--EVF5PPMfhYS0aIcm
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Sat, Jan 22, 2005 at 07:44:19PM -0500, Thor Lancelot Simon wrote:
> On Sat, Jan 22, 2005 at 03:06:28PM -0800, Jason Thorpe wrote:
> > Might as well check it in, then?
>=20
> I dunno -- Gordon's approach seemed a lot better to me -- even if it
> would really give the benefit only for new filesystems.  I also wonder
> if its expanding hash wouldn't give better performance for the truly
> gigantic directories where this version of the dirhash code seems to
> fall down.

I'm sure that there are better solutions, but this particular one
has the benefits of not only existing in a form that is quickly
made ready for integration, but of also having been out in the wild
for several years. We can pull it out easily if a better solution
materialises somewhere down the line that obviates dirhash.

The performance drop with huge directories has me a bit confused,
but I'm more inclined to think that it's something I've missed
(or fudged) rather than a limitation of dirhash itself.

I'm going to commit what I have (it's disabled by default) and let=20
others take a look.

Steve

--EVF5PPMfhYS0aIcm
Content-Type: application/pgp-signature
Content-Disposition: inline

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.0 (NetBSD)

iQEVAwUBQfPtxcuCaQyb9VOkAQKtsggA2kFNXR4j/U5dbmJ27js9r5yuq9TiykHh
Kll0uOulNp0bTvPO/q2jMdJrCmyRZCu8U1dLZK0EM/AS+xFUB/NnKUoGCpzx4G+5
vW+gJX7SQ+T70Jmz13uedX497S0jv2kfYwDoYLZ+Fd7VN7tekXmKBqAM6BPe6Hbx
fIaZcuuJv1zmidGAqnfc+AFPmiSKZ6wpmku4iADuz4agDFUHzxcHov/wHRpgSfIR
+W8fcSl+/ig/hxgyBdHzMo5oB96gdpVq0P+bhZq2jVi/8qA+7JTI7YT1hrJrviPH
eRyFhuYdDoXV/IHOWMhhMcMNZMNjRhk597sD8wIVkt1TbWTTaUb+FA==
=b3/V
-----END PGP SIGNATURE-----

--EVF5PPMfhYS0aIcm--