[Fwd: bgp route oscillation]

[Jeff Pickering <[e-mail]>]

--- Begin Message ---

Just a question on draft-walton as an oscillation solution:

Using the example 2.1 in RFC345, we have in router Ra 3 routes in
step 3):

                 AS-Path         MED        IGP-cost
x               6 100              0              13
y               6 100              1               4
z               10 100             10            5

Oscillations in this example occur because the preference rules result
in
a non-deterministic choice, more specifically because the rules result
in
choices between more than 2 routes that are non-transitive. In the
example
above:

x beats y
y beats z
z beats x

The proposal in draft-walton suggests avoiding oscillation by enabling
Ra
to always have knowledge of all three routes. However, this seems
insufficient since x can be chosen best one time and z chosen another.
To prevent oscillation,
it would seem an additional requirement to always choose the same route,
given
the same set of choices.

If this requirement is applied to the rfc3345 example, and the
draft-walton
solution applied, Ra could always choose route z as best and prevent
oscilliation.
It could also choose x as best with the same result.

But this leads me to question the usefullness of the
neighbor-as-best-path
information. For the example in question, if we required Ra to make
consistent
choices, and dont use the proposed new info, in step 1) we say y beats
z.
Step 2 remains unchanged. In step 3), we force consistency and say y
still beats z.
x then beats y so we send send a withdraw to Rd. This breaks the loop
and
there is no oscillation.

So, it would seem that the draft-walton solution is neither necessary
nor sufficient
unless of course I am missing something, which seems likely. Can anyone
enlighten me?

Jeff

--- End Message ---