> > You've got some apples to oranges here.
>
> Undoubtedly...  However, the "high-level" fact that we perform
> much worse than an ext3-using Linux installation on identical
> hardware remains, and I'm trying to see if there is some way we
> can improve on the NetBSD performance.

Sure. You are comparing out-o-th-box experiences, which is
a good thing. My point, I suppose, is that if NetBSD's out-o-th-box
is no softdep -o async, it would be closer to ext3 out-o-th-box.
Then, they both soundly concede stable storage and look fast.

>
> I've not yet brought NFS into the picture, but interestingly,
> yes, that is eventually what we're looking at in this context as
> well.  Your comments are noted.

It's not about NFS. It's about stable storage rules. What
constitutes stability? The NFS folks, www.spec.org, have had
an answer for over a decade.

>
> > If you mount ext3 with -o sync,data=journal, you'll find FFS faster.
>
> I'll get the exact ext3 parameters used tomorrow.  Do you
> indicate that this set of ext3 parameters make it more similar to
> what FFS does?  (Note, I've run with soft dependencies turned on.)

No. Journaling and softdep are two different animals. I agree
that the directory hashing is a big part of the perofmance difference
you see. But the directories have nothing to do with the difference
in read/write throughput.

>
> > FWIW, I don't use the namei cache at all with my file system,
> > which uses hashed directories, and it outperforms FFS on large
> > directories.
>
> OK, that's good to know.  If you don't mind me asking, which file
> system is that, and what's the base OS?  (I'm just curious, you
> don't have to answer, of course.)  Perhaps this also points in
> the same general direction that the FreeBSD UFS_DIRHASH does?

A commercial, portable, journaled file system (we use NetBSD for
demo purposes) that does meet NFS stable storage rules.

>
> Regards,
>
> - Håvard