My two-month old prize host keeps asking for GPU work for the GPU it lacks, but has not asked for CPU work for days. Any guesses as to why the lack of CPU work request?
Details:
Windows 7 64-bit Xeon E5620 running hyperthreaded, so nominally 8-core.
BOINC version 6.10.58
preference settings:
internet connection interval 0.004 days
additional work buffer requested 6.26 days
% of time BOINC client is running: 98.6201 %
While BOINC running, % of time host has an Internet connection: 99.9882 %
While BOINC running, % of time work is allowed: 99.9933 %
Task duration correction factor: 1.877434
The host most recently actually downloaded CPU work on October 23, though the most recent time it seems actually to have acquired the intended buffer is October 21.
Today, October 28 it is down to 26 GC S5 tasks and 7 ABP2 tasks unstarted, which is about 1.3 days of work. Yet all the work requests which I can find in the message log take the form: "requesting new tasks for GPU", and not the form which indicates that both CPU and GPU tasks were requested. The GPU request quite properly is denied, as the GPU in my system is ATI (of which use is enabled because I am running a Lunatics Astropulse ap on it over at SETI).
I've become accustomed to a bursty work fetch pattern on this box, as the performance of ABP2 and GC vs. the expectation set by the benchmark differ, so as the recent work mix shifts it will overfetch for a while, then go silent for perhaps two days, but this is the first time I have seen such an extreme queue draw-down.
Any ideas?
Copyright © 2024 Einstein@Home. All rights reserved.
wanting what it can't have
)
If your system is also (still) asking for work from Seti... well, they have just decided that there isn't going to be any work for the next month while they wait for the new servers to come in and set them up. So you can just as well set Seti to No new tasks, then BOINC will probably shake awake and start asking work for CPU work from Einstein as well.
RE: My two-month old prize
)
This is probably a 'status message' not a 'problem message'. It probably says:
"message from server your computer has no Nvidia gpu". The Nvidia would be replaced by ATI if you don't have an ATI card in the machine. This is NOT an 'error message" and is "normal" according to Dr. David Anderson the main programmer for Boinc.
RE: So you can just as
)
I had seen that status from the SETI project an hour or two before I read your message, but had not yet acted on it. I agreed with your advice and set SETI to No New Tasks, but after a couple of hours there still had been no requests for Einstein CPU work.
I did notice that BOINCView represented the status of Einstein on this host as "overworked". So I reviewed the debt status and decided to do a bit of tampering. I shut down BOINC and tampered down all nonzero debts by an exact factor of 10. Upon restarting BOINC, it immediately generated a request for Einstein CPU work, and immediately received 10 work units. Overnight it continued to receive work and by now the queue is approximately in a normal six-day shape.
So my self-diagnosis is that the matter had to do with the current state of debt on the host, which in turn was the result of my settings and of an odd episode a week or two ago in which I downloaded accidentally a day or so worth of SETI work, and then crunched most of it in order to avoid a binge purge sequence of EDF behavior alternating with no SETI.
Mikey, I understand that Doctor A thinks we should not mind spurious GPU requests--the problem and the question were that it was not generating CPU requests.
RE: Mikey, I understand
)
My understanding is that dr. A fancies GPUs so much that BOINC CC tries to crunch as much GPU work as possible even if it was to hit deadlines with CPU tasks.
Recently one of my Nvidia hosts received a few CPU tasks in addition to many GPU tasks. Those CPU tasks were just laying on disk platters waiting for deadline to hit in. Meanwhile it was happily crunching GPU tasks, reporting them and downloading new ones.
I couldn't bear watching it so I forced those CPU tasks to crunch through.
All of it was EAH, so inter-project debts couldn't play any role in it.
I think that GPU vs. CPU task scheduling needs some rethinking.
Metod ...
RE: RE: Mikey, I
)
There IS a Boinc Alpha Mailing List you can subscribe too but be advised Boinc is his 'baby' he created it, has maintained it all along, and sees complaints as 'attacks' against it. He is NOT close minded though, but he can be very short with people that have not done their homework and are re-hashing old arguments. But go for it, as a mailing list though it is not like he has to reply if he doesn't like your suggestion. Yes I have had some personal interactions with him in the past, on other subjects, and have my own opinions of him. They could be very different from your experiences!!
RE: RE: So you can just
)
I see that now and am very glad you thought of the debt issues as a possible solution, it obviously was in the end! As long as Seti is on this weird on again off again thing it may not be resolved and you may end up resetting it again and again. Right now Seti is down so is accumulating Long Term Debt against Einstein, meaning when Seti comes back up, you will again not get any Einstein workunits! The only way to prevent the LTD issue is too suspend a project in the Boinc Manager, just setting it to NNW will not stop the debt from accumulating.
RE: I think that GPU vs.
)
Go for it then: http://boinc.berkeley.edu/trac/wiki/ClientSchedOctTen?version=8
RE: RE: I think that GPU
)
Thanks for the link Jord I no longer had that!
RE: RE: I think that GPU
)
Thanks Jord! I've skimmed through that document and seems that at least some of problems I see are described there. I'll wait and see (and hope) if the new scheme works better.
Metod ...
RE: RE: I think that GPU
)
The estimated credit proposal is going to cause problems with apps that assign serverside credit significantly different than what the estimate is