"Not reporting or requesting tasks"
7 Sep 2011 2:47:36 UTC
Topic 195947
(moderation:
so i've noticed a disturbing trend on one of my hosts recently...the queue has been drying up, and despite having my cache set to 5 days worth of work, i probably have less than 10 hours of work left in the queue. it started a day or two ago when i noticed that it had been a while since BOINC requested new CPU tasks. now keep in mind that i crunch only CPU tasks for E@H, and not GPU tasks...in fact, i don't even have a capable GPU in that host. despite that, BOINC has a tendency to ask for GPU tasks, even with the appropriate options checked in the BOINC manager's settings AND the online project settings. that was never a problem though, b/c after asking for GPU tasks and failing to get any, it would ask for CPU tasks and would get them. as of late its only been asking for ATI GPU tasks and (failing to get them b/c i don't have a capable nVidia GPU) and not asking for CPU tasks at all. when i force a project update, it tells give me the following:
Quote:
9/6/2011 10:00:03 PM | Milkyway@home | Sending scheduler request: Requested by user.
9/6/2011 10:00:03 PM | Milkyway@home | Not reporting or requesting tasks
9/6/2011 10:00:04 PM | Milkyway@home | Scheduler request completed
now i've been running E@H successfully on the CPU for months on end without a hitch, even in conjucntion with another BOINC project Test4Theory@Home, splitting the BOINC resources 50/50 between them. i haven't changed any of the BOINC manager settings, nor have a made any major changes to my online project settings recently. just to be sure it didn't have anything with my integrated HD 3300 ATI GPU, i slipped a cc_config.xml file into the BOINC data directory with a directive to ignore GPU 0 (b/c its the only GPU in the host), which BOINC successfully ignored per the startup portion of the event log. after starting BOINC again, i noticed that not only was it no longer asking for GPU tasks, but it still wasn't asking for CPU tasks - it wasn't requesting any tasks at all, and every time i manually forced a project update, i would get the:
Quote:
9/6/2011 10:00:03 PM | Milkyway@home | Sending scheduler request: Requested by user.
9/6/2011 10:00:03 PM | Milkyway@home | Not reporting or requesting tasks
9/6/2011 10:00:04 PM | Milkyway@home | Scheduler request completed
i've selected to run all E@H applications on this host, and the "Run CPU versions of applications for which GPU versions are available" check box is checked. i have "use nVidia GPU" set to no, and obviously i have "use CPU" set to yes. so i'm completely baffled as to why i've had no problems fetching CPU work for the past several months, haven't made nay major changes to either my BOINC or online settings, and all of the sudden my host doesn't want to fetch CPU work.
anyone have an idea what might be going on? if more information is needed, please specify and i'd be happy to oblige.
TIA,
Eric
"Not reporting or requesting tasks"
)
Some of the problems I saw on my laptop suggest that you'll have to do one of the following:
1. Enable GPU use, and then connect to a project that can use an AMD/ATI GPU, regardless of whether that project does anything you care about. Download at least one AMD/ATI workunit from them. Once you download enough workunits, this should avoid the problem until either those workunits run or they reach their deadlines.
2. Find a different method of disabling the GPU.
If BOINC is trying to use the GPU, it will not go on to requesting CPU workunits until it downloads enough GPU workunits (not necessarily from any specific project) to keep all your GPUs busy.
If you like medical research, POEM@HOME appears to be preparing to do something useful with AMD/ATI GPUs, but only after a future BOINC version adds enough support for OpenCL GPU workunits.
http://boinc.fzk.de/poem/
If you prefer math, two are available:
http://boinc.thesonntags.com/collatz/
http://www.primegrid.com/
I don't recall any more - but since I don't have that type of GPU, I haven't looked very hard.
Some of the more recent versions of BOINC (such as 6.12.33) offer a quick way of disabling them from running GPU workunits, but one that does not fully disable downloading them.
Go to the advanced view, click on Activity, then click on Suspend GPU.
thanks for the suggestion,
)
thanks for the suggestion, but i'm looking for a bit more than a workaround or a temporary solution. the thing is, i've run for months on end with my settings the way they are and have never had a problem until the last day or two.
exactly...which is why i never had a problem for months on end and shouldn't have one now - given that i haven't a capable nVidia GPU, the number of GPU tasks required to keep a nonexistent GPU busy is and always has been zero...at which point BOINC would then ask the E@H project server for CPU tasks. sure, it always asked for GPU tasks first, but when it realized that my host couldn't crunch GPU tasks, it never had a problem subsequently fetching CPU tasks. why all of the sudden would it not ask for CPU tasks at all? not to mention that if the cc_config.xml file with a directive to ignore the GPU was successful in the sense that BOINC no longer requested GPU tasks from the E@H server, then it should have strictly asked for CPU tasks. what gives?
even if i did resort to your suggested solution, it wouldn't work for me - my ATI GPU is an integrated HD 3300, and i don't think its powerful enough for any DC projects. besides, i use it solely to run the display so that the CPU can handle all the crunching (hence no GUI lag).
