Re: [RFC] Optimization for use-once pages

Daniel Phillips (phillips@bonn-fries.net)
Wed, 25 Jul 2001 00:22:53 +0200


On Tuesday 24 July 2001 22:27, Marcelo Tosatti wrote:
> On Tue, 24 Jul 2001, Daniel Phillips wrote:
> > Today's patch tackles the use-once problem, that is, the problem of
> > how to identify and discard pages containing data likely to be used
> > only once in a long time, while retaining pages that are used more
> > often.
> >
> > I'll try to explain not only what I did, but the process I went
> > through to arrive at this particular approach. This requires some
> > background.
>
> Well, as I see the patch should remove the problem where
> drop_behind() deactivates pages of a readahead window even if some of
> those pages are not "used-once" pages, right ?

Yes, that was a specific goal.

> I just want to make sure the performance improvements you're seeing
> caused by the fix of this _particular_ problem.

Of course I chose the test load to show the benefit clearly, but I
also tried to avoid disturbing any of the existing cases that are
working well. I did the verification and audit in a hurry, so I
wouldn't be surprised at all if I missed something, but I feel
equally that picking these up is more a matter of adjustment than
change in viewpoint.

> If we knew the amount of non-used-once pages which drop_behind() is
> deactivating under _your_ tests, we could make absolute sure about
> the problem.

Yes, for sure, it's time to use the statistics weapon. I guess we
need to add in a couple more sensor points. An obvious place is in
check_used_once.

--
Daniel
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/