Re: [tsvwg] DSCPs and L4S: Label DSCP

["C. M. Heard" <[e-mail]>]

On Tue, Jun 1, 2021 at 11:18 AM Jonathan Morton <[e-mail]>
wrote:

> On Tue, Jun 1, 2021 at 8:35 pm, C. M. Heard <[e-mail]> wrote:
>
>> What I don't understand is how this proposal prevents an L4S sender from
>> competing unfairly with classic traffic when traversing a non-participating
>> domain.
>>
>> Doesn't the sender need feedback from the receiver that the label DSCP
>> was received? And the the the receiver is itself capable of providing the
>> type of CE feedback that the L4S send needs?
>>
>
> My understanding of this is that L4S traffic is not supposed to *exist*
> outside of the networks participating in the experiment.  The measures to
> enforce that invariant are:
>
> 1: Don't deploy L4S endpoints outside the participating networks.
>
> 2: Use existing Diffserv management infrastructure at the participating
> networks' border to prevent traffic carrying the Label DSCP from crossing
> the border (or, possibly, to limit their volume to a presumed harmless
> level).
>
> If I've got that wrong, I expect to hear a correction.
>

I've seen no corrections, and it seems that both Martin and David agree
with this assessment, particularly point #2.  *BUT*:

On Sun, Mar 28, 2021 at 5:35 PM Bob Briscoe <[e-mail]> wrote:

> Many of the arguments for why a DSCP is not useful e2e also apply when it
> is used as a guard DSCP in a single network. For instance, when an IP
> header with a DSCP is tunnelled in pipe mode, it is not propagated to the
> outer. So it won't be visible, won't be bleached, etc.
>

That was in the "L4S DSCP" thread; see
https://mailarchive.ietf.org/arch/msg/tsvwg/A1HQwlbDSEONIEhU9DXOYreG8HA/. As
a reminder, allow me to also remind folks of a message from Olivier Tilmans
with subject "Prevalence of tunnels in mobile networks" at
https://mailarchive.ietf.org/arch/msg/tsvwg/ij5BqzNtMxKV8VbHE8pUevZjx7k/.
The gist was:

3GPP domains are literally a set of tunnels stitching together multiple NFs
> and nodes/equipments.
>

If the assertions in those messages are substantially correct, there are
good reasons to doubt the efficacy of a Label DSCP even within a
participating L4S domain.

And even if that were not the case, there is the question what would happen
if a Label DSCP did survive, and the traffic so labelled were not filtered
by a participating network at its border. It;s then left to the
non-participating network to do the filtering itself. That is not fair or
reasonable. Participating networks need to protect those who do not choose
to participate.

I think it's clear that what is wanted by the L4S proponents is an
experimental internet-wide deployment, not a set of isolated deployments.
If the WG agrees, it should have the courage of its convictions and declare
that RFC 3168 is obsolete.

No half-measures, please.

Thanks and regards,

Mike Heard