Subject: Re: another anomaly
To: Zeljko Vrba <zvrba@globalnet.hr>
From: Bill Studenmund <wrstuden@netbsd.org>
List: tech-kern
Date: 07/25/2005 12:32:24
--zbGR4y+acU1DwHSi
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Mon, Jul 25, 2005 at 09:09:01PM +0200, Zeljko Vrba wrote:
> I have also noticed the following unfair bufcache assignment to filesyste=
ms:

How is this unfair? If the file system is the component in the system that=
=20
is generating the most dirty pages, why not let it?

> Could the poor performance of ioflush be due to this memory assignment?

No. As I note in a follow-up to your other note, the problem is how we=20
schedule dirty pages for cleaning.

> Because the program is working on /mnt/data partition.
>=20
> There are  34053 metadata buffers using             58972 kBytes of memor=
y.
> There are 714051 pages for cached file data using 2856204 kBytes of memor=
y.
> There are   1397 pages for executables using         5588 kBytes of memor=
y.
>=20
> File System          Bufs used   %   kB in use   %  Bufsize kB   %  Util %
> /usr                     33570  98       55453  94       55593  94     100
> /                          320   0         879   1         885   2      99
> /mnt/data                  162   0        2474   4        2478   4     100
>=20
> Total:                   34052  99       58806 100       58956 100     100

Take care,

Bill

--zbGR4y+acU1DwHSi
Content-Type: application/pgp-signature
Content-Disposition: inline

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

iD8DBQFC5T5IWz+3JHUci9cRApxPAJ9wPwWQoWkr9hp+54366vL8IxNVdwCgjeZ2
K2XQPTG62QT+HKKks2wyOdM=
=qCma
-----END PGP SIGNATURE-----

--zbGR4y+acU1DwHSi--