Network Engineering Stack Exchange is a question and answer site for network engineers. It only takes a minute to sign up.
Sign up to join this community
I'm running a Vyatta based BGP-router with multiple transit providers which send full BGP tables. I'd like to reduce the load generated by route-calculation but want to keep some possibility of choosing AS-paths. Therefore, just asking for only default-routes is not an option.
Does it make a significant difference in performance whether to ask the transit providers to send less (partial) routes or to keep full-tables and filter with route-maps.
If the providers can filter their updates for you, then that's fewer prefixes to be sent over your links, and fewer for you to process.
A common technique is for providers to send you their AS and directly connected customer AS's, in addition to a default route. For AS's that are farther away, it may not matter which provider you use to get there.
Remember that you can control outbound traffic paths, but you're limited in what can you do for inbound traffic. So one link will probably be used more than the other. If you're really interested in balancing (outbound) traffic, you can adjust local preference for particular AS paths. But at some point, it will be more trouble than it's worth.
If you are specifically looking to speed convergence or lower CPU usage, then asking your providers if they offer a customers+peers+default route option would be the simple solution.
If your full-table convergence time/CPU load is acceptable, configuring an as-path-list to limit the number of prefixes you use by AS-Path count provides you more flexibility in the future without involving your provider(s). This can be simpler to support as the providers are providing you full information and limits the number of changes you need to request from third parties (especially important if you scale past 2 providers...) if you ever do need additional/different routing information.
I would suggest you to accept a full BGP feed from your providers since one day or another you might need to access a route which is a /24 prefix and announced by only one of your neighbours.
If you just need to insure you have a full visibility of the 400k routes of the Internet here is a personal receipe I would try to reduce your router convergence time.
Let's say you have 2 BGP neighbours AS1 and AS2. For the 2 feeds, define a prefix-list so as to reject any route that will contain the other BGP neighbour AS.
In front of AS1, reject every AS path that will contain AS2. In front of AS2, reject every AS path that will contain AS1.
Hence an AS path which is provided by your 2 neighbours will never lead to a computation. A route accessible through only one of your neighbours will never be eliminated.
The risk is that a route like: AS1 - AS2 - AS3, which is a shorter route than the one: AS2 - AS3, will be rejected. But this "coming from behind" route being the best is a rare practical case.
