Routing design for DataCenter Core - Aggregation layer

vladrac-ccna · ‎07-25-2007

Hello,

I would like to know if anyone has any expirience in using eBGP beteween the SC layer and the Aggregation layer, and running OSPF only at the Aggregation layer?

Does anyone see a problem here?

Or does anyone see advantages for using BGP between SC and AG?

Let me know.

vlad

mchoo2005 · ‎07-25-2007

Why are you considering running BGP? Why not OSPF across the board?

vladrac-ccna · ‎07-25-2007

Well, BGP was selected as the exterior gateway protocol of choice because it is an open standard,

provides the capability to affect routing paths, and is designed specifically for large and scalable environments supported by multiple organizations or corporations.

Do you see any problems using BGP?

Let me know.

Vlad

mchoo2005 · ‎07-26-2007

There is no problem with running BGP. It's just that from reading your original post, I didn't see any indication that your network is "external-facing". And unless you are going to have a large network with an extraordinarily large number of routers, there is no reason to use BGP. You can use OSPF instead.

If I misread your situation (i.e. your network is external-facing), then please disregard my comment about "why BGP". If you could attach a rough diagram of what you're trying to achieve, I'll be able to advise you if you should use OSPF-BGP or pure BGP.

taylorml · ‎07-26-2007

Vlad,

Yes, there are definite scalability and functionality problems with this setup.

In order to use BGP as your only routing protocol on a router then you must have a full mesh of all you iBGP speaking routers. You can't even use a BGP route reflector (RR) in this setup because client speakers would have no Network Layer Reachability Information (NLRI) for routes they learned from the RR unless they are directly connected to the router and the to the IP the route was announced from (loopback addresses wouldn't be advertised in your setup).

It's a much better idea to run OSPF everywhere, run BGP on your border/external facing routers. Without an underlying IGP you would have to add static routes (which is a type of IGP) every time your topology changed to maintain NLRI.

Remember, BGP is Path-Vector routing protocol, which means it has similar limitations to RIP but without all the routing information. BGP relies IGPs for NLRI to get to the destination addresses it knows about from other BGP peers.

marty

vladrac-ccna · ‎07-27-2007

Hello I was planning on using OSPF on the Aggregation layer, and connect the AG layers to the core with BGP, also run BGP between the CORE and the border routers.

I dont see a problem in using OSPF, I'd like to know if you see a problem running eBGP between the core and the other layers.

Vlad

Danilo Dy · ‎07-29-2007

Hi,

The problem is the convergence time. OSPF converge faster than BGP, therefore OSPF is a good choice below Core Layer (or eBGP speaking routers). With your setup, BGP carries internet prefixes anywhere in your network while OSPF carries infrastructure Prefix. If you can afford longer convergence time, and your AG devices can take BGP routing table, I don't see a problem using BGP from AG to Core.