[88151] in North American Network Operators' Group
qos on access (Re: is this like a peering war somehow?)
daemon@ATHENA.MIT.EDU (Mikael Abrahamsson)
Mon Jan 23 07:02:54 2006
Date: Mon, 23 Jan 2006 13:02:22 +0100 (CET)
From: Mikael Abrahamsson <swmike@swm.pp.se>
To: nanog@merit.edu
In-Reply-To: <Pine.WNT.4.63.0601230651070.3724@jvc>
Errors-To: owner-nanog@merit.edu
On Mon, 23 Jan 2006, Todd Vierling wrote:
> Provided that IPTV or VoIP are the only [large] datagram streams in
> progress, this is not difficult to do on the CPE end -- in fact, this
> sort of thing is already available as a QoS option in some home router
> appliances. It involves scaling back TCP streams so that they leave
> enough headroom for the in-progress IPTV or VoIP.
Doing QoS on the CPE side of the access link is always risky, you can't
really create any guarantees there since it's already "too late", you're
not the one doing the buffering. Especially since a lot of L2/L3 DSLAMs
don't have a lot of buffers (40ms on ADSL2+ 24meg in one of my tests),
you really have to strangle TCP in order to guarantee that you TV packets
aren't dropped.
--
Mikael Abrahamsson email: swmike@swm.pp.se
