A question about RAC
19 Jun 2006 15:39:00 UTC
Topic 191414
(moderation:
Could someone enlighten me as to why, after uploading 14 WU's, (2 of them the long S5 seriws), my RAC, which was 1,201 prior to this event climbs by -3.03, to end up at 1,198.03?
My previous uploads were 4 hours ago today - to get me to 1201.
Seems a bit* screwy to me that one adds a whole heap of work only to be back behind where one started ...
* - or 'extremely'
A question about RAC
)
RAC went down by 0.25% - wow, end of the World!
In general, participant has no influence on when he gets granted credit (on project where quorum is >1) because he can't predict when other results for quorum needed by validator will arive.
Another reason is transition from S4 to S5 which will affect return rate, turn-around time, pending WUs hence pending credit and also credit per hour.
Another reason is that stats get updated in interval not instantly.
Above mentioned change is statistically insignificant.
(to me, RAC doesn't mean much).
Credit and RAC is updated in
)
Credit and RAC is updated in real time, Honza.
You have a lot of pending
)
You have a lot of pending credits, will go up as the others for them come in and you get credit. It is already on the way back up, at 1,200 as I write this.
Like the rest of us your pending credit will go down as you get more of the S5 units.
Try the Pizza@Home project, good crunching.
RE: You have a lot of
)
Thank you very much guys ... I get your point Ray ...
As the results were not matched, they went into the "pending credit" list and were not recorded as work done. When a match is made, credit is granted and the train moves out of the station!
Thanks for that ... now may I ask a subsidiary question?
Is there any value in dividing CPU seconds by RAC granted ... in order to see which machine/WU mix performs best?
I have done that with S4 and S5, and find that I get a better return of "seconds per point of RAC" with S4 Units. So I am still crunching S4 on my main (and fastest) comp.
Any comments?
Many thanks
RE: Is there any value in
)
Yes, but...
1. In Einstein, at least a few months ago, you could get a pretty non-random sample of quorum partners, as a small subset of the total Einstein crunchers might share your major datafile for weeks on end. The "law of small numbers" implies that there was far more variation in actual claim versus standard claim than you might have expected from experience on other large projects.
I have two Coppermine PIII's. For weeks in February they differed in RAC (and also in simple credit/week) by far more than their modest real difference, just because of this chance of the quorum issues.
2. If you use RAC to compare machines, then it is important that the usage history be comparable for some weeks into the past, so you are not just comparing the RAC calculation response. In the most obvious example, and brand new machine will lag greatly in RAC for weeks, as the calculation history term drags it down.
But, reservations aside, your idea does make sense. I've compared RAC ratios across projects to re-allocate my small diverse fleet's individuals to best comparative advantage.
Not realy with the RAC. Just
)
Not realy with the RAC. Just look at the list of computers on your account and it list the RAC for each. The more work credit granted for the higher the RAC.
You may want to figure out the average granted credit per hour, that would mean more, just plug some figures into the Excel file that I have here and it will compair two systems. Easy to add a 3rd system, but I only have two so set up for that.
Ray
EDIT
Of course the projections there are way off as they are for S4 WU's. When tou get enoughf S5 units you can plug them in and get a better projection.
Try the Pizza@Home project, good crunching.
Many thanks Ray and archae
)
Many thanks Ray and archae ... I begin to understand ...
As my comps are so varied in power, I just wanna use them to the max for the programme ..
And as there are still S4's out there, someone's gotta do them ... !!!!!
... and Ray, I'll give that excel file of yours a twirl and see where we get to
Thanks guys ...
RE: But, reservations
)
In fact E@H and the various other projects are trying hard to 'align' their credit values so that all the projects are as close to equivalent as possible. But indeed some types of computers will perform better than others on particular projects.
(This is only meant as a comment, not as a criticism!)
Cheers,
Bruce
Director, Einstein@Home
RE: Credit and RAC is
)
Yes Pav - on project's sites.
My point was that it might be updated in real time but it can't be fully synchronized with other users (i.e. not running 24/7 one project with 100% share, permanent connection etc.) so the RAC (even if the formula was perfect) can't be accurate number - surely not withing 0.25% limit.
Honza, by definition the RAC
)
Honza, by definition the RAC is always accurate, because by definition it's validity is 7 days from the moment of last credit grant :)