[Bug 246940] [wishlist/enhancement, patch incl.]: idle user tasks should be charged as "nice" or "user" CPU time

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=246940

--- Comment #4 from t.eichstaedt at gmx.net ---
> Why is it counter-intuitive that user processes are counted in user CPU time?
When they are neither "nice" nor "idle", it's not.
But when they are in the idle class, IMHO a mortal user would expect them not
to be charged as user CPU time, since their scheduling priority is even lower
than all nice tasks.

> "Idle" is for kernel idle threads, nothing else.
Then what's the idle _user_class_ for?

> Low-priority user programs can be charged as "nice"
So this means I have to apply both "nice" and "idprio".  OK, but IMHO one would
expect that idle user tasks are implicitely also (at least) nice, same
argument: because their scheduling priority is lower than all nice tasks.

I totally agree to a bit of resistance to introduce new sysctl's and in
general, accepting patches, no matter how trivial they are.

In this case, however, I'd kindly suggest the issue that currently, idle user
tasks do affect the CPU freq scaling (becasause they are charged as "user"
time) and with this patch, have an effect that I (and presumably many others)
consider useful, i.e. one can run tasks as idle and know that they do not
affect the CPU freq scaling - like the idle kernel task(s).

This is at least useful for mobile devices running AC-offline (or
solar-powered).  It's useful for desktops, one can switch between
fullpower(noisy) and quite mode for background tasks.
It can be even useful for servers to do such a switch on power-outage.
The patch is non-intrusive, since the default is to retain the current
behaviour.

-- 
You are receiving this mail because:
You are the assignee for the bug.