Hi Ross, On Mar 27, 2014, at 5:28 PM, Ross Callon wrote: >> we can reword the statement so to explain that there are places >> where mpls is not planned to be deployed while ipv6 is and it is >> preferred (by the network operators of such infrastructures) to >> extend ipv6 with sr rather than changing dataplane. > > To me this paragraph above is the real justification for use of IPv6. > (although one could debate whether forwarding based on a not-yet-standardized > new extension header constitutes changing the dataplane).
well, depends on the change... the proposal we've made for v6 it's a minor extension of v6 dataplane: extensions headers are there from rfc2460 (even if the behavior related to them has been updated by rfc7045). The SR extension consists on a new type of an already existing header. I believe that a v6 network operator would find it more attractive than a radical change of it's dataplane. Anyway, I think we are in agreement. > This points to my main concern with > draft-martin-spring-segment-routing-ipv6-use-cases -- specifically that it > confuses some perfectly reasonable use cases for SPRING (which would be > equally valid with either MPLS or IPv6 implementation), with the choice of > data plane. > > Looking at draft-martin-spring-segment-routing-ipv6-use-cases in detail: > > Section 2.1 ("IPv6 Segment Routing in the Home Network") does not seem too > bad. On the one hand as far as I know the MPLS data plane has not been > deployed in home networks -- although to be fair just how my digital cable > box and my digital telephone box talk with the network is a detail that I > don't need to know and in my role as a homeowner I don't want to know. Also > the data speeds needed in the home network might be slow enough that purely > software forwarding might be appropriate (at least for now). I don't disagree but if the reality of the infrastructure is based on v6 dataplane, I will obviously work on a solution that preserves the dataplane. Realistically, I will never go to a large operator running a v6 networks and tell him "ho, btw, just move to mpls and you'll get what you want" if his question was "can you make source routing available in v6". I'm agnostic in terms of architectures, dataplanes, protocols, etc. but I do care about industry reality and customer requirements. I think you and me agree that we should not let spring become an arena where mpls fans and haters try to shoot down each others. > Section 2.2 ("IPv6 Segment Routing in the Access Network") seems to me to be > a perfectly reasonable explanation for why SPRING is useful in the access > network, along with a statement along the lines of the paragraph above (that > some might just not want to use an MPLS data plane). yes. The initial paragraphs of section 2 give, imho, a good description of the context. > The same could be said for sections 2.3, 2.4 and 2.5, which contain perfectly > reasonable explanations for why SPRING might be useful in Data Centers, > Content Delivery Networks, and Core Networks. However, why the data centers, > content delivery networks, and core networks might want to use SPRING, and > whether they choose to use an MPLS forwarding plane or an IPv6 forwarding > plane, seem like independent issues. Given that a SPRING control plane should > be able to work just fine with either data plane, the choice of data plane > might come down to issues such as what products get built and what > performance can be achieved at what cost. I fully agree. SPRING charter is about architecture that, once agreed, is translated into the necessary protocol extensions so to apply it on existing dataplanes. Thanks. s. > > Ross > _______________________________________________ spring mailing list spring@ietf.org https://www.ietf.org/mailman/listinfo/spring