On Tue, Aug 26, 2003 at 12:21:34AM +0900, YAMAMOTO Takashi wrote:
> i did some benchmarks. (attached)

The regular file mapping improvement looks pretty good, and pagefault
overhead is also reduced. This should help with the most common case
by far: paging in executables.

I wonder why the /dev/zero case is slower. I guess there's not much
algorithmic improvement possible there; it's probably the worst case
for the rbtree algorithm. But, it's also an uncommon case, so the
regular file map speedup seems to clearly outweigh that one.

- Frank

-- 
Frank van der Linden                                            fvdl@netbsd.org
===============================================================================
NetBSD. Free, Unix-like OS. > 45 different platforms.    http://www.netbsd.org/