* Thor Lancelot Simon <tls@rek.tjls.com> [041211 15:03] wrote:
> Profiling data from the kernel on www.netbsd.org, covering a period of
> a few days beginning just after the 2.0 release, can be found at
> http://www.panix.com/~tls/nbwww_gprof.txt
> 
> There aren't really any major surprises.  The heaviest hitter turns out
> to be pvtree_SPLAY, called from the pmap code of course.  One thing I
> am wondering is if the splay routines could be optimized with MD code?

That makes sense.

Another possible optimization, I always wondered if not splaying when
the depth traversed is below some threshhold, say 2 or 3, that would
compromise a working set and reduce the write traffic to memory caused
by balancing the tree.

-- 
- Alfred Perlstein
- Research Engineering Development Inc.
- email: bright@mu.org cell: 408-480-4684