Subject: Re: NetBSD iSCSI HOWTOs
To: Aaron J. Grier <agrier@poofygoof.com>
From: Bill Studenmund <wrstuden@netbsd.org>
List: current-users
Date: 02/22/2006 17:30:19
--Xm/fll+QQv+hsKip
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Wed, Feb 22, 2006 at 04:42:32PM -0800, Aaron J. Grier wrote:
> On Wed, Feb 22, 2006 at 10:19:46PM +0000, Alistair Crooks wrote:
> > I'm working on a NetBSD initiator, but it won't be in 4.0.
>=20
> I'm thinking a RAID of iSCSI targets would be interesting.

Yep, it is.

Though don't add too many boxes to the RAID as you then get TCP problems.=
=20
David Nagle, at Panasas, described the problem to me at FAST and refered=20
to it as "In-cast", the opposite of multicast. If a lot of senders (iSCSI=
=20
targets talking to an initiator) flood a receiver, packets get dropped and=
=20
TCP goes into recovery. For a striped read, you then end up having the=20
whole operation wait for one connection to recover. There's probably a PhD=
=20
dissertation in all of this. :-)

Where exactly "too many" is at depends on the drives you have in each box=
=20
and your network. Look hard at the receiving NIC in the initiator.=20
Depending on how beefy the iSCSI targets are, you may need 10G ethernet.

Take care,

Bill

--Xm/fll+QQv+hsKip
Content-Type: application/pgp-signature
Content-Disposition: inline

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

iD8DBQFD/RArWz+3JHUci9cRApgYAJ9sV1M8fG5K68T4IOMBxQVRIJyWvwCeNMYO
FUeO+7mTJ0SIq8KNnhgBQJU=
=cUlX
-----END PGP SIGNATURE-----

--Xm/fll+QQv+hsKip--