> grep conntrack /proc/slabinfo (about 4200 connections at the time) > ip_conntrack 4527 4532 352 412 412 1 > > 8192 buckets, 65536 max > > I guess the hash function fails for my setup (note: this is a test setup, > connection originating from a single ip to a single ip)
... and probably with a small subset of all possible port combinations, right? Your numbers would work out to slightly under 2 entries per hash bucket _on_average_, so if __ip_find_conntrack() gives you bad times, you clearly have a test setup not good WRT the hash function used. If this is similar to your real world situation, you have to investigate whether a different kind of hash function may be applicable. If it does not reflect your real world situation, maybe you are worrying about a synthetic benchmark effect that's not really applicable? best regards Patrick
