> > A recent thread in comp.arch discusses context switch times for various
> > operating systems on modern hardware.  Someone who works at QNX just
> > posted a list of measurements he's taken; it shows context switch times
> > of 210ms and 380ms for 386BSD 0.1 and NetBSD on the same hardware.
> ...
> (3) our switch does a fair amount more, in terms of keeping more
> 	careful process accounting information, than the 386BSD
> 	switch() did.  In particular, mi_switch() calls microtime()
> 	twice, and that's not a cheap function.

Well, as it turns out, there's really no need to be defensive about
it. :-)

First, note that the NetBSD entries in the results list did not have a
version number attached to them.

I just went through my mailbox and dug out some results run with the
ctxsig.c program posted by lm@sun.com (?) on comp.arch about a year ago.
At that time we ran the test on a i486dx/2-66 with NetBSD 0.8, and got
the reported 380ms context switch time.  I just reran the test program
on *identical* hardware with NetBSD 1.0_BETA (as of a few days ago), and
the context switch time on this platform has gone down to 110ms.

Well done!

- Havard

------------------------------------------------------------------------------