Re: PATCH 2.3.26: kmalloc GFP_ZERO
Jamie Lokier (lkd@tantalophile.demon.co.uk)
Tue, 9 Nov 1999 17:37:01 +0100
Jakub Jelinek wrote:
> > Mikulas> GFP_ZERO is not so stupid idea. It allows us to implement
> > Mikulas> something like zero-page cache. While processor has nothing
> > Mikulas> to do, it can clear free pages, and when GFP_ZERO request
> > Mikulas> comes, it returns page directly without clearing overhead.
> >
> > Which I doubt is a win anyway, all you do is to sit there and thrash
> > your cache in the background.
>
> Even if you do it in a smart way (avoid trashing caches by bypassing them),
> it turned out to be very little win (I have implemented it on sparc64 some
> months ago and lmbench numbers were better just by a very small margin so I
> scratched it).
A pre-zeroing thread was discussed a few months ago, and someone
mentioned it was effective on some architecture.... Sadly I forget
which.
Anyway, it's not jus overhead that's improved with pre-zeroing. Latency
when a process suddenly needs new pages comes to mind.
-- Jamie
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/