Re: [tsvwg] Informational: draft-geib-tsvwg-diffserv-interconnection-00.txt

[Joe Touch <[e-mail]>]

On 11/6/2012 4:19 AM, Brian E Carpenter wrote:
> Some thoughts on this draft, based on a preview over the weekend:
>
> 1. I think it is a reasonable idea to specify a small set of traffic classes
> that could be the default set in peering relationships. Getting consensus
> on the best set could be tricky, of course.
>
> 2. I may be old fashioned, but I don't really understand why we bother
> with layer 2 values at all. If diffserv is working properly, layer 2 can
> just give the same service to all packets. (Unless there are non-IP services
> on the same wire.) In fact I'd rather that layer 2 doesn't try to impose
> any queue or drop behaviour other than FCFS.

Because your assumption that IP diffserv is sufficient holds only when 
layer 2 is composed solely of point-to-point links AND all L2 packets 
consist only of IP.

You addressed the latter point, but not the former. When the L2 is 
either switched (ethernet) or coordinated MAC (802.11), the diffserv 
from L3 needs to be mapped to L2 so that competing flows from different 
sources that are merged onto a single L2 link provide that muxing 
mechanism the info needed to decide what to prioritize.

Joe