Hello Kumar,

On Tue, Apr 12, 2022 at 4:13 PM Kumar Amber <[email protected]> wrote:
>
> For packets which don't already have a hash calculated,
> miniflow_hash_5tuple() calculates the hash of a packet
> using the previously built miniflow.
>
> This commit adds IPv4 profile specific hashing which
> uses fixed offsets into the packet to improve hashing
> performance.

The code lgtm.

I did not check AVX512 implementation, but as for the scalar code, I
did some benchmarks and I see:
- a limited, yet consistent accross the runs, -2% (for 1c/2t) change
in maximum throughput when doing a loopback test with one physical
port,
- a +6% when doing a pvp test (physical to vhost, looped back with testpmd),

The drop is due to the additional branch in miniflow_extract.
I don't see a way to enhance this, I tried different forms of
inline/no inline, likely/unlikely but it did not help.

Seeing how PVP gets better, I think the performance drop in the
physical only usecase is acceptable.


-- 
David Marchand

_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to