UDP/IPv6 Optimization Patches Pending For ~5% Improvement In CPU Bound Test

[phoronix]

Phoronix: UDP/IPv6 Optimization Patches Pending For ~5% Improvement In CPU Bound Test

Among the many new features in Linux 5.17 are several notable network optimizations. Optimizing network performance is a never-ending game and already for a future kernel are a new set of UDP/IPv6 optimizations being worked on...

https://www.phoronix.com/scan.php?pa...p-Optimization

[Archprogrammer]

Sounds good for future HTTP/3 adoption.

[Danny3]

Cool, but why only for IPv6?
Is it because of its address length or because IPv4 is already optimized in this area?

[foobaz]

UDP for IPv6 uses different code than UDP for IPv4. The underlying differences in the IP protocols are too great, so it was not possible to write one set of code that works for both. Both provide the same interface to programmers, and UDP looks the same to applications whether it is v4 or v6, but under the hood the way they work is different. IPv6 is not simply IPv4 with bigger addresses. It has many other improvements as well.

[Alex/AT]

Sounds good for future HTTP/3 adoption.

And

-A INPUT -p udp -m multiport --dports 80,443 -j DROP
-A OUTPUT -p udp -m multiport --dports 80,443 -j DROP
-A FORWARD -p udp -m multiport --dports 80,443 -j DROP

sounds even better.