On Thu, Mar 09, 2023 at 05:07:40PM +0100, Claudio Jeker wrote: > On my big IXP RS test setup with 600 peers the rde_filter call in the > output path is rather expensive. By splitting the output ruleset into > output rulesets per peer the performance is improved by around 5min on a > 20min runtime. > > So when a peer is added or on config reloads the output ruleset is > compiled into a per peer output ruleset. The filter code will still > recheck these bits (from/to peer) but will no longer fail on those. > With 600 peers even skipsteps touches too many output rules and so this > results in a big boost.
This makes a lot of sense. The diff reads fine. ok tb
