Re: [Idr] Review Updates to draft-ietf-idr-bgp-prefix-sid

[Robert Raszuk <[e-mail]>]

Hi Ketan,

Sorry for mixing two different and independent comments together in my msg
to Acee. One is about "first prefix selection" and let's keep it separate.

As far as anycast point I was talking about anycast SID/index not anycast
prefix. Yes for IP anycast prefix must be the same across nodes. But for SR
data plane I don't see this as such strong requirement.

Imagine R1 and R2 advertsing P1 and P2 with SID S. If you force that P1 &
P2 must be the same you must enable multipath on all the nodes to install S
with both next hops into dataplane.

Contrary if you allow same SID to be installed with different next hops
when it comes with different prefixes you don't have to worry about
multipath being enabled anywhere :). After all it should be up to operator
to decide how he wants to advertise his MPLS indexes.

Remember for mpls2mpls date plane what prefixes are advertised as part of
NLRI does not really matter ...

Anyhow I have a feeling that this discussion belongs to SPRING not to IDR
so let's move it there or take it offline.

Cheers,
R.


On Sun, Feb 18, 2018 at 1:16 PM, Ketan Talaulikar (ketant) <[e-mail]
> wrote:

> Hi Robert,
>
>
>
> I do hope this is not an open surgery of the draft J
>
>
>
> Regarding Anycast prefixes, my understanding was that this, by definition,
> means the same prefix being multi-home and originated by more than one
> routers. I did not follow anycast with different prefixes.
>
>
>
> Also, the handling of anycast involves either of the following:
>
> 1)     Pointing to the “closest” (best on whatever the decision process
> of the protocol involved) originator of the prefix, OR
>
> 2)     Doing load-balancing between the multiple originators (when this
> is possibled/desired).
>
>
>
> So I am not sure I understood the relationship of anycast to the conflict.
>
>
>
> Thanks,
>
> Ketan
>
>
>
> *From:* Idr [mailto:[e-mail]] *On Behalf Of *Robert Raszuk
> *Sent:* 18 February 2018 16:27
> *To:* Acee Lindem (acee) <[e-mail]>
> *Cc:* [e-mail]
> *Subject:* Re: [Idr] Review Updates to draft-ietf-idr-bgp-prefix-sid
>
>
>
> Hi Acee,
>
>
>
> Since we are at the open surgery of this draft there is one more point
> which did not got resolved during our offline discussion. Feedback from WG
> is welcome here.
>
>
>
> The draft RECOMMENDS the following procedure in case of a conflict:
>
>
>
> ... it is RECOMMENDED that the first prefix using the label index is selected.
>
>
>
> In BGP there is no guarantee about ordered delivery of BGP prefixes.  It
> may work such in some specific topologies but not in any arbitrary one. So
> if we recommend to pick one prefix on R1 and different prefix on R2 and
> install in forwarding the same SID for those I am afraid the data plane
> will get quite severely messed up and even finding it during
> troubleshooting will be quite hard.
>
>
>
> Also as pointed out already MPLS anycast SID may on purpose use the same
> label across nodes according to draft-ietf-spring-mpls-anycast-segments-02.
> Now if we want to accomplish the same in BGP we would need to advertise
> different prefixes with the same SID (in MPLS-SR index). the above
> paragraph does prevents that. Note that adding and advertising the _same_
> prefix from different nodes will not work as BGP speakers may pick only one
> (best path) and use it for forwarding. IBGP or EBGP multipath may not
> always be enabled/use same set.
>
>
>
> The crux of the issue when advertising SIDs in BGP is not so much about
> prefixes, but their next hops which unfortunately the draft does not even
> mention once.
>
>
>
> Kind regards,
>
> Robert.
>
>
>