Hi Bruno, Do you mean « UPA is not enough” or “Hint” is not enough. > > Assuming the later, we can’t force people to use UPA, plus IP aggregation > existed before UPA. > > But we can rephrase to first express the drawback of aggregation (loss of > individual reachability) and then refers to a solution. >
I meant "hint" may not be enough - but I am not pushing you nor I think you/authors should not be pushing operators to UPA. IMO and I and few others were very clear when UPA was discussed removal of service routes can be as fast as UPA (or I would argue even faster if done correctly) therefor the requirement of transport layer signalling of remote unreachability with aggregation may not be a MUST. We have the same amount of entropy (an individual label per specific FEC). > The different is the additional top label for the MPLS hierarchy. Typically > implementations hash multiple labels so this would not make a difference. > (more bits, same entropy) > Well if you hash on more than the top label in the stack :) Likewise what happens if PEs add entropy labels during aggregation ? > > > > For SR, ELC is signaled on a per prefix basis, since rfc9088. > > As aggregation hide the details of each prefix, ELC signaling on a per > egress LER basis would be lost, just like all other prefix specific > signaling. A priori this make sense to me since aggregation is about > summarization, hence hiding the details. > > OTOH, the aggregation segment could probably be advertised with ELC, > allowing for the use of EL from ingress LER up to egress ABR. > > Alternatively, if needed, the ingress LER could learn the specifics of the > egress area using BGP-LS and hence learn ELC on a per egress LER basis. > I guess we fully agree here. I was just making an observation which I did not see much written about in the draft. > Bottom line is that IMO all pros and cons should be listed. > > > > Of IP aggregation that we are doing for 25+ years (e.g., details of each > prefix is hidden)? Would you have a pointer to an existing document? (if > it’s useful text, it probably have already been written in 25 years) > > Or only for the specific SR-MPLS aspects? > I meant for SR-MPLS aggregation of course. But since you mentioned 25+ years .. what would be really cool to compare IP encapsulation with aggregation vs SR-MPLS. Is there any benefit for the latter at all ? Cheers, R.
_______________________________________________ spring mailing list -- spring@ietf.org To unsubscribe send an email to spring-le...@ietf.org
