Max Daily WU Quota...
31 Dec 2005 4:54:41 UTC
Topic 190477
(moderation:
Thanks for increasing the quota to 16 :) ....we will monitor the faster machines, though, as I think you may have to raise it higher for them if they are assigned the low frequency WUs.....Cheers,Rog.
Max Daily WU Quota...
)
My 2.0 GHz G5 does work units in 51 minutes. I exchanged e-mails with Bruce about this and he will be monitoring and may increase the quota some more.
Though he also indicated that this may be an abnormal, or uncommon, set. As stated in the very postings about Albert, the work may vary in length. Though, I am still seeing 51 minutes per on the G5 ... :)
(And I don't even have the fastest G5 :))
That's a good news!
)
That's a good news!
Well, almost ... :) WIth
)
Well, almost ... :)
WIth 16 quota, I will only be out of Einstein@Home work 8 hours a day ... or another 16 work units ... or so ...
Hmmm... hope staff will take
)
Hmmm... hope staff will take this into account and increase max daily WUs even further. I don't like to waste CPU cycles and I don't want to join other BOINC projects just to keep my machines crunching. I joined BOINC because of E@H, not the opposite.
RE: Hmmm... hope staff will
)
There is a balance to be struck.
The quota wants to be big enough not to cramp anyone on 100% Einstein, but only just that big to keep the downside under control.
The downside of an incresased quota is that a rogue machine will take more wu out of circualtion. With the welcome decrease in the initial allocation from 4 down to 3 this means that every wu lost in a rogue box will not complete till past the original deadline. Complaints from participants about credit pending; delay for scientists to get their data, and size of scheduler databases will all increase each time the quotas go up.
So the project team are right to increase the quote slowly and in steps, and no doubt intend to go on increasing in small steps till those with the fastest boxes stop complaining. That way they get the smallest possible quota.
And every so often when AMD bring out a faster chip, the complaints will start again and the quota can be edged up just a little again.
In the meantime, if anyone has a fast box it would be as well to have another project on it, even if the long term aim is to go 100% Einstein.
By the way, in case anyone is concerned about multi-cpu boxes, that is not an issue, if you are using N cpus for BOINC then that box gets N times the quota.
If you have an HT box it is time to enable that as you will get 2x previous quota but not be crunching twice as fast, which is why it will be the AMD chips that define the borderline case.
~~gravywavy
Yes, I know ... you are right
)
Yes, I know ... you are right that they should not raise the quota too fast or too far.
And, if the work length does increase some it will not be an issue for me. THought the 51 minute run times are in line with what I had been seeing with the older application which was at 4+ hours per work unit.
So, it may have been luck of the draw that my G5 got all fast work ... I have noticed my Xeons have some work that is closer to 10 hours, much like the older application, so, maybe the next chunk I will get for the PowerMac will be longer ... we will see ...
I had been trying to get to be all Einstein@Home for a little bit to try to get my EAH total over SAH ... sigh ...
RE: ...maybe the next chunk
)
You will, it will just take a little longer.
And in the meantime, give the time to the _next_ project you want to get above SETI -- Rosetta maybe? -- so you are bringing the following goal that much closer.
I've said before, myself I don't have much problem getting projects above SETI ;-)
Except Orbit of course...
~~gravywavy
RE: My 2.0 GHz G5 does work
)
Paul just FYI...I'm running E@H on my G5 1.6GHz, I'm getting about
the same time on the Albert WU's. 51 minutes over the last 20 or so
WU's..reached my quota of 16....I detached from S@H for awhile to get
a better idea on the varying WU/Times..
RE: Paul just FYI...I'm
)
Well, it does seem to be WU dependent. I have a number of them that are still showing 51 minutes, but I also have a few that are listed as 2.5-3.5 hours. An hour in to one it looks to be only 33% done, so, the three hour time may be good.
With slightly longer processing times, it looks like the 16 quota may be sufficient for the time being. Though I did like the idea of doing a lot of work in a short period of time. :)
RE: With slightly longer
)
Well, it will have to be raised soon if E@H wants to maintain its resource share. Our AMD Athlon 3000+ is processing these LF WUs in about 65 minutes and it's not the fastest machine here. There has been no communication about this problem from the Devs so I guess all we can do is shift more resources over to R@H. I hope they realise that dedicated E@H resouces are bleeding away to other projects.....