stefano, et al,
my read of SR and IPv6-SR in the past year is:
1. that in 2013 you have started out with a *unified* control-plane
based on stacked-tunnels technology which supports more than
one data-plane.
2. because it supports 'more than' dataplane and there is
a high congruency between dataplane forwarding semantics you need an
overarching, common 'architecture' with new terminology etc.
now lets have a closer look after more than one-year of 'SR-promised-land':
1. There is little to no SP traction on IPv6-SR.
those SPs who have voiced support in all reality
can be served well with nested IP tunnels based on
exisiting protocols (GRE, IPIP, L2TPv3 )
2. For certain usecases (e.g. protection, OAM, controller-based TE, etc)
there is SP traction in SR-over-MPLS.
3. There is little to no congruency in terms of forwarding-semantics
between the IPv6-SR dataplane and the MPLS data-plane.
a) MPLS forwarding
= loose and strict forwarding based on local assigned tags
b) IPv6-SR forwarding
= best-matching prefix forwarding based on global assigned adresses
4. the cost of adding explicit routing forwarding capabilities
to IPv6 forwarding hardware is now visible (and guess what - its expensive
to implement on high speed (100GBIT/s lookup engines -
(parsing IPv6-extensions headers does require a lot of memory references)
5. IOW the need to support for 'more-than-one dataplane' has just shrunk down
to 'one dataplane' of practical relevance (which is MPLS).
6. for *one* practical dataplane we do not need new architecture, nor new
terminology.
all what is required is some further clarification to rfc3031, which
'draft-gredler-spring-mpls' really is all about.
what really irritates me is your repeated delay tactics around publication
of 'IPv6-SR' in sufficient details, such that one can see the clear need for a
*common* architecture. my trust that this will happen has gotten down to almost
zero.
before asking the WG to adopt a common SR architecture and terminology
i'd rather ask you to *proof* the need for it. right now there
is *zero* evidence on the table that we need a 'common, dataplane neutral
SR architecture' at all, for supporting the practical relevant use-cases.
/hannes
On Mon, Jun 02, 2014 at 07:36:47PM +0000, Stefano Previdi (sprevidi) wrote:
[ ... ]
| > i hope it does not take another 12 months
| > (e.g. like the last time when we have asked for the data ipv6-sr plane
specification
| > presented at mplswc2013 - actual publication data was feb 2014 ...) -
|
|
| Indeed, SR-IPv6 was initially presented at mplswc2013, then it
| triggered interest and collaboration from many parties. Note that
| we met during a couple of hours in Paris at that time and we gave
| you all the details of what we had in mind, including the ipv6
| part.
|
| Then, at some point in the cycle, we published a draft describing
| what the authors worked out in terms of architecture and SRH details.
| I don't think I have to come with a justification about the amount
| of time it took.
|
| Anyway, in the mean time, implementation work from multiple parties
| started (for both mpls and v6 dataplanes). Now that we have multiple
| implementations, we are working on interoperability and we will,
| probably, again update some of the details but don't take this as a
| commitment from my side to publish anything in a delay that suits
| your time perception.
|
|
| s.
_______________________________________________
spring mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/spring
