Jeff, Your explanation of distinct Transport SIDs and service labels (which appear at BoS) applies to Point-to-Point services. Same model can be applied when a Point-to-Multipoint service is realized by one end-to-end replication segment; "Downstream Replication SID" as specified in draft is effectively the service label at a specific downstream node of a Replication segment with Transport SIDs imposed on top to take replicated packet to that node.
However, when a Point-to-Multipoint service is over replication segments stitched together to form a P2MP tree, as described in PIM WG draft, this model no longer holds since all service egress nodes would have to agree on one common service label. So P2MP services implicitly map the P2MP transport label (Replication SID at BoS in this case) to the P2MP service. Of course, this implies one-to-one association between a P2MP service and a P2MP transport. There are techniques to share one P2MP transport across different P2MP services using either upstream assigned label or a global context from "Domain-wide Common Block" (draft-ietf-bess-mvpn-evpn-aggregation-label). These and other gory details are described in Section 2 of PIM WG draft and to some extent in BESS MVPN-EVPN draft. -Rishabh On Wed, Jul 1, 2020 at 5:50 PM Jeff Tantsura <jefftant.i...@gmail.com> wrote: > > Rishabh, > > Transport SID with a service on top can’t be a BoS label, there’s s service > label below, since a service is associated with a particular node, there > would be at least a N-SID associated with the service node. > It seems like B-SID behavior is the correct one, when R-SID is looked up and > popped, it would yield: replication (as programmed by a controller, since > there’s no state) + new label stack associated with the new, post > replication/branching path that is imposed on top of existing label stack, so > service label ( + optionally more transport labels) are preserved. > > Cheers, > Jeff > On Jul 1, 2020, 12:40 PM -0700, Rishabh Parekh <risha...@gmail.com>, wrote: > > Robert, > > A) Firmly state that replication SID MUST be the last one on the stack > B) Instead of real SID after the replication SID provide a binding SID which > locally will be mapped to a different SID list imposed to each replicated > flow. > > > We would be fine with A), but we don't want to exclude possibility of > something like what you describe in B. > > -Rishabh > > On Wed, Jul 1, 2020 at 12:27 PM Robert Raszuk <rob...@raszuk.net> wrote: > > > Hi Rishabh, > > Of course, care must be > taken to avoid the "explosion" as you describe it. G-SID-2 has to map > to a unique node; for example, it may be an Anycast-SID that takes > packet to distinct nodes from each of the downstream node, or the > downstream nodes can be border nodes connecting to other segment > routing domains where G-SID-2 resolves to distinct nodes in each > domain. > > > I think you are stretching it too thin. > > See even if G-SID-2 is anycast SID you have zero assurance that physical > nodes packets will land on would be at all diverse. > > Likewise crossing domains yet providing identical global SID now to be a > different node in each such domain to me is not a realistic example. > > I think we have two options: > > A) Firmly state that replication SID MUST be the last one on the stack > > B) Instead of real SID after the replication SID provide a binding SID which > locally will be mapped to a different SID list imposed to each replicated > flow. > > What is currently in the draft seems to be very counterintuitive and IMHO > will result in operational difficulties. > > Thx a lot, > R. > > > _______________________________________________ > spring mailing list > spring@ietf.org > https://www.ietf.org/mailman/listinfo/spring _______________________________________________ spring mailing list spring@ietf.org https://www.ietf.org/mailman/listinfo/spring
