Re: WGLC on the draft-ietf-idr-best-external-05
Robert Raszuk <robert <at> raszuk.net>
2012-02-11 16:45:56 GMT
Hi Jay,
> optional knob, with the existing "announce only one's overall best"
> behavior continuing as the default.
+
> optional knob, with the existing "announce only one's overall best"
> behavior continuing as the default.
The default today is not as you say above. Today if IBGP path is
selected as best path ASBR/PE does not advertise anything into IBGP
mesh. (If it advertised before ebgp learned path it will withdraw it).
Typically this happens due to MED or Local-Preference.
"best-external" proposal does not reduce information advertised into
IBGP mesh. Contrary it enhances it with additional paths which otherwise
would be suppressed from getting advertised.
If ASBR/PE chooses as best path it's ebgp path it will advertise it as
usual regardless if best-external knob is enabled or not.
To summarize I see no point of stating in "Deployment Considerations"
that networks should configured their IBGP mesh correctly. I sort of
take this as given ;)
Conclusion: I am fine to progress the draft as is.
Best,
R.
> Susan Hares writes:
> >
> >
> > This is to start a WGLC on draft-ietf-idr-best-external-05.txt which will run
> > from 2/11 - 2/25.
> >
> > Please send comments to the list.
> >
>
> I cannot support this draft in its current form, but I will suggest a
> change that I would be able to support.
>
> Prior to best-external when a router X in AS1 learns a route via ibgp
> having a bgp next-hop of border router Y (also in AS1) for a route
> coming from Y's neighboring AS2, X can be assured that if it can
> somehow deliver packets destined for that route to router Y, that Y
> will forward those packets out of AS1 into AS2.
>
> With best-external, this is no longer necessarily the case, as Y's
> announcement may be its best-external route that Y does not itself
> use for its forwarding.
>
> I think it's possible that many network designs today implicitly rely
> on the current behavior. Certainly many network operators believe it
> to be so, and would be misled by best-external.
>
> I do not think that the Deployment Considerations section goes far
> enough to address this. I agree that in most network designs, router
> X would also learn the advertisement that router Y chooses as its
> overall best and router X would make the same overall best-path choice
> as router Y had made, but it is not a certainty that this is so and it
> should not be assumed so here.
>
>
> My suggestion to remedy this concern is to add text requiring that any
> router software implementing best-external MUST enable it only as an
> optional knob, with the existing "announce only one's overall best"
> behavior continuing as the default.
>
> If this suggestion is adopted, then the Deployment Considerations
> section should be adjusted to say something like "Networks should
> enable best-external only if they know their bgp mesh is constructed
> in such a way that routers learning a non-overall-best best-external
> route from some other router in their AS will also learn the same
> overall-best route the other router preferred." (Or to continue using
> my example from above: routers in AS1 should enable best-external only
> if their bgp mesh is constructed such that router X will always learn
> the overall-best route preferred by router Y.)
>
> Thanks.
>
> Jay B.
>
> _______________________________________________
> Idr mailing list
> Idr <at> ietf.org
> https://www.ietf.org/mailman/listinfo/idr
>
>
_______________________________________________
Idr mailing list
Idr <at> ietf.org
https://www.ietf.org/mailman/listinfo/idr
> > -- Jeff > > Likewise. > So, is the WG generally happy with this? Does it feel that this is sufficiently baked to request WGLC? Get your final hits in now… W > Tony > _______________________________________________ > Idr mailing list > Idr <at> ietf.org > https://www.ietf.org/mailman/listinfo/idr -- Don't be impressed with unintelligible stuff said condescendingly. -- Radia Perlman.
RSS Feed