Re: [Tsv-art] Tsvart early review of draft-ietf-sfc-nsh-ecn-support-08

[Donald Eastlake <[e-mail]>]

Hi Olivier,

Thanks for the review, see below:

On Fri, Nov 19, 2021 at 6:58 AM Olivier Bonaventure via Datatracker
<[e-mail]> wrote:
>
> Reviewer: Olivier Bonaventure
> Review result: Almost Ready
>
> I read the draft as part of the tav-area review team and found some parts that
> are unclear and should be clarified before publication.
>
> I have one technical concern regarding the tunnelEcnCEMarkedRatio information
> element. This IE is defined as "The ratio of CE-marked packets at the
> Observation Point".
>
> This definition is unclear. It does not indicate the period over which the
> ratio has been computed. This looks critical if the ingress plans to use the
> ratio to adjust its rate or provide ECN feedback upstream. All other IEs refer
> to the initialization of the metering processes the start of the period. My
> impression is that there are two possibilities:
>  - clearly specify the period for the computation of tunnelEcnCEMarkedRatio,
>  but different deployments will probably need different periods
> - let the ingress compute the tunnelEcnCEMarkedRatio from the other IEs that it
> received and remove the tunnelEcnCEMarkedRatio IE that would then be redundant

Alternatively, the ingress could communicate a time window duration to
the egress with an Options Template Record. Use of the ratio rather
than the marking type counts reduces the volume of the Data Records.

> Editorial comments
>
> 1. Introduction
>
> What does an interior SFC node need to implement ECN ? If this similar to the
> requirements for routers or do they need to do something special ?

ECN would ideally be implemented at all bottlenecks, that is, points
where sufficient congestion could cause packet drop. This is typically
the output queue at an output port but could occur at other places in
a device. It makes little difference if the device is an SFC node, a
router, a bridge, ...

> 1.2 ECN Background
>
> Why do you indicate RFC3168 as an example solution to carry ECN information
> while this is a standards track document ?

I'm not sure what the question is. Both this draft and RFC 3168 are
standards track.

> 3.1
>
> You mention non-IP protocols that support similar services as ECN. Could you
> refer to one as an example ?

OK.

> 4.1
>
> Third paragraph, last sentence "not affectED by packet drop"

OK.

> 5. Example of Use
>
> I have difficulties to understand the examples.
>
> In figure 9, a message is an IPFIX message, am I right ? Does the figure
> indicate that IPFIX messages will be sent very frequently ? How often should
> they be sent ?
>
> Figure 10, why putting egress on the left and ingress on the right ?
> I cannot figure what A1, B2 etc really mean and don't understand this figure
> and the associated text.

I'll see what I can to clarify the example.

Thanks,
Donald
===============================
 Donald E. Eastlake 3rd   [phone] (cell)
 2386 Panoramic Circle, Apopka, FL 32703 USA
 [e-mail]