Re: [PATCH] mm, memcg: cg2 memory{.swap,}.peak write handlers

[Tejun Heo <tj@kernel.org>]

Hello,

On Tue, Jul 16, 2024 at 05:01:15PM +0000, Roman Gushchin wrote:
> > If we want to allow peak measurement of time periods, I wonder whether we
> > could do something similar to pressure triggers - ie. let users register
> > watchers so that each user can define their own watch periods. This is more
> > involved but more useful and less error-inducing than adding reset to a
> > single counter.
> 
> It's definitely a better user interface and I totally agree with you regarding
> the shortcomings of the proposed interface with a global reset. But if you let
> users to register a (potentially large) number of watchers, it might be quite
> bad for the overall performance, isn't it? To mitigate it, we'll need to reduce
> the accuracy of peak values. And then the question is why not just poll it
> periodically from userspace?

I haven't thought in detail but it's the same problem that PSI triggers
have, right? PSI seems to have found a reasonable trade-off across accuracy,
overhead and usage restrictions? Maybe peak usage isn't as widely useful and
it doesn't justify to engineer something as sophisticated. I don't know.

Thanks.

-- 
tejun