Re: Revising CPU usage calculations

SODA Noriyuki <sodasra.co.jp> wrote:
> Is there any reason that autonice is SCHED_4BSD only?
Different schedulers may do that differently. 
At least, this should be per-LWP and with better heuristics
for I/O and
CPU-hog.

At least, autonice must go into the scheduler specific part,
but I think it
could go out at all.

-- 
Best regards,
Mindaugas
www.NetBSD.org

>>>>> On Wed, 28 Mar 2007 15:18:10 +0300, 
      Mindaugas R." <rmindNetBSD.org> said:

> At least, this should be per-LWP and with better
heuristics for I/O and
> CPU-hog.

That's questionable.
For example, one of the problems of autonice is it makes X
server
slower, and even per-LWP something cannot solve the
problem.
The job that X server is doing is a CPU thing (e.g. font
scaling) from
scheduler's viewpoint, but it is an I/O thing from users'
view point.

> At least, autonice must go into the scheduler specific
part, but I
> think it could go out at all.

I rather think the feature itself could be removed.

Is there any one who is still using autonice even in
NetBSD?
Note that our default is off.

Apparently FreeBSD people don't use the feature in this
decade.
-- 
soda

On Wed, Mar 28, 2007 at 03:18:10PM +0300, Mindaugas R.
wrote:
> SODA Noriyuki <sodasra.co.jp> wrote:
> > Is there any reason that autonice is SCHED_4BSD
only?
> Different schedulers may do that differently. 
> At least, this should be per-LWP and with better
heuristics for I/O and
> CPU-hog.
> 
> At least, autonice must go into the scheduler specific
part, but I think it
> could go out at all.

Well, as long it is a sysctl tunable and disabled by
default, I don't
think it is really neccessary to put autonice into the
scheduler-dependent part.

However, is anyone strongly objecting against dropping
autonice
completely? If not, I'd like to go this way.

Regards,
Daniel

-- 
Daniel Sieger
Faculty of Technology
Bielefeld University
wwwhomes.uni-bielefeld.de/dsieger