Tim Chown wrote:
On Tue, May 03, 2005 at 05:41:13PM +0200, Brian E Carpenter wrote:
The point is that (at the ISPs' request) the diffserv model allows
each ISP along the path to apply a different policy - so a packet
marked for EF treatment inside one ISP might be marked for AF
treatment inside another ISP. (You can argue that would be stupid, but
that's what the diffserv WG heard from the ISPs.) So different SPs will
interpret the same label differently.
How weird :)
The example I chose was deliberately weird to indicate the principle.
Perhaps a more realistic example is if ISP A marks VoIP traffic with
the EF code point, and ISP B marks it with Class Selector 5 (formerly
known as Precedence 5).
In the academic community, a lot of effort has been put in to agree some
common DSCP semantics; this has been particulary useful for LBE, where
end-to-end immutability of the DSCP is highly desirable (DSCP=8).
Well, it wouldn't be needed if everyone agreed that all
sources would use Flow Label = 8 for all LBE flows, but different
domains chose to use (say) DSCP=0 or DSCP=8 for LBE traffic.
If an ISP wants to apply local marking then it can, but one would hope
the value would be restored on exit from the ISP's scope.
No, that's explicitly not a requirement in the diffserv architecture.
And it isn't algorthmically possible without adding a signalling
protocol - there are no 1:1 correspondences rules to invoke.
I'm not aware of any research network that changes the value.
You can test this with the traceroute -t option to set a DSCP value, and
it'll report any change along the path.
I tried this to a US commercial server, and no DSCP change was reported.
WOuld be interesting to see who is seeing this happen in practice.
What I'm telling you is the way we were more or less told to design
diffserv by the operator community. For what people are now thinking,
see draft-ietf-tsvwg-diffserv-service-classes-00.txt.
Given this architecture, IPv6 has the advantage of the e2e Flow Label,
which IPv4 doesn't.
Brian
--------------------------------------------------------------------
IETF IPv6 working group mailing list
ipv6@ietf.org
Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------
