On Mon, Jun 1, 2015 at 5:05 PM, Blake Hudson <blake@ispn.net> wrote:
After studying failure modes and attempting to optimize BGP using partial routing tables, I am of the opinion that BGP with a full routing table to directly connected devices is by far the best way to gain the availability benefits of BGP. Many attempts to cost save through multi-hop BGP or traffic engineering end up breaking down when a fault occurs. Some faults, like link state, are easy to detect and work around. Other faults, like where a peer is up, but has no outside connectivity, are harder to detect if you're taking anything less than full routes.
Hi Blake, Yes, it's better to take full routes. But taking a default from two ISPs still has a reliability advantage over using a single ISP. And of course if you have two connections to the same ISP there's limited in taking full routes. Between default routes and full routes there is a range of options with increasing reliability. For example, years ago I had routers with a 256k TCAM as the BGP table approached 256k. The organization I worked for was US-centric. We needed world connectivity, but high reliability to Asia or Europe was not essential. And a large cash expenditure that year would have been bad. By slaving the APNIC /8's to a single accepted BGP route, backed by static routes for those /8's should the master BGP route fail, I maintained full connectivity while suppressing the route count to what the hardware could handle. And of course maintained maximum reliability to the destination region I most cared about. Moral of the story: if you can afford it, always take full routes. If you can't afford it, try to. If you really can't afford it, there's some trickery that can last you a year or two until you can afford it, but make sure new hardware makes it into your budget. Regards, Bill Herrin -- William Herrin ................ herrin@dirtside.com bill@herrin.us Owner, Dirtside Systems ......... Web: <http://www.dirtside.com/>