Subject: Re: Preparing callout(9) for a HZ-less kernel
To: Jason Thorpe <firstname.lastname@example.org>
From: Simon Burge <simonb@NetBSD.org>
Date: 10/24/2007 02:20:31
Jason Thorpe wrote:
> On Oct 23, 2007, at 5:10 AM, Joerg Sonnenberger wrote:
> > For callout_stop, I would expect it to be a relative rare operation,
> > so
> > optimising that doesn't matter. callout_reset/callout_schedule
> > happens a
> > lot more often, but I don't think it is performance critical either.
> > It
> > happens a lot to extend the trigger period and it should be analyzed
> > if
> > a lazy update of the tree is more efficient. I doubt it though.
> callout_stop() is actually a fairly frequent operation. It's used to
> detect timeouts on I/O operations for e.g. SCSI and ATA commands.
Just for kicks, here some event counters on a Core Duo laptop. This is
after a kernel had finished building after a reboot. The system had
been up for about 8 minutes at this stage.
event total rate type
callout late 9 0 misc
callout ack 10 0 misc
callout active 13865 33 misc
callout destroy 11366 27 misc
callout init 11660 27 misc
callout invoking 47 0 misc
callout pending 113 0 misc
callout reset 83970 201 misc
callout schedule 46165 110 misc
callout setfunc 11467 27 misc
callout stop 101361 243 misc