Re: [mpls] [EXTERNAL] Indicators in the stack and ancillary data after the BoS

[John E Drake <[e-mail]>]

Hi,

Snipped, comment inline [JD1}.

Yours Irrespectively,

John


1.        I have sent the previous email as a response to your aside claim that "RFC 8662 does not define what a transit LSR should do when it finds an [ELI, EL] pair at the top of the MPLS label stack, either when it understands the label pair or when it doesn't". As I see it, both Bruno and I have demonstrated that this behavior is defined

[JD]  Perhaps I missed it, but nowhere in RFC 8662 do I see a statement of the form:  "This document updates the transit LSR behavior defined in RFC 6790 to allow a transit LSR that understands entropy labels to remove an [ELI, EL] pair that it sees at the top of an MPLS label stack.  This requires a transit LSR to indicate this capability to other LSRs in the SR domain and it requires the ingress LSR to ensure that when it inserts [ELI, EL] pairs in the MPLS label stack they will only be at the top of the MPLS label stack when received by a transit LSR with this capability."
[[Sasha]] I have not seen any such statement either. However, I think that we may be dealing with a terminology issue. I believe that in SR a node that originates advertisement, say, of an IGP Prefix SID, is considered as the egress node for the segment represented by this SID, even if it is a transit node for an LSP that includes this SID in the middle of the stack of SIDs. This is why they have assumed that considerations of 6790 pertaining to egress LERs are sufficient.

[JD1]  This seems like quite a stretch.  Also, as I mentioned in my first email, there is no discussion of how an [ELI, EL] pair at the top on a MPLS label stack is to be processed or ensuring, when building an MPLS label stack, that [ELI, EL] pairs need to be inserted such that they rise to the top of an MPLS label stack only when they are received by a transit LSR that understands entropy labels.

[JD1}  Do you think that it would be safe to re-use the RFC 8662 techniques for GAL?  I.e., a node indicates whether it can receive a GAL at the top of an MPLS label stack and the ingress LSR constructs MPLS label stacks in which a GAL rises to the top of the stack only at such nodes.  Or do implementations of RFC 5586 explicitly scan an MPLS label stack for GAL labels that are not at the bottom of the stack?  This seems like silly behavior but one never knows.



Juniper Business Use Only