Hello. I had the same problem
)
Hello. I had the same problem in past. This was because of hardware failure on my side when I suddenly trashed all my queue. Server then said me "Reaching daily quota" and lowered it down to 2 per core. But this is not your case right now. All you have to do is to start lowering your work cache by changing the following:
1) Go to the "Tools"->"Computing preferences"
2) Change "connect about every ... days" parameter. Lower it by one day.
3) You may also try in parallel to chage "Additional work buffer" by lowering it down too.
It is strange. But it helped me many times. The lowest number I tried when BOINC suddenly starts to ask for WUs was 0.01 for both parameters.
Let me know if it doesn't help. Remember to wait for a minute after each server request else it will reject your request saying "last request too recent".
thanks for the suggestion,
)
thanks for the suggestion, but before i even had a chance to try changing some of my settings, i woke up this morning to find that BOINC downloaded 60 new CPU tasks around 3:50am...go figure. i don't think the queue ever went completely dry...that is to say, i think the new work got downloaded just before i actually ran out of CPU work. i'm not sure how to explain this, as i've always been able to maintain a decent sized cache according to my settings...but this time it seems like i almost had to run out of tasks in order to get more...
Yes, sometimes when you think
)
Yes, sometimes when you think the queue is completely drained out it suddenly downloads a new bunch of work enough for say a week. The cause is DCF (duration correction factor). It can be changed manually but rarely needs to be corrected, because BOINC since version 6 or so corrects it (when it is not correct like in your case) right when you run out of work. So you should look for the host for about a week. There may be some more moments when cache will grow up and then (not fully) drain out. It is normal.
RE: ...So you should look
)
well that's the thing - i'm able to monitor this host on daily basis...several times a day in fact. and i've never seen the cache almost completely drain out like this during the better part of a year. in fact, this is the first time i've seen it happen after almost a year of flawless crunching & cache filling. i'm semi-familiar with the DCF. what confuses me is how the DCF could have changed so drastically in such a short period of time to have caused this startling event on my host. when a host has been crunching tasks of a particular project constantly and consistently for months on end, the DCF should have converged on a specific value after such a long period of consistent crunching has passed (in an effort to keep tasks flowing constantly and consistently). why the sudden change, i don't know...
lol i have the complete
)
lol
i have the complete opposite i want only gpu units but in a month i had a few thats it Server just reporting no work available for weeks in a row
then ofcourse i turn onto other projects with cuda but almost all which i favor have issues with delivering units (seti@home, distrRTgen) gpugrid is not an option cause the cards are too slow for such heavy project.
I really not want units on my cpu cores cause is way too slow, and heats up my machine too much.
I have noticed the same thing
)
I have noticed the same thing as sunny129. I will have my requested cache size (say 10 work units) and it will crunch all 10 of them right until I need more (there is only 1 or 2 left running and almost done) then download 10 new ones, rather than keeping 10 work units in queue constantly.
RE: when i force a project
)
Note that the message is about Milkyway at Home . not Einstein at Home. Actually the reason why BOINC will not fill the queue for some of your projects might be problems to get work for Test4Theory@Home. Even tho you assigned to it a high resource share (50:50?), it doesn't seem to get much work lately:
So it would be only reasonable for BOINC to try desperately to reserve some share of the pipeline for this project to finally catch up, and only get work from E@H at the last moment if it runs a risk to have no work at all.
Heinz-Bernd
RE: Note that the message
)
my mistake...i accidentally copied and pasted from the wrong section of the event log. but rest assured that i got the same result when i manually updated Einstein@Home as well.
hmm...i hadn't thought about that yet. but is it not strange that i've been running T4T@H for a whole month now along side E@H without any E@H cache problems until very recently?
by the way, my cache seems to be all over draining again...it isn't the end of the world if it continues to do this, so long as it refills the cache before i actually run out of work. but what happens when i start to run low and the E@H servers go down right before my BOINC client goes to fetch more CPU work? then i'm SOL.