W dniu 24.02.2021 o 22:40, Alexander V. Chernikov pisze: > 24.02.2021, 10:50, "Olivier Cochard-Labbé" <oliv...@freebsd.org>: >> On Wed, Feb 24, 2021 at 1:22 AM Marek Zarychta < >> zarych...@plan-b.pwste.edu.pl> wrote: >> >>> > >>> >>> Please let me ask only one final question. Does adding: >>> >>> net.route.algo.inet.algo=dpdk_lpm4 >>> net.route.algo.inet6.algo=dpdk_lpm6 >>> >>> to /etc/sysctl.conf make any sense? I see that right FIB_ALGO is >>> automatically picked up when the module is available. >> >> My understanding is the network stack will switch to the optimum algorithm >> regarding the number of routes installed. >> cf "Automatic algorithm selection" chapter here: >> https://reviews.freebsd.org/D27401 >> >> So you don't need to set this sysctl. > To add some points to Olivier's explanation: > indeed, algorithm selection SHOULD happen automatically. > Some things to keep in mind: when the BGP sessions flaps kernel routes may go > from the full-view to near-zero and vice versa. > In that case, automatic selection will switch to lockless_radix for the range > between 10..1000 routes (or similar). Lockless radix schedules rebuild for > every change, so the speed of deleting/adding first 1k routes may be slower > than expected.
Hello Olivier and Alexander, thank you for explaining this. As far as I have tested setting this from sysctl.conf works for current fib. So to avoid early switching of route algorithms and speed up initial filling of route tables with 850k+ ip routes and 100k+ ip6 routes I'll keep both those setting in /etc/sysctl.conf. -- Marek Zarychta
OpenPGP_signature
Description: OpenPGP digital signature
