Re: A Question about Tie breaking rules (draft-ietf-idr-bgp4-17.t xt)

[Enke Chen <[e-mail]>]

Jeff,
Just consider the case of parallel sessions (using different interfaces
with different speeds) between two routers, the route selection algorithm
would not terminate without Step g).  EBGP parallel sessions are used widely
in practice.

-- Enke

> Date: Fri, 18 Jan 2002 09:22:38 -0500
> From: Jeffrey Haas <[e-mail]>
> To: Russ White <[e-mail]>
> Cc: "Reddy, Sudhakar" <[e-mail]>,
> 	"'Manav Bhatia'" <[e-mail]>, [e-mail]
> Subject: Re: A Question about Tie breaking rules (draft-ietf-idr-bgp4-17.t xt)
> Message-ID: <[e-mail]>
> References: <E7E13AAF2F3ED41197C100508BD6A328251438@india_exch.hyderabad.mindspeed.com> <[e-mail]>
> 
> On Fri, Jan 18, 2002 at 08:12:13AM -0500, Russ White wrote:
> > Actually, this rule is designed to prevent the route flap when a
> > new path, identical in all respects other than the peer it is
> > learned from, is newly learned. The older route wins, which means
> > the decision algorithm won't flip over to the newer route if the
> > route from which it is learned has a lower rid.
> 
> A different question is why you're announcing the route when
> it is identical, but just from another peer and you've already
> announced that reachability.
> 
> Flapping the stuff on the wire just because you've changed the stuff
> in your internal route instances is silly.
> 
> IOW - g) wasn't necessary at all.  g) even causes flaps where there
> shouldn't need to be any.
> 
> > Russ
> 
> -- 
> Jeff Haas 
> NextHop Technologies