[c-nsp] Growing BGP tables

[lee.e.rian at census.gov]

On 11/30/2004 10:30 PM, Rodney Dunn <rodunn at cisco.com> wrote:

> I'm trying to put it together in my head
> how it would be done for dual EBGP sessions.
>
> ie:
>
> ISPA                    ISPB
>  |                       |
> RouterA  --- IBGP --- RouterB
>
> Say I get a /16 and a /24 for a prefix from ISPA with
> the same next hop so I wouldn't want to install the
> /24.  But what happens if I get a /16 from ISPB and
> that gets sent to RouterA.  That prefix would have
> a next hop of Router B or ISPB so it would be a different
> prefix and installed in the RIB.  Now the traffic that
> was originally flowing to the /24 would take the backup
> path (due to a longest match lookup) rather than the
> path it would have taken if we had installed the original
> /24. Would that be acceptable?

I think not.

Say RouterA advertises a /16 and a /24 within the /16 to ISPA, RouterB
advertises the same /16 and a different /24 within the /16 to ISPB.

If I'm understanding the proposal correctly, ISPA drops the /24 since it
has the same next hop as the /16.  ISPB also drops their /24 since it has
the same next hop as the /16.  If the link between RouterA and RouterB goes
down then ISPA, and all their customers, aren't going to be able to get to
siteB (the /24 advertised by RouterB) while ISPB, and all their customers,
aren't going to be able to get to siteA (the /24 advertised by RouterA).

Lee