BGP redistribution into iGP for optimal routing

Hi ]\/[arco,

Thanks for your reply. I'm having a read into GBLP now. Am I correct in saying its just a gateway load balancing protocol. I.e the routing decisions arent educated they are just to split the load over multiple gateways. The traffic that hit those routers though would probably have a good chance of needing to be forwarded to another? I.e better then my current setup but not ideal.

My thoughts behind re-distributing some of the routes onto the 3750 was so that it can make an optimal routing decision (I believe it can route somewhat rapidly as it uses hardware rather than software?) - so if there was 3 routers (2 transit, 1 peering) it sends the traffic to the ideal gateway.

Know any good consultants who can help with these kinda questions 'offline'?

Thanks,
Mike

Just to follow this up - at the moment all the clients gateways are set to the VLAN IP's on the 3750 stack. GBLP appears from what I've read work by giving a different MAC out as a reply to different ARP requests. Surely in my setup where it is the cisco 3750 issuing the ARP request for the virtual IP (and then caching it?) it would not result in the desired effect? (I understand how it would work if hosts were issuing those queries though...).

Hi JWCastile,

Thanks for your input. I'll check out the URL you linked to - not seen that before.

We take full routes at the moment. No firewall between router and 3750 - we firewall either on the end devices or in between the 3750 and client vlan on some other kit.

Just clarify your comment "directly peer with your 3750" - as in set up an iBGP peer between the 3750 and routers as I suggested above with the necessary filters to restrict the number of routes?

Thanks,
Mike

You can peer eBGP or iBGP, just remember to do next-hop-self. if you do iBGP  There's no harm in configuring a private AS on your 3750 and peering that way.  Regardless of how you do the peering, you'll want to configure filters.

Hi Mike,

yes GBLP is just like HSRP (it provides redundancy for the gateway) but just is able to load balance the traffic between the two or more gateways. The final routing decision is still done by the routers. Of course based on your upstream you'll need to tweak it. i.e. if now one of the borders is forwarding 80% of the traffic because it has better paths, and the other is forwarding the other 20%, you can tweak the weights to that 80% would go to the router with better upstream connectivity, and save bandwidth on your internal path.
Only your 3750s will see the different mac addresses of your routers, so yes, you can achieve load balancing from the 3750 point of view, not from the clients point of view. Still better than nothing of course! ;)
If you prefer to go the routing way, I'd just make the 3750 speak eBGP with the other routes via a private AS. But now you'll have to choose which subset of the routing table giving to them! 3750s support at most 11000 routes in the sdm routing template, so you'll probably want to feed them just with your providers route, and some big providers one. They'll never be able to make the 'perfect' decision, as they won't have the full table anyway!

Cheers,
]\/[arco

Hi ]\/[arco,

Ok - just been playing with some regexp's for say 2 as path in length and its actually quite tricky to fit in with that 11K - so should be interesting that one :-). Primarily I'm only really after the peering routes as In this scenario that should take the majority of the traffic luckily.

Re eBGP using private AS - presumably the 3750 on private AS advertises no routes in this config - just receives and filters. Out of interest what is the advantage of doing eBGP rather than iBGP in this scenario?

Presumably on the routers I only advertise the regexp'd routes and I put the same filters on the 3750 along with a maximum prefix statement to prevent it ever over flowing the TCAM?

Could I also replace my HSRP setup in this scenario by advertising default on both routers to the 3750? or better to leave in place with corresponding static route as a last resort?

Thanks,
Mike