Re: BGP multihoming with two address spaces

[Michael Hallgren <m.hallgren () free fr>]

Le 29/01/2014 20:34, Owen DeLong a écrit :
> > This sort of local-pref default seems to be a common practice with
> > backbones. It's very annoying. I wish they'd stop.
> Most of their customers would actually be very unhappy if they stopped. This local-pref default prevents many many 
> problems and in the vast majority of cases provides the desired result. Yes, it's important to know what is going on 
> and to be able to ask your providers to make necessary changes in circumstances where this is not optimal (either via 
> community or ticket), but I assure you that if every backbone turned this off suddenly, most internet users would be 
> very !HAPPY.

The underlying reason for this type of local preference has to do with
an assumption of cost
(which, given current transit prices, may be true or not):

        cost(customer) < 0 < cost(peer) < cost(provider) ..., thus

        local_pref(customer) > local_pref(peer) > local_pref(provider),

and as you say Owen, many actors sharing a policy simplifies our view of
things a bit. Another
way of assigning local preference of a route may factor in measured or
perceived path quality,
slightly more complex, but not a crime at all :-).

Cheers,
mh


> Owen