Russ - Happy New Year. Responses inline.
> -----Original Message----- > From: Lsr <lsr-boun...@ietf.org> On Behalf Of 7ri...@gmail.com > Sent: Saturday, December 19, 2020 4:49 AM > To: 'Les Ginsberg (ginsberg)' <ginsberg=40cisco....@dmarc.ietf.org>; > 'Shraddha Hegde' <shraddha=40juniper....@dmarc.ietf.org>; lsr@ietf.org > Subject: Re: [Lsr] New Version Notification for draft-white-lsr-distoptflood- > 00.txt > > > > > Given the base work defined in draft-ietf-lsr-dynamic-flooding why is > > the use of Circuit Scoped LSPs required/useful? > > Circuit scoped flooding is required to prevent unnecessary floods from being > carried beyond where they are needed in the network. > [Les:] Signaling is required in your solution to indicate whether the neighbor should/should not propagate the LSP(s) received from a given neighbor. But Circuit Scoped LSPs as defined in RFC 7356 do not provide this functionality. A Circuit Scoped LSP is to be used to send information originated by a node to a specific neighbor of that node. This information has circuit scope - meaning it is NEVER meant to be flooded on other circuits. It is not an encapsulation mechanism to forward area/domain scoped LSPs originated by other nodes in the network. The obvious conflict arises when a given Node needs to originate circuit scoped information. The LSP-ID associated with this information will be indistinguishable from the LSP-ID of an area/domain scoped LSP originated by that same node. So your overloaded use case cannot be supported. BTW, this isn't simply a theoretical issue. One of the documented use cases for Circuit Scoped LSPs is defined in https://tools.ietf.org/html/draft-ietf-lsr-isis-spine-leaf-ext-02 - which might well be relevant to the same deployment cases this draft targets. > One thought ... the way I originally worded the draft was not to have two > separate databases, but rather just to insert LSPs received as circuit local > in the normal database and then clear the SRM for those LSPs, so the local > flooding process believes they have already been flooded, but they will be > included in the CSNP, etc., as normal. I'm not certain why we need two > databases here. > > > [Les:] If you use the mechanisms defined in > > draft-ietf-lsr-dynamic-flooding there would be no need to use Circuit > > Scoped Flooding to send normal area scoped LSPs. > > > Each node in the topology would know to which neighbors it should have > > flooding enabled - either because you operate in centralized mode and > > the Area Leader distributes the flooding topology or because you operate > > in distributed mode and all nodes employ the same algorithm. (The latter > > Using an area leader, even to centrally calculate a flooding topology, > breaks the distributed nature of the solutions, and creates a much more > complex solution. If you operate in distributed mode, with a common > algorithm, I think you still need to specify what that algorithm is ... this > draft describes such an algorithm ... I think ... ?? > > > The fact that you apparently do NOT want to use the extensions defined > > in the dynamic-flooding draft means that (as you agree above) you have > > to introduce additional protocol extensions which aren't really > > necessary. This makes the solution much less appealing to me - > > independent of the merits/demerits of the algorithm itself. > > Not really... we are modifying already existing protocol extensions, rather > than creating new ones. [Les:] What you have done is abuse an existing mechanism and try to use it in a way which conflicts with its defined functionality. > > The process described in this draft has already been tested, and is widely > used, in OSPF, btw (and doesn't use two databases there). > [Les:] I assume what you are referring to here is MANET and the Multipoint Relay (MPR) functionality. But MPR depends upon enhanced signaling of a node's willingness to participate in enhanced flooding - which you have not defined. So in addition to abusing Circuit Scoped LSP functionality, you haven't defined a means to operate with partial deployment of the extensions or even to know whether a node supports the extensions. Seems to me you have skipped some key steps here. > > The use of Circuit Scoped LSPs to send traditional area scoped LSPs is > > not at all what RFC 7356 intends. And it causes difficulties (as you > > have noted above) in what the content of CSNPs should be and how the > > different LSPDBs are used internally. > > > The dynamic flooding draft wasn't in existence when Russ wrote the > > early versions of this draft. I am suggesting you might want to rethink > > your approach to take advantage of what dynamic flooding draft has > > defined. > > The dynamic flooding draft assumes an area leader, which is what this draft > is attempting to avoid. > [Les:] It does seem that you do NOT want to use dynamic flooding extensions. Which makes me wonder why you suggest (Section 3) the election of an Area Leader. What purpose does this serve in your solution? Les > /r > > > > _______________________________________________ > Lsr mailing list > Lsr@ietf.org > https://www.ietf.org/mailman/listinfo/lsr _______________________________________________ Lsr mailing list Lsr@ietf.org https://www.ietf.org/mailman/listinfo/lsr
