decay factor on your previous value becomes larger due to the longer time interval and the new RAC goes down. On the other hand you may observe your RAC increase even if your host hasn't reported another completed result recently, due to the pendings getting granted as the other hosts complete the result and report.

FWIW, my experience has been most BOINCer's run a CI of around 3 to 5 days, so if you set your's smomewhere in there you should see your RAC stabilize eventually to a fairly tight tolerance band.

Hope that clarifies things a little. :-)

Alinator

Lt. Cmdr. Daze
Lt. Cmdr. Daze
Joined: 19 Apr 06
Posts: 756
Credit: 82,361
RAC: 0

In addition: I run in RAM as

In addition: I run in RAM as I don't have a HD in my laptop. So, I have a small CI. Still, my RAC is quite stable, although with pending credits for two crunching days. IMO, the length of CI has no real influence on RAC. In the beginning I got almost zero credits. When the pending credits got piled, I got credits from this. This results in a stable RAC.

The length of CI does not depend on stability on RAC, but on other factors like a smaller risk of running out of WUS due to server outage and personal circumstances/preferences.

Happy crunching,
Bert

Somnio ergo sum

neubaum
neubaum
Joined: 11 Sep 06
Posts: 23
Credit: 61,959
RAC: 0

RE: RE: RE: 3.) You

Message 47532 in response to message 47530

Quote:
Quote:

Quote:
3.) You are correct with your analysis here. If you set your CI to a longer interval, you increase the probability your machine will be granted credited as soon as it reports. Therefore this would tend to make RAC a better metric for gauging overall performance for that host.

My RAC-curves (S@h and E@h) turned downwards a couple of weeks ago, after constantly been raising since I joined BOINC. I think it was due to my change of "CI" from 0.01/0.05 days to 0.25 days in BOINCs "General Preferences" and after I went back to 0.05 a couple of days ago the curves seem to turn upwards again. It may be a coincidence of course. However it doesn't seem to match what you say about "longer intervals". Going from "Conect to network about" 0.25 days to 0.05 days should diminish "the interval" and, according to your analysis, lower the RAC or ...? I'd be happy for further comments on this matter- ;-)

Keep in mind that there is always a "Luck of the Draw" factor when it comes to RAC. As I said before RAC is recomputed for your account overall, every host host you have attached, and every team you're a member of on every credit granting event. Pendings do not count towards RAC until they become granted.

That being said, 0.25 days or less is still a pretty short CI if you get teamed up with users running a CI of 7 days or more for example. In this case the odds are your host will finish and report long before the others do, therefore you will have a tendency to build a fair amount of pending credit which isn't being counted towards your RAC. Now lets suppose your host completes one which gets granted right away. Since there hasn't been a credit granting event for awhile, when the RAC computation is performed the decay factor on your previous value becomes larger due to the longer time interval and the new RAC goes down. On the other hand you may observe your RAC increase even if your host hasn't reported another completed result recently, due to the pendings getting granted as the other hosts complete the result and report.

FWIW, my experience has been most BOINCer's run a CI of around 3 to 5 days, so if you set your's smomewhere in there you should see your RAC stabilize eventually to a fairly tight tolerance band.

Hope that clarifies things a little. :-)

Alinator

I crunch on my own with one box only (Dual 2.7 2.7 GHz PPC G5 and 1.5 GB DDR SDRAM with OSX 10.4,8). It may simplify matters some. I'll stick to CI=0.05 for a while and see if the RAC-curve will levels off and stay around some value. Besides I may of course read in things that are not there. Especially I feel unsure about the scale of variation in these RAC by date diagrams. I am tempted to stay very relgular as conserns my computer usage and not change any preferences for a long time to get a picture of the variation of RAC due to other factors.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.