Rehashing the idea of F/RIB Compaction

[f0o]

This is a nudge towards https://lists.frrouting.org/pipermail/frog/2019-March/000504.html - a very old request to take a look at RIB/FIB Compaction.

The specific message linked shows a valid first-iteration that would already greatly help when dealing with multiple full table providers by reducing the actual size of the RIB transmitted between iBGP drastically. The efficiency of this compression will decrease sharply in relation to the amount of peering (non-transit) sessions as routing decisions leaves the "full table" more and more fragmented.

The problem/pain point to solve here is iBGP convergence when multiple iBGP speakers have their own full-tables as well as individual peering sessions. Right now the iBGP speakers take ~15 minutes to sync and process 2.5m paths (900k v4 routes from iBGP + local variants) on 3rd Generation Xeon Platinum, during this time CPU lockup is possible and in some cases can even lead to network degradation while the FIB is churned. This scales the more iBGP speakers you have and also can cause cascading failures on weak hardware (such as low power CPEs).