slub: Disable NUMA remote node defragmentation by default
authorChristoph Lameter <cl@linux-foundation.org>
Tue, 19 Aug 2008 13:51:22 +0000 (08:51 -0500)
committerPekka Enberg <penberg@cs.helsinki.fi>
Wed, 20 Aug 2008 18:50:21 +0000 (21:50 +0300)
Switch remote node defragmentation off by default. The current settings can
cause excessive node local allocations with hackbench:

  SLAB:

    % cat /proc/meminfo
    MemTotal:        7701760 kB
    MemFree:         5940096 kB
    Slab:             123840 kB

  SLUB:

    % cat /proc/meminfo
    MemTotal:        7701376 kB
    MemFree:         4740928 kB
    Slab:            1591680 kB

[Note: this feature is not related to slab defragmentation.]

You can find the original discussion here:

  http://lkml.org/lkml/2008/8/4/308

Reported-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Tested-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Christoph Lameter <cl@linux-foundation.org>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
mm/slub.c

index 4f5b96149458a6712d1592fd64bdf4a4b876f38a..fb486d5540f8bb0a7110245f2368fe2ace28338d 100644 (file)
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -2312,7 +2312,7 @@ static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags,
 
        s->refcount = 1;
 #ifdef CONFIG_NUMA
-       s->remote_node_defrag_ratio = 100;
+       s->remote_node_defrag_ratio = 1000;
 #endif
        if (!init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA))
                goto error;
@@ -4058,7 +4058,7 @@ static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
        if (err)
                return err;
 
-       if (ratio < 100)
+       if (ratio <= 100)
                s->remote_node_defrag_ratio = ratio * 10;
 
        return length;