On Mon, 11 Sep 2000, Neil Schemenauer wrote:
> On Mon, Sep 11, 2000 at 01:12:32PM -0300, Rik van Riel wrote:
> > Your idea /heavily/ penalises libc and executable pages by aging them
> > more often than anonymous pages...
>
> I don't think I age anonymous pages any more than any other type
> of page.
Think again.
You're aging them both in try_to_swap_out() /and/ in
shrink_mmap().
> Perhaps you are saying that shared pages should recieve some
> bonus?
No. I'm saying shared pages should have the accessed bits
propagated and be only aged once. I know we can't handle
this right for 2.4, but for 2.5 I hope to use physical-page
based page aging to get this one right...
> That is a different issue and it is handled naturally with my
> patch. If shared pages are actually used then PageTouch() will
> be called on them more often.
This is /not/ the case. Think of a page from libc, which
is mapped by 30 processes. Now imagine that page is being
heavily used and was used by 5 processes since we scanned
it the last time.
With your patch we'd age the page down 25 (!!) times and
only age it up 5 times. This is clearly not what you want
for a page which was used by 5 different processes since
the last time we scanned it...
regards,
Rik
--
"What you're running that piece of shit Gnome?!?!"
-- Miguel de Icaza, UKUUG 2000
http://www.conectiva.com/ http://www.surriel.com/
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Fri Sep 15 2000 - 21:00:15 EST