I only know of the related commit number as: c308b56b5398779cd3da0f62ab26b0453494c3d4
I would argue that the commit results in more accurate reported load averages, overall.
One has to look at the reported load averages over the entire range of operation. Meaning over all frequencies of cpu's entering and exiting the idle state, and all per cpu loads. An idle desktop environment is but one point in that two dimensional space, and yes, it seems to be too high.
Without the commit code changes, all cpu enter/exit idle frequencies above 25 hertz will result in a reported load average of 0.00, and will scale linearly from 0.00 at 25 hertz to accurate at 0 hertz. (for systems with a basic 250 hertz clock, which is the default)
Reference: https://launchpadlibrarian.net/93077784/load_freq_s15.png
With the commit code changes, and for medium to higher actual load averages the results are much better.
I am working on aquiring more data for low actual load situations, both with control samples using kernels compiled with CONFIG_NO_HZ=n (i.e. the old tick way) and for CONFIG_NO_HZ=y (the newer tickless way) with and without the commit. Reported load averages are very noisey, and take a long time to settle for given conditions, so each test takes a long time. I will post here or on the above referenced bug when done (it'll be a few days).
I only know of the related commit number as: c308b56b5398779 cd3da0f62ab26b0 453494c3d4
I would argue that the commit results in more accurate reported load averages, overall.
One has to look at the reported load averages over the entire range of operation. Meaning over all frequencies of cpu's entering and exiting the idle state, and all per cpu loads. An idle desktop environment is but one point in that two dimensional space, and yes, it seems to be too high.
Without the commit code changes, all cpu enter/exit idle frequencies above 25 hertz will result in a reported load average of 0.00, and will scale linearly from 0.00 at 25 hertz to accurate at 0 hertz. (for systems with a basic 250 hertz clock, which is the default) /launchpadlibra rian.net/ 93077784/ load_freq_ s15.png
Reference: https:/
With the commit code changes, and for medium to higher actual load averages the results are much better.
I am working on aquiring more data for low actual load situations, both with control samples using kernels compiled with CONFIG_NO_HZ=n (i.e. the old tick way) and for CONFIG_NO_HZ=y (the newer tickless way) with and without the commit. Reported load averages are very noisey, and take a long time to settle for given conditions, so each test takes a long time. I will post here or on the above referenced bug when done (it'll be a few days).