On 21/09/2009, at 7:37 AM, Fouant, Stefan wrote:
I don't know if you want to arbitrarily use local-pref and AS-Path prepend in a one-size-fits-all approach, as under certain scenarios it might be more beneficial to route traffic between POPs to take advantage of routes via shortest AS Path or other constraints. Why not just extend your IGP across all POPs and set inter-POP links to a higher metric? In this scenario, if a given route is received via muliple POPs and all things being equal (AS Path, etc.), you'll prefer to route traffic out the shortest-cost path to a POP exit point.
Sorry for the top post, I'm on my BB.
Stefan Fouant Neustar, Inc. / Principal Engineer 46000 Center Oak Plaza Sterling, VA 20166 Office: +1.571.434.5656 ▫ Mobile: +1.202.210.2075 ▫ GPG ID: 0xB5E3803D ▫ stefan.fouant@neustar.biz
I agree with Stefan. You are better off extending your IGP across PoPs as it will give you more flexibility in the long run. If you ever want to go down the path of traffic engineering / MPLS / etc, you will find it much easier as this will allow for CSPF and multi-topology routing architectures. Truman
----- Original Message ----- From: Rick Ernst <nanog@shreddedmail.com> To: nanog@nanog.org <nanog@nanog.org> Sent: Sun Sep 20 16:17:41 2009 Subject: Multi-POP design check/help question
Cross-posted from cisco-nsp. We are a (mostly) Cisco shop, but I'm looking more for BCP and overall design, not provisioning specifics.
-----
My Cisco bookshelf isn't helping me much with this...
We currently have a single POP with border/core/aggregation topology. Upstreams each come in on their own border router and the core is used as a route-reflector for border and aggregation. The internal network uses OSPF as an IGP and each device is dual-connected for redundancy on independent layer-2 networks. OSPF load-shares with loopback IPs and IBGP uses the loopback addresses for peering.
We are looking at turning up two additional POPs in the metro area, each connected by redundant GigE loops to the original POP. Each POP may have zero or more direct upstream connections. I'd like local traffic at each POP to prefer both in and outbound traffic via the local upstream, but still be able to failover to upstreams at other POPs if needed.
My initial thoughts are to BGP peer between POPs with a higher local- pref for the local outbound traffic and to prepend between the POPs so inbound traffic is more likely to take the shortest path inbound.
Is this too simplistic? Prone to trouble? What gotchas should I be looking at, or other designs should I be considering?