Re: [sfc] [mpls] Working Group adoption of draft-farrel-mpls-sfc

[" 徐小虎(义先) " <[e-mail]>]

Hi Stewart,
Thanks for your response. For the SR-based SFC mechanism that has been described in draft-xuclad*, it's not helpful to describe it again in another draft. The most simple and efficient way to address the overlapping issue is to reference draft-xuclad* rather than "using a different name for the same thing". I'm looking forward to seeing the revision of draft-farrel* that would address the overlapping issue concretely.
If co-authors of draft-farrel* believed the current text as described in draft-xuclad* is not good enough or misses something important, any comments and suggestions are more than welcome.
Best regards,Xiaohu------------------------------------------------------------------Stewart Bryant <[e-mail]>2018年4月13日(星期五) 13:27徐小虎(义先) <[e-mail]>mpls <[e-mail]>; "Bernier, Daniel" <[e-mail]>; Robert Raszuk <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>Re: [mpls] [sfc]  Working Group adoption of draft-farrel-mpls-sfc
Hi Xiaohu
What an earlier version of the draft said is of no importance. What it says going forward is what counts.
Perhaps the way to address your concern is to include some text of the form that I used in my email of yesterday to describe to the reader the difference in approach. This is consistent with earlier advice in this discussion to reference the work from which this forked.
- Stewart


Sent from my iPad
On 13 Apr 2018, at 03:35, 徐小虎(义先) <[e-mail]> wrote:

Hi Stewart,
If draft-farrel* was just describing an MPLS-based SFC technology that is different from the MPLS-SR-based SFC technology that has been described in draft-xuclad*, that would be fine. However, draft-farrel* also described the technology that has been described in draft-xuclad* (see section 6) by "using a different name for the same thing". Note that the title of section 6 in those pervious versions of draft-farrel* is "MPLS Segment Routing". One co-author of draft-farrel* said they worked very hard to change the "Segment Routing" term to "label stack" term in the new version of draft-farrel* in order to deal with the overlapping issue. However, such change is just "using a different name for the same thing", and it doesn't solve the overlapping issue at all, as had been pointed out by many people. As said by one co-author of draft-farrel*, in a thread which is irrelavant to this overlapping issue, "using a different name for the same thing is not so clever:)". In fact, it would cause unneccessary confusions to implementors by describing the same technology within different drafts. More badly, it would set a bad precedant in the IETF of copying the idea of the existing draft by "using a different name for the same thing".
Best regards,
Xiaohu------------------------------------------------------------------Stewart Bryant <[e-mail]>2018年4月12日(星期四) 23:04"Bernier, Daniel" <[e-mail]>; Robert Raszuk <[e-mail]>[e-mail] <[e-mail]>; [e-mail] <[e-mail]>Re: [mpls] [sfc]  Working Group adoption of draft-farrel-mpls-sfc

Rather than have a process discussion, I think we should go up a level 
and better understand the technical differences between the two drafts.

draft-farrel-mpls-sfc describes the actions at a hop in terms of a tuple 
that mirrors the SFC approach that allows a short indication of 
potentially re-entrant chains. In its simplest form it uses a compact 
MPLS stack to describe an arbitarily complex path that is compatile with 
simple edge routers which are often challenged in terms of the number of 
labels that they can push.

draft-xu-clad-spring-sr-service-chaining unrolls the path and explicitly 
calls out each hop and each function into the label stack. This results 
in a much larger MPLS label stack that will challenge some edge routers. 
The way that we generally deal with imposition limits is through the use 
of binding-SIDs, which in the limiting case resolves to the approach in 
draft-farrel with the limitation that the position on the path is 
implicit in the label stack size rather than explicitly specified by the SI.

Mid-flight path changes (if such things are needed) is clearly simpler 
with draft-farrel.

The short stack in draft-farrel comes at the cost of greater network 
forwarding stack, and the long stack is the price that draft-xu-clad 
pays for the reduction in forwarding state.

The optimal design point between forwarding and control plane state is 
something that is dependent on many parameters, and is dependent on many 
network and operational factors, so much so, that don't think it is wise 
to rule either out of scope at this stage.

The hybrid mode in section 6 of draft-farrel supports the mixed mode in 
section 7 of the draft. This allows the construction of SFCs that are 
the concatination of two or more compacted sub-chains. This allows the 
operator to deploy a solution with the advantages of draft-farrel 
together with some of the flexibility of draft-xu-clad.

At this stage the two drafts are sufficienly different that I think we 
need to proceed with both at least to the point where we fully 
understand the detailed consequences of the two approachs and the 
scenarios where each finds it's niche.

After developing a better understanding the detail of each design, their 
control plane, and operational contexts and how each maps to customer 
network requirements, we can move the drafts to the appropriate IETF 
track. Such tracks may be anything from abandonment to IETF standard for 
one or both of these approaches.

Meanwhile I think that we need to focus our efforts on a deeper 
understanding of the technology and how each might make the Internet 
work better,  rather than spending effort on arguing about IETF process.

- Stewart

_______________________________________________
mpls mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/mpls
_______________________________________________
mpls mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/mpls