The data file LATeah3012L09.dat didn't last very long. It's been replaced with LATeah4021L00.dat. We've had the 4000L series before - the most recent one was LATeah4013L04.dat in Dec 2021.
I seem to recall that there was a crunch time difference. Overall the 4000L series used to take a bit longer, particularly for the main calcs (0 - ~90%). The followup stage was quite quick - a small fraction of what the 3000L tasks have been taking lately. For my older hosts, the net result for both these things was a total crunch time about 5-10% longer.
It might be different this time so we'll have to wait and see.
The task shortage seemed to end with the last 3012L12221101 tasks - those with a frequency component of 900Hz. After a couple of hours, 3012L09 tasks started flowing freely. The first tasks started with 732Hz and continued up to 772Hz.
The 3012L09 tasks had started back in May and my memory is that they finished abruptly with a frequency in the low 700s - probably 724Hz. It seems we are just finishing off these tasks at higher frequencies - probably all the way to 900Hz eventually.
My guess is that the task shortage is being alleviated by running two separate task generation streams, one for 3000 series tasks and one for 4000 series tasks. While one is being distributed, the other is building up a stock so that the available tasks never fully deplete.
The very latest 3012L09 tasks started at 780Hz - the next frequency point above the previous end (772Hz), presumably when the stock of 4021L00 had depleted. That series had tasks starting as low as 4Hz and eventually getting up to 276Hz. My guess is that once the current 3012L09 tasks deplete (somewhere around 804 - 828Hz, there will be a switch back to 4021L00 starting at 284Hz.
These are just guesses - maybe Bernd might be kind enough to give us the full story :-).
I don’t know if Bernd posted it publicly in such detail, but he told me via PM that the work shortage was due to pre-processing bottlenecks caused by the energy shortage in Europe. Some servers were offline or intermittently working and not processing tasks at a high enough rate. This limited overall WU production for us to a rate lower than we were processing.
Quote:
Actually we're having trouble producing (i.e. pre-processing) enough FGRPB1G work. Some machines went down due to the unexpected power outage in our datacenter a month ago, and probably won't be revived for some time due to the energy crisis here in Europe.
Bernd had said at the time that he thought the situation would improve around the beginning of November, but it just took a little longer in reality.
If you don't, you'll just
)
If you don't, you'll just keep getting what the server decides to send.
Cheers,
Gary.
So, assuming the flood of
)
So, assuming the flood of grp#1 tasks continue how long before races "recover".
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Assuming you meant to say
)
Assuming you meant to say “RAC”.
6 weeks
_________________________________________________________________________
The data file
)
The data file LATeah3012L09.dat didn't last very long. It's been replaced with LATeah4021L00.dat. We've had the 4000L series before - the most recent one was LATeah4013L04.dat in Dec 2021.
I seem to recall that there was a crunch time difference. Overall the 4000L series used to take a bit longer, particularly for the main calcs (0 - ~90%). The followup stage was quite quick - a small fraction of what the 3000L tasks have been taking lately. For my older hosts, the net result for both these things was a total crunch time about 5-10% longer.
It might be different this time so we'll have to wait and see.
Cheers,
Gary.
looks like the 4000-seies was
)
looks like the 4000-seies was the short lived one. getting mostly 3012L09 again.
_________________________________________________________________________
The task shortage seemed to
)
The task shortage seemed to end with the last 3012L12221101 tasks - those with a frequency component of 900Hz. After a couple of hours, 3012L09 tasks started flowing freely. The first tasks started with 732Hz and continued up to 772Hz.
The 3012L09 tasks had started back in May and my memory is that they finished abruptly with a frequency in the low 700s - probably 724Hz. It seems we are just finishing off these tasks at higher frequencies - probably all the way to 900Hz eventually.
My guess is that the task shortage is being alleviated by running two separate task generation streams, one for 3000 series tasks and one for 4000 series tasks. While one is being distributed, the other is building up a stock so that the available tasks never fully deplete.
The very latest 3012L09 tasks started at 780Hz - the next frequency point above the previous end (772Hz), presumably when the stock of 4021L00 had depleted. That series had tasks starting as low as 4Hz and eventually getting up to 276Hz. My guess is that once the current 3012L09 tasks deplete (somewhere around 804 - 828Hz, there will be a switch back to 4021L00 starting at 284Hz.
These are just guesses - maybe Bernd might be kind enough to give us the full story :-).
Cheers,
Gary.
I don’t know if Bernd posted
)
I don’t know if Bernd posted it publicly in such detail, but he told me via PM that the work shortage was due to pre-processing bottlenecks caused by the energy shortage in Europe. Some servers were offline or intermittently working and not processing tasks at a high enough rate. This limited overall WU production for us to a rate lower than we were processing.
Bernd had said at the time that he thought the situation would improve around the beginning of November, but it just took a little longer in reality.
_________________________________________________________________________