On Tue, 14 May 2024 at 19:19, Yafang Shao <laoar.shao@xxxxxxxxx> wrote:
>
> I believe this change is still necessary. Would you prefer to commit
> it directly, or should I send an official patch?
Sending an official patch just for people to try out sounds good
regardless, but I'd really like some real performance testing on other
loads too before just taking it.
It *may* be acceptable, and it's certainly simple. It's also almost
certainly safe from a correctness angle.
But it might regress performance on other important loads even if it
fixes an issue on your load.
That's why we had the whole discussion about alternatives.
I'm still of the opinion that we should probably *try* this simple
approach, but I really would hope we could have some test tree that
people run a lot of benchmarks on.
Does anybody do filesystem benchmarking on the mm tree? Or do we have
some good tree to just give it a good testing? It feels a bit
excessive to put it in my development tree just to get some
performance testing coverage, but maybe that's what we have to do..
Linus
