Subject: Re: multiple Tx queues
To: Erik E. Fair <>
From: Daniel Carosone <>
List: tech-net
Date: 01/08/2004 12:48:30
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Wed, Jan 07, 2004 at 05:31:33PM -0800, Erik E. Fair wrote:
> So, you're suggesting a driver-specific queueing mechanism based on=20
> TOS, rather than a more general packet queueing framework?

In discussion on icb, there were two main areas of thought:

 - we need a good, general, appropriate interface from the driver, to
   expose this hardware feature into something like altq, so the
   administrator can set policy based on his own choices.  This partly
   presumes that some other chips (wired or not) likely have some
   similar features.

 - a collection of ideas of quick hacks for types of packets that
   could meaningfully benefit "in the meantime". This set included
   tiny ACKS, perhaps TCP PSH frames (though that potentially creates
   ordering problems), and now TOS bits. Another idea was to allow
   individial source addresses (ie, aliases) to bind to l/m/h queues,
   and use the ip addresses accordingly in the external network.
   All cheesy in one way or another.


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

Version: GnuPG v1.2.3 (NetBSD)