[PATCH v8 00/14] Multi-Gen LRU Framework

[Yu Zhao]

On Tue, Mar 8, 2022 at 5:07 PM Linus Torvalds
<torvalds at linux-foundation.org> wrote:
>
> On Tue, Mar 8, 2022 at 3:48 PM Yu Zhao <yuzhao at google.com> wrote:
> >
> > The current page reclaim is too expensive in terms of CPU usage and it
> > often makes poor choices about what to evict. This patchset offers an
> > alternative solution that is performant, versatile and
> > straightforward.
>
> So apart from my complaints about asking users config questions that
> simply should not be asked, I really think this just needs to start
> getting merged.
>
> We've seen several numbers on the upsides, and I don't think we'll see
> any of the downsides until we try it. And I don't think there is any
> question that we _shouldn't_ try it, given the numbers posted.
>
> But yeah, I certainly _hope_ that all the benchmarking has been done
> with a unified set of config values, and it's not some kind of bogus
> "cherry-picked config values for this particular machine" kind of
> benchmarking that has been done.
>
> Because that isn't valid benchmarking - comparing some "tuned for this
> paeticular machine or load" setup to a default one is just not worth
> even setting numbers to, and debases the whole value of posting
> results.

All benchmarks were done with the default config values. I'm removing
those config options now.

This sounds self-serving: our data centers want them, so I had to try.