Pending tasks
7 May 2017 0:53:24 UTC
Topic 207383
(moderation:
Has anyone else noticed an increase in pending tasks? The reason I ask is in the last 2 days my pendings have gone from a typical 70 or so to over 150. Only 4 are inconclusive so I'm pretty sure my boxes are OK. I know pendings go up and down but this is very abnormal.
Something is realy wrong,
)
Something is realy wrong, besides the pendings, if you look the daily production of the Top hosts all down about 40%, that can´t be just a coincidence.
That is consistent to what
)
That is consistent to what I'm seeing.
There's really no need to be
)
There's really no need to be alarmist about it.
You can't really draw that conclusion. It's pretty clear that there has been some sort of drop in the rate at which results are being validated. It's likely that is just the same thing as the increase in pendings. Perhaps there is some blockage in the pipeline that gets returned results to the validator. There is no backlog showing on the server status page so it is perhaps before the point where queued numbers are measured. I imagine at some point someone will come along and release the blockage :-). This is not a major crisis (yet) :-). It seems to be a drop in apparent production which, later on, will probably turn into a spike in apparent production. I think we should just wait calmly until the Devs resolve the issue.
I don't know where you get the 40% from - I haven't even tried to check. In my case, my fleet has shed about 7% so far which seems to be just a bit less than 40% :-).
Cheers,
Gary.
This is not a major crisis
)
There's really no need to be
)
My apologies if my bad english leave you to that conclusions, i was not in <Panic> mode, just showing another side of the problem.
Maybe that´s because i look at the daily production of the host, not the RAC. Yes the RAC drops a little (can´s say how much i don´t follow the RAC numbers).
What i can tell is what the numbers said... few examples could easely find in the Free DC last stasts.
On the top of the list, the E@H top host is a 4 MM/Day crucher, it´s yesterday´s production was about 2 MM.
My own main cruncher it´s a 1.8-2MM day and yesterdays production was in the range of 1.3 MM.
The same is happening with most of the hosts of the list if you exclude of course the new rising hosts who´s averages are not acurate.
Do the single math and you will see the aproximately 40% i was talk about. But seems like some hosts are more afected than others. Why i have no ideia.
Please be sure not was my intention to be alarmist, rise any problem or questions, just reporting something i was looking.
That was my mistake, i forget i was in an open (not our team) thread with people who dont´t know my dificulties to sometime choose the right words in english.
Wish you all a good weekend and happy crunchings.
Gary Roberts wrote:I don't
)
Gary, as it happens I daily update a spreadsheet which gives me daily credit award rate for the last day and the last five days.
In the wonderful days when the website provided summary counts of pending jobs I also maintained an alternate column which added in the pending work, which aids detecting problems in my systems by smoothing out variation just attributable to luck of quorum partners. But that has not been available for about nine months.
Watching for unexpected changes in these results is a primary tool to alert me to problems in my small fleet, but it also can show me trouble at home base.
Based on numbers from that spreadsheet, I support Juan's observation of roughly a 40% reduction in credit award rate in the last couple of days. To be specific, I've been running at about 2,000,000/day, and the last two daily lines were first 1,293,208, then 1,214,236. These are wildly out of pattern, so something is going on somewhere. As all three of my hosts are similarly affected, it seems likely that "home base" rather than my systems is the somewhere.
There are currently two
)
There are currently two things happening. One is that EinsteinAtHome is a BOINC Pentathlon project which means that a lot of users are downloading work but not reporting it back (known as bunkering). The number of GPU in progress tasks has doubled over the last days which means the throughput declined). This means there is not a lot of validation happening for GPUs right now until next Tuesday when the challenge starts and Penathlon users are reporting those tasks back.
The other thing is the new gravitational wave testrun which lacks windows support right now. The high rate of pending there is expected and will solve itself once the windows app is active and more users are able to crunch gravitational wave tasks.
juan BFP wrote:Quote:There's
)
There's no need to apologise and your English is fine. If anyone should apologise, it is me :-).
I'm in the middle of commissioning a 34kW solar array (100 x 340W panels) and was concerned that statements like, "there must be other things as well as just an increase in pendings" might lead a whole lot of people to jump to unwarranted conclusions and start making a lot of noise like, "get this fixed now or I'm leaving", sort of thing. I was just keen to nip any such reaction 'in the bud' as I don't have time to get distracted at the moment. So I'm sorry for over-reacting and I really wasn't intending to 'pick on you'.
I'm fully aware of how slow RAC is to react to a sudden change. My RAC had dropped 7% so I did realise that the instantaneous drop would be a lot bigger. I was hoping the 'blockage' might be fairly quickly fixed so that most people probably would not have even noticed. A statement of "40% drop" does tend to grab attention however :-).
Now that Christian has explained the cause, we can all relax. I've never participated in any of these 'challenges' or whatever they call them. I know nothing about them and I really don't want to know either :-). It's the first time I've heard this term 'bunkering'. I understand why it's an appropriate term - a coal or oil fired power plant taking on fuel :-). However it seems to me that 'bunkering' is not really a good analogy. Taking on fuel doesn't give a power plant the ability to produce a greater instantaneous output. Pre-crunching tasks and releasing them only when they can be measured seems to be an artificial way of making your system appear to be more powerful than it really is. I really don't get the point of that?? It must play havoc with the orderly running of a project.
Cheers,
Gary.
BTW: soon can be a
)
BTW: soon can be a rubberband! :-(
https://einsteinathome.org/content/report-web-site-problems-here?page=5#comment-155007
Greetings from the North
Christian you seem to have
)
Christian you seem to have hit the nail on the head. Now that that the disruptive BOINC Pentathlon has moved over to this project My RAC is starting to go up and the pendings going down.