Re: Performance degradation seen after using one list for hot/cold pages.

[KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>]

On Mon, 22 Jun 2009 17:52:36 +0100
Mel Gorman <mel@csn.ul.ie> wrote:

> On Mon, Jun 22, 2009 at 11:32:03AM +0000, NARAYANAN GOPALAKRISHNAN wrote:
> > Hi,
> > 
> > We are running on VFAT.
> > We are using iozone performance benchmarking tool (http://www.iozone.org/src/current/iozone3_326.tar) for testing.
> > 
> > The parameters are 
> > /iozone -A -s10M -e -U /tmp -f /tmp/iozone_file
> > 
> > Our block driver requires requests to be merged to get the best performance.
> > This was not happening due to non-contiguous pages in all kernels >= 2.6.25.
> > 
> 
> Ok, by the looks of things, all the aio_read() requests are due to readahead
> as opposed to explicit AIO  requests from userspace. In this case, nothing
> springs to mind that would avoid excessive requests for cold pages.
> 
> It looks like the simpliest solution is to go with the patch I posted.
> Does anyone see a better alternative that doesn't branch in rmqueue_bulk()
> or add back the hot/cold PCP lists?
> 
No objection.  But 2 questions...

> -        list_add(&page->lru, list);
> +        if (likely(cold == 0))
> +            list_add(&page->lru, list);
> +        else
> +            list_add_tail(&page->lru, list);
>          set_page_private(page, migratetype);
>          list = &page->lru;
>      }

1. if (likely(coild == 0))
	"likely" is necessary ?

2. Why moving pointer "list" rather than following ?

	if (cold)
		list_add(&page->lru, list);
	else
		list_add_tail(&page->lru, list);


Thanks,
-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>