[[e-mail]: Re: [[e-mail]: Re: BGP4 stuff: Local Prefe

[[e-mail]]

*** Resending note of 09/06/96 11:12
Subject: Local Preference response
Cristina,

With respect to what local preference calculation should be used:

> The exact nature of this policy information and the computation
> involved is a local matter.

The implementations with which I am most familiar allow the router
operator to assign a fixed LOCAL_PREF value to BGP routes that match a
particular policy filter. These implementations (Cisco and gated) add an
explicit AS path length comparison step, because it is hard to
incorporate the AS path length into policy filters with static
LOCAL_PREF mappings, and minimizing AS path length is usually considered
to be desirable.

A notable exception is the Bay Networks implementation, that scales
LOCAL_PREF values according to AS path length and ORIGIN by default.

In my Simple LOcal Preference (SLOP) scheme, one can choosing the
relative priority of AS path length and ORIGIN, and there are explicit
details on how to change "SLOP factors" and such, to customize the route
selection (obviously, I'm biased). I would loosen its language to allow
options for vendor differentiation, based on comments from Dimitry and
Curtis -- allow AS_SETs to be counted as a single AS in the AS path
length, allow weights to be assigned to individual ASs, treat EGP and
Incomplete ORIGINs as identical, etc.

Any combination of the above local preference schemes in a single AS may
not be ideal, but it would produce loop-free BGP routing (so long as
there were no additional comparison steps -- such as AS path length --
which unfortunately are common in today's router implementations).

I would like to see a flexible local preference computation, such as
SLOP, included in as many implementations as possible. Why? I would like
ISPs to have more control of their BGP path selection priorities, rather
than their router vendors (through hard-coded defaults and the like)...

-- Richard Woundy, IBM