[c-nsp] Segment Routing

[Mark Tinka]

On 4/Jan/17 06:25, Aaron wrote:

> I run an MPLS network for an ISP and have heard about SR/SPRING but I don't
> know much about it.
>
> What would you tell someone like me as to how I would benefit from SR/SPRING
> in my MPLS network ?   ...and if there isn't immediate benefit, are there
> inevitable long-term benefits that I could reap by moving towards a segment
> routed mpls network ?

I recently chaired an SR panel at the just-concluded APRICOT 2017
meeting in Ho Chi Minh City, last week.

Personally, I think SR has a limited use-case today, for the majority of
MPLS networks. If you run an MPLS network that is heavily into RSVP-TE,
whether for primary label signaling or for steering traffic, SR could be
useful for you as it is stateless, but still achieves the same TE
end-goals (that's the theory, anyway).

If you run purely (or mostly) LDP, then adding SR will not be very
useful, apart from ensuring you now have time to lose sleep hours.

And for the TE side of things, I think SR will only really be helpful
once it is bundled with the controllers that have been touted for it. I
expect vendors to get more serious on this in 2H'17, but mostly in 2018
and going forward. Without "the controller", SR is not going to deliver
what it advertises on the tin.

There is also a major concern where lots of vendors are now shipping
boxes based on merchant silicon, and the hardware requirements that SR
will impose on forwarding chips. I am deeply worried about what this
will mean when you buy boxes that are based on merchant silicon that can
only handle a certain label depth, and what that could mean for your SR
core in a year or two from now.

On our side, I've been mulling over SR for 2 years now, and I've figured
out why I was not fully convinced to plan a roll-out. If you have a
stable MPLS network now, especially if it's driven by LDP, I'd say wait.
SR is not something I'd rush into, but do keep an eye on it.

Mark.