North American Network Operators Group

Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

Re: eBGP, iBGP, injecting networks

  • From: william(at)elan.net
  • Date: Fri Feb 20 18:08:21 2004

Ok. The way I read this is that you're redundant as far as one of your 
upstream links going down - it'd not cause complete meltdown as that 
router that had that link would still be announcing that space to the 
other router (over EBGP) and then to the net. 

What you're worrying then is what happens if actual router is down, right?
But that begs the question of how you're getting the routes that router is 
announcing in the first place. Is it coming from some other "edge" router
(that is also talking over local net to your 2nd core router)? 

If so each of your routers has complete local routes table through IBGP 
and you are not announcing it all because you're using static "network" 
statements in BGP config. In that case my suggestion would be to drop EBGP 
connection between routers and have each router announce entire ip space 
but put up 'as-path prepend' statements with the other adding the other 
router's ASN for routes that you want to be considered as being primary 
from that other router. Now exact configuration suggestion would depend on 
what hardware the routers are, i.e. is it cisco, etc. 

P.S. I've never been in situation of having to merge two ASN's or in situation
you describe, so possibly people who have would have better suggestions. 

On Fri, 20 Feb 2004 [email protected] wrote:

> 
> greetings list,
> 
> hoping someone can hook me up on the right way to do this.    
> 
> ---
> 
> we have two ASN's we control.
> 
> we have two border/edge routers (1 in each ASN) that talks to a
> different backbone provider.
> 
> the two border routers peer with eachother over eBGP and also are in
> the same OSPF process.  (we are working to merge them into the same
> BGP ASN)
> 
> my question is this:
> 
> how do we achieve router redundancy between these two routers?
> 
> currently if we lose a transit link, the traffic will flow fine out
> the other pipe.
> 
> but we don't have BGP network statements in router 2 that exist in
> router 1 and we don't have BGP network statements in router 1 that
> exist in router 2.
> 
> so the routes injected into BGP from router 1 will get withdrawn right
> if router 1 dies?
> 
> is it a problem to announce the same networks from two different eBGP
> peers to two different upstreams?
> 
> ------
> 
> if you are still reading, thanks!
> 
> to clearify some more-
> 
> current setup:
> 
> current setup:
> 
> ASN 1 (we're not Genu!ty- just using for an example)
> 
> :)
> 
> ASN 1 injects all of its own space and announces this space to
> Above.net and ASN 2
> 
> ASN 2 injects all of its own space and announces this space to Savvis
> and ASN 1.
> 
> so stuff out on the net looks like:
> 
> 1 6461 etc etc
> 
> and
> 
> 1 2 6347
> 
> -------
> 
> 2 6347 etc etc
> 
> and
> 
> 2 1 6461 etc etc
> 
> -------
> 
> so, you see we are prepending on of our AS's on the way out.
> 
> the problem is tho, we only have 1 router in each respective Autonmous
> System injecting address space.  if we lose that router, we lose
> announcing that ASN's space.
> 
> is it totally going to cause probs to have routes originating from two
> different AS's?  routing loops would be a real drag.
> 
> what about having an iBGP router in AS 1 inject the same space as the
> border router in AS 1?  this other router also peers with AS 2....
> 
> thanks a lot!
> jg
>