[87046] in North American Network Operators' Group
Re: QoS for ADSL customers
daemon@ATHENA.MIT.EDU (Sean Donelan)
Wed Nov 30 10:15:39 2005
Date: Wed, 30 Nov 2005 10:15:11 -0500 (EST)
From: Sean Donelan <sean@donelan.com>
To: Joe Shen <joe_hznm@yahoo.com.sg>
Cc: NANGO <nanog@merit.edu>
In-Reply-To: <20051130030746.88413.qmail@web53601.mail.yahoo.com>
Errors-To: owner-nanog@merit.edu
On Wed, 30 Nov 2005, Joe Shen wrote:
> To Kim's situation, IP packet header based (or access
> interface based) traffic classification is pratical.
> If application based traffic classification is
> required, tools from sandvine or packeteer may have to
> be sitted between ERX1440 and Cisco7609. IMHO, ISP
> network should NOT trust any TOS/DSCP set by their
> customers; so, classifying and (re)tagging must be
> done in PE or BRAS. On the other hand, anti-spoofing
> configuration must be enabled in ERX1440 or 7609.
> Anyway, I don't trust current router's ability on
> content based traffic delivery.
The problem with waiting until the PE or BRAS to do the classification is
most access providers use traffic aggregation in the access network (e.g.
ATM/DSL, Cable, WiFi, etc). This means the interfaces on the BRAS or PE
are oversubscribed and the access network interface will experience
inbound cell/frame drops.
If you don't trust the router's ability, imagine a dslam's ability to do
it at the ATM layer.
Some networks let users tag their traffic, other networks re-tag all
traffic according the network's policies. At the moment it seems to be a
business decision. But the result is users shouldn't expect unmangled
TOS/DSCP bits over the Internet. Coordinating the IP layer QOS with the
access network/physical layer QOS is a bit of a challenge.
