shrink dentry cache before inode cache

The shrink_slab() function must often be called twice to get significant slab cache reduction.

shrink_slab() walks the shrinker_list to call functions that can release kernel slab memory.

The shrinker_list is walked head to tail and, as it is now, comes across the inode cache shrinker first. This releases inodes found on the inode_unused list. Afterwards the dentry cache shrinker moves many freeable inodes to the list. But those inodes are not freed until a second invocation of shrink_slab().

The dentry cache shrinker (shrink_dcache_memory()) should run before the inode cache shrinker (shrink_icache_memory()).

This can be accomplished by queuing the dentry cache shrinker earlier - simply calling inode_init() before dcache_init().

Diffed against 2.6.15-rc5

---
 fs/dcache.c |    5 ++++-
 1 files changed, 4 insertions(+), 1 deletion(-)

Index: linux-2.6.16-rc5/fs/dcache.c
===================================================================
@@ -1738,8 +1738,11 @@ void __init vfs_caches_init(unsigned lon
 	filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
 			SLAB_HWCACHE_ALIGN|SLAB_PANIC, filp_ctor, filp_dtor);
 
-	dcache_init(mempages);
 	inode_init(mempages);
+	dcache_init(mempages); /* place after inode_init so that the dentry
+				  cache shrink goes onto the shrinker list
+				  before the inode cache shrink;
+				  freeing dentry's does iput's of inodes */
 	files_init(mempages);
 	mnt_init(mempages);
 	bdev_cache_init();

Cliff Wickman
Silicon Graphics, Inc.
[email protected]
(651) 683-3824

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

Is logical, although the implementation is fragile. This came up a year or so ago and we decided it wouldn't make any difference - iirc because slab shrinking tends to occur in little bites across the whole list.

Do you have any measurements or instrumentation which show improvement?

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