On Thu, 22 Oct 2015, Nick Hilliard wrote:
rib copies for each rsclient. So if you have 30k prefixes injected from a single peer, the best path calculation needs to handle ~2m rib calculations. This gets much worse for ixps with larger numbers of rsclient peers.
Hard to think how to solve the scaling issue of the per-client-RIB RS approach, while still speaking classical BGP to other peers. How did Bird solve that?
One thing that's quite possible is to run multiple bgpds, and scale out over multiple cores. With some kind of all-paths extension to bgpd (e.g. using ADD_PATH), this could still get the full-view, per-client-RIB behaviour, while still supporting classic BGP clients.
regards, -- Paul Jakma [email protected] @pjakma Key ID: 64A2FF6A Fortune: Don't try to have the last word -- you might get it. -- Lazarus Long _______________________________________________ Quagga-dev mailing list [email protected] https://lists.quagga.net/mailman/listinfo/quagga-dev
