I'm trying to understand the principles of routing in the internet, especially regarding (E)BGP.

So far, I understand the situation of a "typical" ISP (e.g. with the ASN 2) that has its own backbone, and peering/transit/customer interconnections at various points.

Since all of the border routers are connected over some internal routing protocol and backbone, if some AS with ASN 1 has a route to ASN 3 via ASN 2, there is a path (1, 2, 3), and ASN 2 will forward the traffic over its internal backbone.

However, I'm wondering if it's possible to have an ASN present at multiple locations without a backbone to connect those locations internally. Let's say another ISP, with ASN 4, is connected to both ASN 1 and 3, but not 2; also, let's assume that ASN 1 and 3 have no connectivity other than over 2. Will that work (e.g. ASN 4 has connectivity to ASNs 1, 2 and 3)? If so, how?

So far, I've found a pretty good explanation that specifically mentions that situation, and says that it can be resolved by statically configuring routes for the "split" ASN 4, but also mentions that everything has to be configured statically.

Is such a configuration common, or are such situations handled differently? Are there other ways to enable such "edge" networks? I'm thinking especially about content delivery networks with a lot of caches at various locations - do they all have a backbone, or is it done differently?


Of course, after hours of googling, I found the solution right after posting this question:

This seems to be a common configuration; by disabling BGP loop detection on the routers connecting the edge networks to their upstream ISP, connectivity between the separate subnetworks can be established.

  • 1
    If you've answered your own question, go ahead and mark it as accepted. allow-as in is the proper 'tool' for what you're trying to do in this situation. Keep in mind, though, that if your networks are really disparate, they should have their own ASN. Just because a feature works now does not mean it will always work the same way. – Aaron May 3 '13 at 22:16
  • Notably many carriers (at least by default) will never announce an AS' routes back to it, so be careful and make sure to test. Unless you're planning on deploying anycast you probably do want multiple AS'. – LapTop006 May 13 '13 at 8:32

You can either disable rejection of AS-paths with the router's own AS in it, or you can spin up a GRE/ipip tunnel between them and peer iBGP over the tunnel.

Obviously, the latter is generally preferable since allowing prefixes containing your own AS could cause nasty loops.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.