[Bug 173541] High (0.60+, 1.00) idle load averages
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=173541 --- Comment #46 from Alexander Motin --- It is a long known problem of very idle systems. To save the power kernel tries to group different timer events together to reduce number of CPU wakeups from idle state. The problem is that load average accounting also based on timer events, and if system is idle, each time it fires, there is often something else to do, that is getting accounted into load average. -- You are receiving this mail because: You are the assignee for the bug.
[Bug 173541] High (0.60+, 1.00) idle load averages
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=173541 --- Comment #45 from Viktor Štujber --- fwiw the dualcore HT atom minipc I originally reported this for has been showing realistic averages (0.01 to 0.10) since my last post around 10.2 in late 2015, and has remained that way until now, running 13.1. Here's the system vars that might be affecting processor time accounting. kern.timecounter.choice: TSC(1000) ACPI-fast(900) HPET(950) i8254(0) kern.eventtimer.choice: HPET(450) HPET1(440) HPET2(440) LAPIC(100) i8254(100) RTC(0) dev.cpufreq.0.freq_driver: p4tcc0 dev.cpu.0.freq_levels: 1800/-1 1575/-1 1350/-1 1125/-1 900/-1 675/-1 450/-1 225/-1 dev.cpu.0.cx_method: C1/mwait/hwc dev.cpu.0.cx_supported: C1/1/3 -- You are receiving this mail because: You are the assignee for the bug.
[Bug 173541] High (0.60+, 1.00) idle load averages
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=173541 --- Comment #48 from Alexander Motin --- I am not sure what he means. There could be some optimizations, but I am not sure it can be easily fixed, unless we implement some mechanism to somehow strictly prevent timer other events grouping with the load average ones. -- You are receiving this mail because: You are the assignee for the bug.
[Bug 173541] High (0.60+, 1.00) idle load averages
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=173541 --- Comment #50 from Mark Millard --- (In reply to Alexander Motin from comment #46) Quoting your #9 : . . . if system is idle, its CPU may wakeup as low as two times per second. But, the load average code is based on fixed 5 second sampling and then a small variability is tolerated/introduced to avoid matching a (nearly) fixed frequency activity accidentally matching. The "fixed" turns into a fixed number of adjustments, such as 12, 60, or 180. Having anything like 30sec between steppings of the load average adjustment, spreads the 12 steps case (60 sec @5 sec each intended) out to something like 360 sec total for the 12 steps. (Also has implications for widely variable time interval sizes.) If that is really the case, not much else can matter as long as the 5 sec interval is not well approximated and calling the average a "1 minute" average can be highly misleading. Descriptions that would be accurate are "12 step load figure", "60 step load figure", and "180 step load figure". No fixed conversion to time intervals available --and not really arithmetic means. Sound right? Note: For what I use the "load averages" for, the "N step load figure" interpretation works fine without knowing specific times. Inaccuracies in small figures (relative to the hardware thread count involved) make no noticeable difference. Nor do small percentage errors in the larger figures: Rule of thumb material for making some judgments that are not highly sensitive to the details. -- You are receiving this mail because: You are the assignee for the bug.
[Bug 173541] High (0.60+, 1.00) idle load averages
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=173541 --- Comment #52 from Mark Millard --- (In reply to Stavros Filargyropoulos from comment #51) Good question: bad substitution in my head that I did not notice, despite "second" being in what I quoted. So Ignore my #50. Sorry for the noise. Too bad I can not delete to avoid my confusing anyone besides myself . . . -- You are receiving this mail because: You are the assignee for the bug.
[Bug 173541] High (0.60+, 1.00) idle load averages
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=173541 --- Comment #49 from Stavros Filargyropoulos --- (In reply to Alexander Motin from comment #48) Thanks Alexander. Gives me some more confidence to know the root cause and understand that it shouldn't functionally affect the system in any other ways. -- You are receiving this mail because: You are the assignee for the bug.
[Bug 173541] High (0.60+, 1.00) idle load averages
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=173541 --- Comment #51 from Stavros Filargyropoulos --- (In reply to Mark Millard from comment #50) > 30sec Where did you get the 30secs from? He said twice per second. Not per minute? -- You are receiving this mail because: You are the assignee for the bug.
[Bug 173541] High (0.60+, 1.00) idle load averages
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=173541 --- Comment #47 from Stavros Filargyropoulos --- (In reply to Alexander Motin from comment #46) Alexander, in the comment above g...@freebsd.org suggested that this has been fixed. I tried to have a look through the recent commits in ACPI but can't seem to find anything that would have fixed this. Do you happen to know commit was Gordon referring to? -- You are receiving this mail because: You are the assignee for the bug.
