Re: [Ext2-devel] Re: inode cache, dentry cache, buffer heads usage
From: Sonny Rao
Date: Wed Mar 09 2005 - 17:40:03 EST
On Thu, Mar 10, 2005 at 03:23:49AM +0530, Dipankar Sarma wrote:
> On Wed, Mar 09, 2005 at 01:29:23PM -0800, Badari Pulavarty wrote:
> > On Wed, 2005-03-09 at 13:27, Dipankar Sarma wrote:
> > > On Wed, Mar 09, 2005 at 10:55:58AM -0800, Badari Pulavarty wrote:
> > > > Hi,
> > > >
> > > > We have a 8-way P-III, 16GB RAM running 2.6.8-1. We use this as
> > > > our server to keep source code, cscopes and do the builds.
> > > > This machine seems to slow down over the time. One thing we
> > > > keep noticing is it keeps running out of lowmem. Most of
> > > > the lowmem is used for ext3 inode cache + dentry cache +
> > > > bufferheads + Buffers. So we did 2:2 split - but it improved
> > > > thing, but again run into same issues.
> > > >
> > > > So, why is these slab cache are not getting purged/shrinked even
> > > > under memory pressure ? (I have seen lowmem as low as 6MB). What
> > > > can I do to keep the machine healthy ?
> > >
> > > How does /proc/sys/fs/dentry-state look when you run low on lowmem ?
> >
> >
> >
> > badari@kernel:~$ cat /proc/sys/fs/dentry-state
> > 1434093 1348947 45 0 0 0
> > badari@kernel:~$ grep dentry /proc/slabinfo
> > dentry_cache 1434094 1857519 144 27 1 : tunables 120
> > 60 8 : slabdata 68797 68797 0
>
> Hmm.. so we are not shrinking dcache despite a large number of
> unsed dentries. That is where we need to look. Will dig a bit
> tomorrow.
Here's my really old patch where I saw some improvement for this scenario...
I haven't tried this in a really long time, so I have no idea if it'll
work :-)
Sonny
--- fs/dcache.c.original 2004-08-02 15:43:42.629539312 -0500
+++ fs/dcache.c 2004-08-03 18:16:45.007809144 -0500
@@ -31,6 +31,7 @@
#include <linux/seqlock.h>
#include <linux/swap.h>
#include <linux/bootmem.h>
+#include <linux/rbtree.h>
/* #define DCACHE_DEBUG 1 */
@@ -60,12 +61,61 @@ static unsigned int d_hash_mask;
static unsigned int d_hash_shift;
static struct hlist_head *dentry_hashtable;
static LIST_HEAD(dentry_unused);
+static struct rb_root dentry_tree = RB_ROOT;
+
+#define RB_NONE (2)
+#define ON_RB(node) ((node)->rb_color != RB_NONE)
+#define RB_CLEAR(node) ((node)->rb_color = RB_NONE )
/* Statistics gathering. */
struct dentry_stat_t dentry_stat = {
.age_limit = 45,
};
+
+/* take a dentry safely off the rbtree */
+static void drb_delete(struct dentry* dentry)
+{
+ // printk("drb_delete: 0x%p (%s) proc %d\n",dentry,dentry->d_iname,smp_processor_id());
+ if (ON_RB(&dentry->d_rb)) {
+ rb_erase(&dentry->d_rb, &dentry_tree);
+ RB_CLEAR(&dentry->d_rb);
+ } else {
+ /* All allocated dentry objs should be in the tree */
+ BUG_ON(1);
+ }
+}
+
+static
+struct dentry * drb_insert(struct dentry * dentry)
+{
+ struct rb_node ** p = &dentry_tree.rb_node;
+ struct rb_node * parent = NULL;
+ struct rb_node * node = &dentry->d_rb;
+ struct dentry * cur = NULL;
+
+ // printk("drb_insert: 0x%p (%s)\n",dentry,dentry->d_iname);
+
+ while (*p)
+ {
+ parent = *p;
+ cur = rb_entry(parent, struct dentry, d_rb);
+
+ if (dentry < cur)
+ p = &(*p)->rb_left;
+ else if (dentry > cur)
+ p = &(*p)->rb_right;
+ else {
+ return cur;
+ }
+ }
+
+ rb_link_node(node, parent, p);
+ rb_insert_color(node,&dentry_tree);
+ return NULL;
+}
+
+
static void d_callback(struct rcu_head *head)
{
struct dentry * dentry = container_of(head, struct dentry, d_rcu);
@@ -189,6 +239,7 @@ kill_it: {
list_del(&dentry->d_child);
dentry_stat.nr_dentry--; /* For d_free, below */
/*drops the locks, at that point nobody can reach this dentry */
+ drb_delete(dentry);
dentry_iput(dentry);
parent = dentry->d_parent;
d_free(dentry);
@@ -351,6 +402,7 @@ static inline void prune_one_dentry(stru
__d_drop(dentry);
list_del(&dentry->d_child);
dentry_stat.nr_dentry--; /* For d_free, below */
+ drb_delete(dentry);
dentry_iput(dentry);
parent = dentry->d_parent;
d_free(dentry);
@@ -360,7 +412,7 @@ static inline void prune_one_dentry(stru
}
/**
- * prune_dcache - shrink the dcache
+ * prune_lru - shrink the lru list
* @count: number of entries to try and free
*
* Shrink the dcache. This is done when we need
@@ -372,7 +424,7 @@ static inline void prune_one_dentry(stru
* all the dentries are in use.
