Re: [PATCH 6/9] clockpro-clockpro.patch

[Marcelo Tosatti <marcelo.tosatti@cyclades.com>]

On Sun, Jan 01, 2006 at 11:37:34AM +0100, Peter Zijlstra wrote:
> On Sat, 2005-12-31 at 20:40 -0200, Marcelo Tosatti wrote:
> > On Fri, Dec 30, 2005 at 11:43:34PM +0100, Peter Zijlstra wrote:
> > > 
> > > From: Peter Zijlstra <a.p.zijlstra@chello.nl>
> > 
> > Peter,
> > 
> > I tried your "scan-shared.c" proggy which loops over 140M of a file
> > using mmap (on a 128MB box). The number of loops was configured to "5".
> > 
> > The amount of major/minor pagefaults was exactly the same between
> > vanilla and clockpro, isnt the clockpro algorithm supposed to be
> > superior than LRU in such "sequential scan of MEMSIZE+1" cases?
> 
> yes it should, hmm, have to look at that then.
> 
> What should happen is that nr_cold_target should drop to the bare
> minimum, which effectivly pins all hot pages and only rotates the few
> cold pages.

I screwed up the tests. Here are the real numbers.

Test: scan 140MB file sequentially, 5 times.
Env: 128Mb machine

CLOCK-Pro:	0:49:98elapsed	18%CPU
		7358maj+95308min

vanilla:
		1:28.05elapsed	11%CPU
		12950maj+166374min

Kicking some large arses!

-
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/