--imjhCm/Pyz7Rq5F2
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Mon, Jul 25, 2005 at 09:42:13PM +0200, Martin Husemann wrote:
> On Mon, Jul 25, 2005 at 12:29:42PM -0700, Bill Studenmund wrote:
> > No, he's not.
>=20
> How do you know? He has given no details AFAICT, and the description coul=
d be
> a broken disk with lots of retries or PIO wdc transfers or what have you.

Because his description fully matches an issue I've personally=20
experienced.

Things are going along fine and the app is writing to an mmap'd file quite=
=20
happily, then things just slam to a stop. Then things resume.

You are right that there could be media reliability issues too, but I=20
think they would generate a different i/o pattern. In the dead times, he's=
=20
seeing (according to the original note) 1 MB/sec, which doesn't sound like=
=20
a drive experiencing i/o errors.

> I agree with you, of course, that ioflush has optimization potential (and=
 the
> maxphys issue should be solved, etc.)

I actually think this issue is not just an optimization, but correcting a=
=20
bug. :-)

Take care,

Bill

--imjhCm/Pyz7Rq5F2
Content-Type: application/pgp-signature
Content-Disposition: inline

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (NetBSD)

iD8DBQFC5UbIWz+3JHUci9cRAmpAAJ96iNF4mxKJqKjl8qSqGiwXEQlkVQCfTzpk
sbp0zpZIkEUBRasFDehBs8k=
=1B5x
-----END PGP SIGNATURE-----

--imjhCm/Pyz7Rq5F2--