Commit 89288623 authored by Christoph Lameter's avatar Christoph Lameter Committed by Linus Torvalds

[PATCH] Optimize off-node performance of zone reclaim

Ensure that the performance of off node pages stays the same as before.
Off node pagefault tests showed an 18% drop in performance without this
patch.

- Increase the timeout to 30 seconds to reduce the overhead.

- Move all code possible out of the off node hot path for zone reclaim
  (Sorry Andrew, the struct initialization had to be sacrificed).
  The read_page_state() bit us there.

- Check first for the timeout before any other checks.
Signed-off-by: default avatarChristoph Lameter <clameter@sgi.com>
Signed-off-by: default avatarAndrew Morton <akpm@osdl.org>
Signed-off-by: default avatarLinus Torvalds <torvalds@osdl.org>
parent 537421be
......@@ -1589,24 +1589,20 @@ int zone_reclaim_mode __read_mostly;
/*
* Mininum time between zone reclaim scans
*/
#define ZONE_RECLAIM_INTERVAL HZ/2
#define ZONE_RECLAIM_INTERVAL 30*HZ
/*
* Try to free up some pages from this zone through reclaim.
*/
int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
{
int nr_pages = 1 << order;
int nr_pages;
struct task_struct *p = current;
struct reclaim_state reclaim_state;
struct scan_control sc = {
.gfp_mask = gfp_mask,
.may_writepage = 0,
.may_swap = 0,
.nr_mapped = read_page_state(nr_mapped),
.nr_scanned = 0,
.nr_reclaimed = 0,
.priority = 0
};
struct scan_control sc;
if (time_before(jiffies,
zone->last_unsuccessful_zone_reclaim + ZONE_RECLAIM_INTERVAL))
return 0;
if (!(gfp_mask & __GFP_WAIT) ||
zone->zone_pgdat->node_id != numa_node_id() ||
......@@ -1614,12 +1610,17 @@ int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
atomic_read(&zone->reclaim_in_progress) > 0)
return 0;
if (time_before(jiffies,
zone->last_unsuccessful_zone_reclaim + ZONE_RECLAIM_INTERVAL))
return 0;
sc.may_writepage = 0;
sc.may_swap = 0;
sc.nr_scanned = 0;
sc.nr_reclaimed = 0;
sc.priority = 0;
sc.nr_mapped = read_page_state(nr_mapped);
sc.gfp_mask = gfp_mask;
disable_swap_token();
nr_pages = 1 << order;
if (nr_pages > SWAP_CLUSTER_MAX)
sc.swap_cluster_max = nr_pages;
else
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment