Re: [tsvwg] Another tunnel/VPN scenario (was RE: Reasons for WGLC/RFC asap)

[Steven Blake <[e-mail]>]

On Tue, 2020-12-08 at 21:29 +0000, Bob Briscoe wrote:
> Steven,
> 
> On 04/12/2020 01:54, Steven Blake wrote:
> > On Thu, 2020-12-03 at 16:40 +0000, Gorry Fairhurst wrote:
> > > Just on that last point, perhaps we should be clearer by what
> > > "experiment" means.
> > > 
> > > On 03/12/2020 15:52, Jonathan Morton wrote:
> > > > Frankly, the sooner the WG understands and accepts that basic
> > > > fact,
> > > > the sooner we can move to a viable solution - one which does
> > > > not
> > > > redefine CE from the semantics established by RFC-3168 and RFC-
> > > > 8511, and hence does not place burdens on networks which have
> > > > no
> > > > interest in the L4S experiment.
> > > I think we need to be clear that I understand the proposal is to
> > > an
> > > Internet-wide experimental deployment.
> > Have single network experiments been attempted? Have cooperating
> > peering network experiments been attempted? Why is an *Internet-
> > wide*
> > experiment the appropriate next step, especially when potential
> > detrimental impacts to traffic flows of non-participating networks
> > &
> > hosts have been identified?
> 
> [BB] As I understand it, until the IETF assigns ECT(1) to the L4S 
> experiment, no-one should be deploying L4S on the public Internet, 
> whether in single networks, or pairs of peered networks. This
> confines 
> us to experiments where one can only try to generate realistic
> traffic.
>
> Or do you mean a single *isolated* network (or peered *isolated* 
> networks)? If so, isn't that what DCTCP has been doing for some years
> now?

Yes, I mean isolated network(s). IETF is not a regulatory agency. It
cannot prevent interested operators from conducting experiments *within
the confines of their own networks* using whatever signaling they want
(hint: there already is a field defined with an experimental range of
values allocated for steering packets into queues).

Has DCTCP already been tested using Dual-Q outside of the lab or
datacenters?

> Put another way, if implementers/operators started deploying L4S on 
> ECT(1), even in single networks but within the public Internet,
> wouldn't 
> that mean that ECT(1) had become de facto assigned to L4S?

L4S experimentation *within the confines of participating operator
networks* is not dependent on assignment of ECT(1). This has been true
and obvious for years. Every counter argument that I have seen against
using experimental DSCPs to signal L4S traffic in these experiments (at
least initially) has been predicated on ease of widespread deployment,
which should not even be an issue at this stage of experimentation.

The safe and sane thing to do in these initial experiments is to use an
experimental DSCP value to signal L4S/TCP Prague packets and use ECT(1)
to signal congestion in the Dual-Q low-latency queue, avoiding
ambiguity with CE signaling. 


Regards,

// Steve