On Wed, 11 Apr 2001, Lennart Augustsson wrote:
> That looks very interesting.  Was that test with our without softdep?

I was using softdeps at the time.

> I think we should definitely have a look at this and bring it in if at all
> possible.  The time it takes to untar (and rm) a directory tree is
> annoying in NetBSD at the moment.

I'm playing with another idea right now wrt ffs_dirprefs(), and I'll post
test results for it as well if they're any better than the results I have
now.

> Still, there seems to be room for huge improvments.  Look at those
> rm figures, why is the kernel waiting for the disk so much?

Haven't got a clue.. I've always found 'rm' speeds on NetBSD slow -- I did
a test of NetBSD on this K6-3 vs. FreeBSD on a 486DX/4 75 and the 486 beat
the pants off of the K6-3 in an 'rm -rf' test I did (please see
http://invisible.yi.org/~jmcneill/netbsd/benchmark.txt for the numbers).

Jared