Linux 5.7 Netfilter To See AVX2 Optimizations For Big Performance Boost - Can Be Up To ~420%
Queued on Sunday in nf-next as the Netfilter staging grounds are AVX2 optimizations to provide great speed benefits.
Red Hat's Stefano Brivio has provided an AVX2-optimized nft_set_pipapo implementation. AVX2 works out well for optimizing the packet lookup routines of the Netfilter Pile Packet Policies.
In a test of injecting packets directly onto the in-going device path, an AMD EPYC 7402 "Rome" server where the testing took place was found to be significantly faster. The improvements varied from +26% to +420% with many of the tests being above the +100% range with this AVX2 implementation.
With Red Hat caring about all architectures, Stefano is also looking at a similar Arm NEON optimized version moving forward too.
This Netfilter AVX2 nft_set_pipapo implementation should see the light of day with the upcoming Linux 5.7 cycle with it already being part of nf-next.
15 Comments
