> On 13 Jul 2017, at 12:27, Hrvoje Popovski <hrv...@srce.hr> wrote: > > On 13.7.2017. 0:26, Per-Olov Sjöholm wrote: >> I increased net.inet.ip.ifq.maxlen in steps of 256… I had to increase the >> net.inet.ip.ifq.maxlen 9 times to 2309 for the net.inet.ip.ifq.drops to stop >> increasing. At a maxlen of 2309 the drops stopped completley. But all values >> of net.inet.ip.ifq.maxlen higher than 756 did not give any performance >> boost (well… Not that I could see). At maxlen of 756 and over, the below >> output represents the average tests very well when testing against the ISP >> test servers. Yes I love my OpenBSD FW :) :) :) > > > maybe this sysctls would give better performance? > > kern.pool_debug=0 > net.inet.ip.ifq.maxlen=8192 > > or update to latest current and if you're brave enough compile kernel > with "option WITH_PF_LOCK" >
AMD64 SMP kernel gives better throughput performance than UNI kernel (not much though). >>> That I do not understand??? <<< net.inet.ip.ifq.maxlen=4096 is enough to avoid drops at Gbit speed for me. To avoid increased latency, I stop increasing it as the drops has stopped at 4096…. The below very well represent the average performance from my linux server behind my OpenBSD 6.0 FW against the internet test servers. I have had even betters and a few worse tests. [root@server2 ~]# bbk_cli --live Start: 2017-07-16 03:19:10 ISP: Bahnhof Internet AB Support ID: sth60128e5b7 Latency: 11.291 ms Download: 836.488 Mbit/s Upload: 947.120 Mbit/s Subscription: 500-1000 Mbit/s fiber I have only tuned the net.inet.ip.ifq.maxlen in the OpenBSD FW sysctl.conf due to the drops I saw. No other settings in sysctl.conf (well except ip and ipv6 forwarding…). Maybe I should be satisfied with this performance :)
