Re: Question about atomic FIFO/LIFO queues

[Taylor R Campbell <campbell+netbsd-tech-kern%mumble.net@localhost>]

> Date: Thu, 5 Nov 2020 23:39:22 -0500
> From: Ruslan Nikolaev <nruslan_devel%yahoo.com@localhost>
> 
> Would you like us to explore if pcq can be replaced with our new DISC'19 
> algorithm potentially? How critical for the overall networking 
> performance is this code?

I don't know offhand how much of a bottleneck the pcq(9) code is -- my
guess is not much, since we use it mostly just for distributing each
input packet across CPUs once to balance load, and we try to confine
the rest of the packet-processing path to a single CPU (other than
various legacy giant-locked subsystems, many of which are still in use
and are much more important to fix).

But I don't have measurements handy.  Maybe the IIJ folks have a
better idea?  (Ozaki-san, Nakahara-san, Yamaguchi-san?)  And, of
course, if you're interested, you could certainly try to measure for
yourself!