*/
-static void prune_dcache(int count)
+static void prune_lru(int count)
{
spin_lock(&dcache_lock);
for (; count ; count--) {
@@ -410,6 +462,93 @@ static void prune_dcache(int count)
spin_unlock(&dcache_lock);
}
+/**
+ * prune_dcache - try and "intelligently" shrink the dcache
+ * @requested - num of dentrys to try and free
+ *
+ * The basic strategy here is to scan through our tree of dentrys
+ * in-order and put them at the end of the lru - free list
+ * Why in-order? Because, we want the chances of actually freeing
+ * all 15-27 (depending on arch) dentrys on a given page, instead
+ * of just in random lru order, which tends to lower dcache utilization
+ * and not free many pages.
+ */
+static void prune_dcache(unsigned requested)
+{
+ /* ------ debug --------- */
+ //static int mod = 0;
+ //int flag = 0, removed = 0;
+ /* ------ debug --------- */
+
+ unsigned found = 0;
+ unsigned count;
+ struct rb_node * next;
+ struct dentry *dentry;
+#define NUM_LRU_PTRS 8
+ struct rb_node *lru_ptrs[NUM_LRU_PTRS];
+ struct list_head *cur;
+ int i;
+
+ spin_lock(&dcache_lock);
+
+ cur = dentry_unused.prev;
+
+ /* grab NUM_LRU_PTRS entrys off the end of lru list */
+ /* we'll use these as pseudo-random starting points in the tree */
+ for (i = 0 ; i < NUM_LRU_PTRS ; i++ ){
+ if ( cur == &dentry_unused ) {
+ /* if there aren't NUM_LRU_PTRS entrys, we probably
+ can't even free a page now, give up */
+ spin_unlock(&dcache_lock);
+ return;
+ }
+ lru_ptrs[i] = &(list_entry(cur,struct dentry, d_lru)->d_rb);
+ cur = cur->prev;
+ }
+
+ i = 0;
+
+ do {
+ count = 4 * PAGE_SIZE / sizeof(struct dentry) ; /* abitrary heuristic */
+ next = lru_ptrs[i];
+ for (; count ; count--) {
+ if( ! next ) {
+ //flag = 1; /* ------ debug --------- */
+ break;
+ }
+ dentry = list_entry(next, struct dentry, d_rb);
+ next = rb_next(next);
+ prefetch(next);
+ if( ! list_empty( &dentry->d_lru) ) {
+ list_del_init(&dentry->d_lru);
+ dentry_stat.nr_unused--;
+ }
+ if (atomic_read(&dentry->d_count)) {
+ //removed++; /* ------ debug --------- */
+ continue;
+ } else {
+ list_add_tail(&dentry->d_lru, &dentry_unused);
+ dentry_stat.nr_unused++;
+ found++;
+ }
+ }
+ i++;
+ } while ( (found < requested / 2) && (i < NUM_LRU_PTRS ) );
+#undef NUM_LRU_PTRS
+
+ spin_unlock(&dcache_lock);
+
+ /* ------ debug --------- */
+ //mod++;
+ //if ( ! (mod & 64) ) {
+ // mod = 0;
+ // printk("prune_dcache: i %d flag %d, found %d removed %d\n",i,flag,found,removed);
+ //}
+ /* ------ debug --------- */
+
+ prune_lru(found);
+}
+
/*
* Shrink the dcache for the specified super block.
* This allows us to unmount a device without disturbing
@@ -604,7 +743,7 @@ void shrink_dcache_parent(struct dentry
int found;
while ((found = select_parent(parent)) != 0)
- prune_dcache(found);
+ prune_lru(found);
}
/**
@@ -642,7 +781,7 @@ void shrink_dcache_anon(struct hlist_hea
}
}
spin_unlock(&dcache_lock);
- prune_dcache(found);
+ prune_lru(found);
} while(found);
}
@@ -730,6 +869,7 @@ struct dentry *d_alloc(struct dentry * p
if (parent)
list_add(&dentry->d_child, &parent->d_subdirs);
dentry_stat.nr_dentry++;
+ drb_insert(dentry);
spin_unlock(&dcache_lock);
return dentry;
--- include/linux/dcache.h.original 2004-08-03 18:20:40.800963136 -0500
+++ include/linux/dcache.h 2004-08-02 15:36:19.886846432 -0500
@@ -9,6 +9,7 @@
#include <linux/cache.h>
#include <linux/rcupdate.h>
#include <asm/bug.h>
+#include <linux/rbtree.h>
struct nameidata;
struct vfsmount;
@@ -94,6 +95,7 @@ struct dentry {
struct hlist_head *d_bucket; /* lookup hash bucket */
struct qstr d_name;
+ struct rb_node d_rb;
struct list_head d_lru; /* LRU list */
struct list_head d_child; /* child of parent list */
struct list_head d_subdirs; /* our children */