On Sun, 10 Mar 2024 21:19:50 -0700 Shradha Gupta <shradhagupta@xxxxxxxxxxxxxxxxxxx> wrote: > On Fri, Mar 08, 2024 at 11:22:44AM -0800, Jakub Kicinski wrote: > > On Fri, 8 Mar 2024 18:51:58 +0000 Haiyang Zhang wrote: > > > > Dynamic is a bit of an exaggeration, right? On a well-configured system > > > > each CPU should use a single queue assigned thru XPS. And for manual > > > > debug bpftrace should serve the purpose quite well. > > > > > > Some programs, like irqbalancer can dynamically change the CPU affinity, > > > so we want to add the per-CPU counters for better understanding of the CPU > > > usage. > > > > Do you have experimental data showing this making a difference > > in production? > Sure, will try to get that data for this discussion > > > > Seems unlikely, but if it does work we should enable it for all > > devices, no driver by driver. > You mean, if the usecase seems valid we should try to extend the framework > mentioned by Rahul (https://lore.kernel.org/lkml/20240307072923.6cc8a2ba@xxxxxxxxxx/) > to include these stats as well? > Will explore this a bit more and update. Thanks. > Remember, statistics aren't free, and even per-cpu stats end up taking cache space.
