How To Analyze Effects of Adding a New Provider to BGP Routing?

[patrick.peters]

I'm thinking about adding a circuit from another internet provider to our environment, to provide some backup in the event of a failure and, hopefully, to share the load a bit when we're not in the middle of a failure.  I'd really like to find a way to simulate what would happen with BGP if our ASN was available via that provider.  The questions running through my mind are things like:

- If my main provider goes down, how successfully will this secondary provider deliver traffic to us?  I've got a secondary internet provider today that does most of their peering with (you guessed it) my primary provider.  If my primary goes down, my secondary is hurting about as much as I am.  Not very useful...

- I live in hope that my secondary circuit might actually be the shortest route for some percentage of my traffic and help take some load off my primary.  I realize that BGP topology is a fleeting thing, but it would be nice to know what to expect.  I'm thinking of something like picking 100 random points on the internet and tracerouting them back to me with a simulated connection to the prospective ISP and see what percentage of the traffic goes via each ISP.

Does anyone know of a good way to play "what if" with the BGP routing tables in some kind of simulated environment?  

Thanks

[Philip D'Ath]

You wont be able to get this answer.

If you google "BGP Looking Glass" you'll find lots of "looking glasses" where you can examine existing connecting to existing providers, but this wont tell you about the change.

Also the global routing tables are not static, and things can change from minute to minute.

[patrick.peters]

I'm familiar with the looking glass sites, but (as you point out) they won't really let me do a "what if" analysis.

I understand that the global routing tables are constantly updating.  My theory (possibly wrong) is that they don't change in large, substantive ways from minute to minute.  I'm guessing even a snapshot would give me an idea that would be 90% accurate.  Circuits go up and down in the course of business, but I'm guessing that ISPs don't add and drop peering partners very often.

I've never looked at the details of the BGP routing protocol, but I'm wondering if a full set of routes would be enough to compute the topology of who's connected to who (which is what you'd really need to run "what if" simulations).

[Philip D'Ath]

Having a "full routing table" still won't allow you to calculate it for many reasons.  One reason being routing policy.  BGP might say the best path is via circuit 1 but the provider may implement routing policy to use circuit 2 instead.  Perhaps circuit 2 is cheaper, bigger, or better in some other way.