Re: [tsvwg] COMMENT PLEASE: Which DSCP value should we use for LE PHB?

["Bless, Roland (TM)" <[e-mail]>]

Hi Rüdiger,

Am 02.08.2017 um 10:48 schrieb [e-mail]:
> in an MPLS domain, one generally may assign a bunch of DSCP to a 
> single Treatment Aggregate / PHB. That I think is quite useful. 
> Unfortunately the top label is removed before the last MPLS 
> domain router is reached. Then an unknown and not re-marked 
> DSCP is not acceptable. Hence your guidance below isn't applicable in 
> all networks.

Thanks for pointing out MPLS/DiffServ interworking problems.
I'm more a DiffServ purist so I wasn't aware of such operational
problems. In summary, the whole DiffServ deployment seems to be getting
more complex by integrating MPLS constraints. I'm also not aware how
common an AF deployment is, since you quoted (in your last mail) that
bit-masking AF classes can be quite useful. So if I understood your
statements correctly, it is not possible to carry a DSCP x in an
MPLS aggregate and then use the same DSCP x at the egress to the
next domain?

> There is a fairly simple and useful option allowing to combine Treatment 
> Aggregates, Diffserv forwarding across an MPLS domain and 
> maintenance of multiple DSCP. Per Treatment Aggregate:
> - classify DSCPs by bit-mask at ingress.
> - re-mark DSCPs by bit mask at ingress.
> - Diffserv forwarding by Treatment Aggregate
> - classify and forward by individual DSCP at the egress (if desired)
> - same holds for next domain
> 
> The other simple option I see without bit mask based re-mark is, 
> again per Treatment Aggregate:
> - classify DSCPs by bit-mask at ingress.
> - re-mark to _a single DSCP_ at ingress.
> - Diffserv forwarding by Treatment Aggregate 
> - classify and forward by this single DSCP at the egress
> - same holds for next domain
> 
> I prefer a simple solution (and a standardized one, if possible). 

Speaking as DiffServ purist, DSCP classification _by bit-mask_ is
non-conforming to RFC 2474.

> Of course it is technically possible to set up and maintain per DSCP to 
> PHB and individual DSCP re-mark tables, customized for each 
> interconnection. Interconnection DSCPs are not standardized.
> Many carriers operate multiple interconnections. 

The DiffServ architecture indeed requires to have agreements
for each interconnection. However, in case you don't have negotiated
any special treatment, a mapping of "any DSCP -> default PHB" should
work. But if I understood you correctly, an incoming DSCP x may not
be acceptable for the next DS domain, so you must remark?

Regards,
 Roland