[c-nsp] Load Sharing multihomed BGP with two providers

[Scott Granados]

Why don't you set up a session between router A and router B where you
basically give each router the other's full view.  Say set up a common
/30 between them or set the session up over some common network segment
where both A and B touch.  Use OSPF and the like to distribute your loop
backs and if you're taking full routes just remove the defaults.  


-----Original Message-----
From: cisco-nsp-bounces at puck.nether.net
[mailto:cisco-nsp-bounces at puck.nether.net] On Behalf Of andrew2 at one.net
Sent: Thursday, August 03, 2006 11:48 AM
To: cisco-nsp at puck.nether.net
Subject: [c-nsp] Load Sharing multihomed BGP with two providers

Trying to figure the best way to accomplish this:

I have two routers each connected to a different provider, each taking
full routes from their respective providers.  Currently all outbound
traffic is directed to Router A, so long as it is available and
announcing itself via IGP.  Router A has a default route that points to
Router B.  Since the routes learned from Provider A are more specific
than the default route, currently 99.99% of outbound traffic goes out
Provider A unless there is a problem with Router A or Provider A.  Which
is how I wanted it.

Now, however, I'd like to share some the outbound load with Provider B
on Router B.  I'd like to get the full routes from both providers on
both routers, turn on bgp multipath and see how things go.  I'm just
unsure on what the best way to get those routes into the routers is.
One idea I've been kicking around is to check with both providers and
see if they'll allow a second BGP session via ebgp multihop sessions.
In this scenario I'd be sure to inject the next_hop from each provider
into IGP and limit the max hops so that if the circuit to a provider
goes down the BGP session doesn't try to establish over the other
provider's circuit.  Any major gotchas in this scenario?  Is this
something the big "Tier 1's" typically permit?

Any better way of doing this?  Route Reflector?  iBGP?  Redistributing
the routes into an IGP seems evil, but technically possible...

Thanks for any pointers,

Andrew

_______________________________________________
cisco-nsp mailing list  cisco-nsp at puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/