Qing, Inline with [Shunwan].
Thanks, Shunwan From: Qing Yang [mailto:[email protected]] Sent: Saturday, October 13, 2018 2:27 AM To: Jeffrey Haas <[email protected]> Cc: Zhuangshunwan <[email protected]>; [email protected] Subject: Re: [GROW] bmp loc-rib monitoring scope question (was Re: I-D Action: draft-ietf-grow-bmp-local-rib-02.txt) Right. And the locRib peer header doesn't really have way to do more than 1 anyway. [Shunwan] IMO, Section 5.2 of draft-ietf-grow-bmp-local-rib-02 describe a method to send best ecmp group to BMP Station. BMP client can signal add-paths capability to BMP Station via BMP Peer UP message, then BMP Station will know that the client will send multiple NLRI for one destination. That is my understanding. Some of our customers are also interested in knowing the 'best ecmp group' instead of just locRib. For that, I think we really need to go with the proposal that was thrown here about half year ago. Which is to allocate another TLV to denote various flag bits, 'best ecmp' would be just one of them. This allows the BMP feed to also serve 'multi purpose', in a way. So one can actually correlate a post policy adjRibIn with the best ecmp, for instance. I wonder if there is a general interest in pursuing this further? On Fri, Oct 12, 2018 at 7:06 AM Jeffrey Haas <[email protected]<mailto:[email protected]>> wrote: Shunwan, On Fri, Oct 12, 2018 at 12:16:39AM +0000, Zhuangshunwan wrote: > > From: GROW [mailto:[email protected]<mailto:[email protected]>] On > > Behalf Of Jeffrey Haas > > [Trimming original thread to re-ask my core question.] > > > > On Thu, Oct 11, 2018 at 09:18:17PM +0000, Tim Evens (tievens) wrote: > > > The local RIB in BMP should only contain what is/would be used/installed. > > > > >From where? BGP's best route? The routing table's active route? > > > > > In other words, the local rib sent via BMP should not contain the > > > suppressed prefixes that were not installed due to another routing > > > protocol/direct/static having a better preference. > > > > Which ties into the RFC question about where other protocols are injected > > into the Decision Process. > > > > If you read the RFC as literally saying it's injecting it into the Decision > > Process (section 9.4), the LocRib should be the best route and thus the > > active route regardless of whether it was learned from BGP or not. > > [Shunwan] IMO, the LocRib should include selected and used routes. We can > consider some concrete examples. > Example 1: > If a router enable six parallel load balancing, and now there are 8 available > candidate routes for one destination D. > In this case, there will be 6 routes of the same destination D exist in the > Loc-RIB, but only one of them is the best route. Loc-Rib in the RFC 4271 sense only covers selecting the best route, singular. The fact that your FIB implementation may balance over that for ECMP purposes is an implementation detail. > Example 2: > If we enable BGP Fast Reroute (FRR) in a router, then there may exist a best > route and a backup/alternate route in BGP for one destination D. > In this case, there will be 2 routes of the same destination D exist in the > Loc-RIB, but only one of them is the best route. Again, Loc-Rib in the RFC 4271 sense would consider only the best route. The implementation is using the contents of the Adj-Rib-In to select additional information for local forwarding purposes. I.e. similar to the ECMP case above. -- Jeff _______________________________________________ GROW mailing list [email protected]<mailto:[email protected]> https://www.ietf.org/mailman/listinfo/grow
_______________________________________________ GROW mailing list [email protected] https://www.ietf.org/mailman/listinfo/grow
