Re: [RFC PATCH V3] mm readahead: Fix the readahead fail in case ofempty numa node
From: Jan Kara
Date: Mon Jan 06 2014 - 05:56:28 EST
On Mon 06-01-14 15:51:55, Raghavendra K T wrote:
> Currently, max_sane_readahead returns zero on the cpu with empty numa node,
> fix this by checking for potential empty numa node case during calculation.
> We also limit the number of readahead pages to 4k.
> Signed-off-by: Raghavendra K T <raghavendra.kt@xxxxxxxxxxxxxxxxxx>
> The current patch limits the readahead into 4k pages (16MB was suggested
> by Linus). and also handles the case of memoryless cpu issuing readahead
> failures. We still do not consider [fm]advise() specific calculations
> here. I have dropped the iterating over numa node to calculate free page
> idea. I do not have much idea whether there is any impact on big
> streaming apps.. Comments/suggestions ?
As you say I would be also interested what impact this has on a streaming
application. It should be rather easy to check - create 1 GB file, drop
caches. Then measure how long does it take to open the file, call fadvise
FADV_WILLNEED, read the whole file (for a kernel with and without your
patch). Do several measurements so that we get some meaningful statistics.
Resulting numbers can then be part of the changelog. Thanks!
> mm/readahead.c | 15 +++++++++++++--
> 1 file changed, 13 insertions(+), 2 deletions(-)
> diff --git a/mm/readahead.c b/mm/readahead.c
> index 7cdbb44..be4d205 100644
> --- a/mm/readahead.c
> +++ b/mm/readahead.c
> @@ -237,14 +237,25 @@ int force_page_cache_readahead(struct address_space *mapping, struct file *filp,
> return ret;
> +#define MAX_REMOTE_READAHEAD 4096UL
> * Given a desired number of PAGE_CACHE_SIZE readahead pages, return a
> * sensible upper limit.
> unsigned long max_sane_readahead(unsigned long nr)
> - return min(nr, (node_page_state(numa_node_id(), NR_INACTIVE_FILE)
> - + node_page_state(numa_node_id(), NR_FREE_PAGES)) / 2);
> + unsigned long local_free_page;
> + unsigned long sane_nr = min(nr, MAX_REMOTE_READAHEAD);
> + local_free_page = node_page_state(numa_node_id(), NR_INACTIVE_FILE)
> + + node_page_state(numa_node_id(), NR_FREE_PAGES);
> + /*
> + * Readahead onto remote memory is better than no readahead when local
> + * numa node does not have memory. We sanitize readahead size depending
> + * on free memory in the local node but limiting to 4k pages.
> + */
> + return local_free_page ? min(sane_nr, local_free_page / 2) : sane_nr;
Jan Kara <jack@xxxxxxx>
SUSE Labs, CR
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/