On Mon, May 20, 2019 at 4:09 PM Seth Mattinen <sethm@rollernet.us> wrote:
On 5/20/19 3:05 PM, William Herrin wrote:
The technique you describe was one variant of FIB Compression. It got some attention around 8 years ago on the IRTF Routing Research Group and some more attention about 5 years ago when several researchers fleshed out the possible algorithms and projected gains. As I recall they found a 30% to 60% reduction in FIB use depending on which algorithm was chosen, how many peers you had, etc.
A good start would be killing any /24 announcement where a covering aggregate exists.
Only when the routes are identical -- same origin, same path. Otherwise you're potentially throwing away your only path to that destination. And if you lose the aggregate, the /24 has to be reintroduced to the FIB. Which means you have to interlink the routes in the RIB data structure so that the update algorithm dealing with the aggregate knows there's an associated /24. There's some real subtlety a FIB Compression implementor must take in to account. Regards, Bill Herrin -- William Herrin bill@herrin.us https://bill.herrin.us/