>> >> Maybe I'm confused, but I'm still not hearing why the LSB buys us >> anything above and beyond what we have to have anyway: i) mappings, and >> some way to detect that an ITR has an outdated copy of one, and ii) >> reachability (which necessarily has to discover, on a fairly real-time >> basis, a superset of 'up'). >> > > > Noel, Dino, > > Thanks for clearing up my misconceptions regarding LSB function. Now, I think > that we are in a better position to discuss Noel's question, above. > > Dino has suggested that the LSB provides a convenient way for ETRs to inform > ITRs that locators have been taken in or out of service. This can be done > without involving the mapping service.
What the LSBs buy us is fast convergence. Let me explain. (1) An IGP operating in a small part of the network can converge faster than propagating control messages to remote ASes. (2) Once the IGP tells the data-plane, the data-plane sets these control bits. That means the control bits are sent from one AS to another very fast, via fast-switching of packets in the core routers. (3) The LSBs are NOT reachability status. They are source up/down status that the source knows about. Where the source here is the site that operates the xTRs. > However, the mechanism only works in "controlled environments". Furthermore, > even in controlled environments, when the LSB==1, the ITR still has to test > for reachability. This is because a locator can be in service, but > unreachable from a particular ITR. (I am not sure about this, but when the > LSB==0, might the ITR have to test for reachability anyway, just to ensure > that things haven't changed). If your LSB bit is 0, it doesn't matter if the reachability to you is up, because if you are down, then it doesn't matter. However, if your LSB bit is 1, it means you are up. But from my part of the topology you may be unreachable, but from Noel's part of the topology you may be reachable. So up/down is global and reachability is pairwise. > Do I have this much right? > > If so, does the LSB offer enough to be worth the complexity and processor > load. I believe it is worth it and it is being used in different use-cases and environments. Dino _______________________________________________ lisp mailing list [email protected] https://www.ietf.org/mailman/listinfo/lisp
