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

[Pedro Roque Marques <[e-mail]>]

Jeffrey Haas writes:

> On Fri, Jan 18, 2002 at 10:47:11AM -0800, Pedro Roque Marques wrote:
>> The conter argument is that reaching a deterministic state is worth
>> that extra flap...

> Not flap, singular, flaps plural.

> Consider:

+---+
: A :
+---+
 1 2
+---+
: B :
+---+

> Routers A and B are redundantly connected via networks 1 and 2.  A
> sends to B two routes that are equal preferable in the route
> selection process (prior to the new step g).

> Network 1 has a lower neighbor address than 2 and thus would be
> chosen.

> If network 1 flaps (bad circuit) then we unnecessarily flap the
> reachability.

Jeff,
If you consider that both link 1 and 2 have the same probability of
failure then there is 50% change that it is link 2 that flaps case in
which the first path selection change due to the tie break is helpful.
If they have different probabilities of failure then bgp needs to be
informed of that through one of the several metric parameters...

I still believe that having a deterministic path selection algorithm
(i.e. independent of temporal order of arrival) is a very significant
advantage in terms of being able to debug network/equipment.

  Pedro.