Re: [RFC net-next 0/5] Suspend IRQs during preferred busy poll

From: Joe Damato
Date: Wed Aug 14 2024 - 11:46:48 EST


On Wed, Aug 14, 2024 at 11:08:45AM -0400, Willem de Bruijn wrote:
> Joe Damato wrote:

[...]

> > On Tue, Aug 13, 2024 at 11:16:07PM -0400, Willem de Bruijn wrote:
> > Using less CPU to get comparable performance is strictly better, even if a
> > system can theoretically support the increased CPU/power/cooling load.
>
> If it is always a strict win yes. But falling back onto interrupts
> with standard moderation will not match busy polling in all cases.
>
> Different solutions for different workloads. No need to stack rank
> them. My request is just to be explicit which design point this
> chooses, and that the other design point (continuous busy polling) is
> already addressed in Linux kernel busypolling.

Sure, sounds good; we can fix that in the cover letter.

Thanks for taking a look.