Disproportionate amount of workunits

tullio
tullio
Joined: 22 Jan 05
Posts: 2118
Credit: 61407735
RAC: 0

I am running 6 projects, +LHC

I am running 6 projects, +LHC when available , all equally shared. My cache is 0.25 days. I get a new WU when the former has finished and I am never out of work. My CPU, an Opteron 1210 running Linux at 1.8 GHz, has two cores and runs 2 projects simultaneously. Only AQUA, which is multithreading, grabs all two cores, but I rarely get AQUA units. I have many pending units in SETI and a few ones in Einstein. The other projects have all a quorum of 1.
Tullio

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5883
Credit: 119076347914
RAC: 24100430

RE: Before I set it to NNT

Message 98685 in response to message 98683

Quote:
Before I set it to NNT on 7/30, E@H loaded Computer ID: 995651 up with over 100 WUs for S5GCESSE2. The work buffer is set to 1.0 day. It can process about 5 per day, so something is clearly amiss.


Yes, something is indeed amiss, and that something is most likely BOINC's inability to properly manage the cache of work for your GPU in addition to the cache of work for your CPUs. You have a CUDA capable card that needs ABP tasks whilst any GW tasks need to be run on the CPU only. I suspect that at the time you mention above, your GPU had probably run dry and BOINC was repeatedly attempting to get more ABP tasks without taking proper regard of the large number of GW tasks that were being sent.

The scheduler is programmed to favour GW tasks since that is the main focus of the project. So it's not surprising that when requested for work, the scheduler will very often give only GW work. Also, on 29 July, ABP work was discontinued for about 2 days whilst the new 'anti-center' ABP search was brought into play - see the tech news forum for details. So I'm guessing that when your machine was trying to get more ABP work for your GPU, there was a temporary shortage and BOINC went beserk with GW downloads.

Quote:
It will grind through about 25 of the remaining 80+ WUs before the deadline. Once the report deadlines have passed and it has done all it can, what should I do, if anything? Reset? I don't want to again have a huge backlog of WUs that won't be processed in time.

Please DON'T reset. That is entirely unnecessary. All you need to do (based on your figures - you might like to double check your calculations) is abort about 50 to 60 tasks, leaving about 25 or so, if that is what could be completed by the deadline. You should do this sooner rather than waiting for the deadline so that the tasks can be reissued quickly to other crunchers. Also, there is usually a bit of a delay before tasks get reissued so don't worry if the last of the tasks you decide to keep actually overrun the deadline by several hours. They will still be accepted.

Because the shortage of ABP was a rather unusual occurrence, hopefully BOINC wont get the opportunity to be so profligate in the future.

Cheers,
Gary.

Jord
Joined: 26 Jan 05
Posts: 2952
Credit: 5893653
RAC: 0

RE: The work buffer is set

Message 98686 in response to message 98683

Quote:
The work buffer is set to 1.0 day.


I see in the connection logs that your "connect to server" value is set to one day (86400 seconds). What's your Additional work buffer set to, though? And are you setting that one on the web-site, or in the local advanced preferences? These latter will override the web-site preferences.

glohr
glohr
Joined: 27 Aug 05
Posts: 5
Credit: 280104111
RAC: 156492

RE: What's your Additional

Message 98687 in response to message 98686

Quote:
What's your Additional work buffer set to, though? And are you setting that one on the web-site, or in the local advanced preferences?


Set locally to 1.00 day. Settings have not changed for several years.

Olaf
Olaf
Joined: 16 Sep 06
Posts: 26
Credit: 190763630
RAC: 0

Well, similar problem, maybe

Message 98688 in response to message 98687

Well, similar problem, maybe related to the change of the sheduler?
One day it took as much work units as possible for GPU (deadline 2010-08-16)
until the server stopped to send more work. Not sure, if they will get ready
before deadline, especially because the behaviour changed too - before I had at
the same time 7 units with CPU and one with CPU+GPU, this was switched at the
same time to 8 units for CPU and one for 0.41CPU+GPU. Because there are only
8 CPUs on an i7, it is difficult to estimate, how long it will take for the
GPU tasks. Estimated times are still funny for CPU units - around 100 hours
instead of realistic 7. These for the GPU tasks seem to be more realistic now
however, but I think, this is the computing time now, not the real time, which
can be much longer due to the problem, having sometimes 9 task for 8 CPUs. The
computer notes a current workload enough for about 9.5 CPUs just for e@h ;o)

glohr
glohr
Joined: 27 Aug 05
Posts: 5
Credit: 280104111
RAC: 156492

Thanks, the explanation is

Message 98689 in response to message 98685

Thanks, the explanation is helpful. I've aborted the extra WUs. Once the queue had drained, I'll put everything back to normal and see what happens.

Henri Ala-Peijari
Henri Ala-Peijari
Joined: 11 Sep 05
Posts: 5
Credit: 6428544
RAC: 0

RE: [ Let us know how you

Message 98690 in response to message 98682

Quote:
[
Let us know how you get on.


I decided to run seti@home with a 10 day cache. But since I have three dual core intel atom machines (with hyperthreading it's 4 cores), I might put one of those into einstein@home work. I did buy those boards with experimentation in mind since they're pretty much a PC for 80euros + PSU,memory and reused old housing and hdd.

Comment viewing options

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