dm cache: switch the "default" cache replacement policy from mq to smq

The Stochastic multiqueue (SMQ) policy (vs MQ) offers the promise of
less memory utilization, improved performance and increased adaptability
in the face of changing workloads.  SMQ also does not have any
cumbersome tuning knobs.

Users may switch from "mq" to "smq" simply by appropriately reloading a
DM table that is using the cache target.  Doing so will cause all of the
mq policy's hints to be dropped.  Also, performance of the cache may
degrade slightly until smq recalculates the origin device's hotspots
that should be cached.

In the future the "mq" policy will just silently make use of "smq" and
the mq code will be removed.

Signed-off-by: Mike Snitzer <[email protected]>
Acked-by: Joe Thornber <[email protected]>

Documentation/device-mapper/cache-policies.txt

@@ -25,10 +25,10 @@ trying to see when the io scheduler has let the ios run.
 Overview of supplied cache replacement policies
 ===============================================
 
-multiqueue
-----------
+multiqueue (mq)
+---------------
 
-This policy is the default.
+This policy has been deprecated in favor of the smq policy (see below).
 
 The multiqueue policy has three sets of 16 queues: one set for entries
 waiting for the cache and another two for those in the cache (a set for
@@ -73,6 +73,67 @@ If you're trying to quickly warm a new cache device you may wish to
 reduce these to encourage promotion.  Remember to switch them back to
 their defaults after the cache fills though.
 
+Stochastic multiqueue (smq)
+---------------------------
+
+This policy is the default.
+
+The stochastic multi-queue (smq) policy addresses some of the problems
+with the multiqueue (mq) policy.
+
+The smq policy (vs mq) offers the promise of less memory utilization,
+improved performance and increased adaptability in the face of changing
+workloads.  SMQ also does not have any cumbersome tuning knobs.
+
+Users may switch from "mq" to "smq" simply by appropriately reloading a
+DM table that is using the cache target.  Doing so will cause all of the
+mq policy's hints to be dropped.  Also, performance of the cache may
+degrade slightly until smq recalculates the origin device's hotspots
+that should be cached.
+
+Memory usage:
+The mq policy uses a lot of memory; 88 bytes per cache block on a 64
+bit machine.
+
+SMQ uses 28bit indexes to implement it's data structures rather than
+pointers.  It avoids storing an explicit hit count for each block.  It
+has a 'hotspot' queue rather than a pre cache which uses a quarter of
+the entries (each hotspot block covers a larger area than a single
+cache block).
+
+All these mean smq uses ~25bytes per cache block.  Still a lot of
+memory, but a substantial improvement nontheless.
+
+Level balancing:
+MQ places entries in different levels of the multiqueue structures
+based on their hit count (~ln(hit count)).  This means the bottom
+levels generally have the most entries, and the top ones have very
+few.  Having unbalanced levels like this reduces the efficacy of the
+multiqueue.
+
+SMQ does not maintain a hit count, instead it swaps hit entries with
+the least recently used entry from the level above.  The over all
+ordering being a side effect of this stochastic process.  With this
+scheme we can decide how many entries occupy each multiqueue level,
+resulting in better promotion/demotion decisions.
+
+Adaptability:
+The MQ policy maintains a hit count for each cache block.  For a
+different block to get promoted to the cache it's hit count has to
+exceed the lowest currently in the cache.  This means it can take a
+long time for the cache to adapt between varying IO patterns.
+Periodically degrading the hit counts could help with this, but I
+haven't found a nice general solution.
+
+SMQ doesn't maintain hit counts, so a lot of this problem just goes
+away.  In addition it tracks performance of the hotspot queue, which
+is used to decide which blocks to promote.  If the hotspot queue is
+performing badly then it starts moving entries more quickly between
+levels.  This lets it adapt to new IO patterns very quickly.
+
+Performance:
+Testing SMQ shows substantially better performance than MQ.
+
 cleaner
 -------
 

drivers/md/dm-cache-policy-mq.c

@@ -1431,21 +1431,12 @@ static struct dm_cache_policy *mq_create(dm_cblock_t cache_size,
 
 static struct dm_cache_policy_type mq_policy_type = {
 	.name = "mq",
-	.version = {1, 3, 0},
+	.version = {1, 4, 0},
 	.hint_size = 4,
 	.owner = THIS_MODULE,
 	.create = mq_create
 };
 
-static struct dm_cache_policy_type default_policy_type = {
-	.name = "default",
-	.version = {1, 3, 0},
-	.hint_size = 4,
-	.owner = THIS_MODULE,
-	.create = mq_create,
-	.real = &mq_policy_type
-};
-
 static int __init mq_init(void)
 {
 	int r;
@@ -1455,36 +1446,21 @@ static int __init mq_init(void)
 					   __alignof__(struct entry),
 					   0, NULL);
 	if (!mq_entry_cache)
-		goto bad;
+		return -ENOMEM;
 
 	r = dm_cache_policy_register(&mq_policy_type);
 	if (r) {
 		DMERR("register failed %d", r);
-		goto bad_register_mq;
-	}
-
-	r = dm_cache_policy_register(&default_policy_type);
-	if (!r) {
-		DMINFO("version %u.%u.%u loaded",
-		       mq_policy_type.version[0],
-		       mq_policy_type.version[1],
-		       mq_policy_type.version[2]);
-		return 0;
+		kmem_cache_destroy(mq_entry_cache);
+		return -ENOMEM;
 	}
 
-	DMERR("register failed (as default) %d", r);
-
-	dm_cache_policy_unregister(&mq_policy_type);
-bad_register_mq:
-	kmem_cache_destroy(mq_entry_cache);
-bad:
-	return -ENOMEM;
+	return 0;
 }
 
 static void __exit mq_exit(void)
 {
 	dm_cache_policy_unregister(&mq_policy_type);
-	dm_cache_policy_unregister(&default_policy_type);
 
 	kmem_cache_destroy(mq_entry_cache);
 }

drivers/md/dm-cache-policy-smq.c

@@ -1748,6 +1748,15 @@ static struct dm_cache_policy_type smq_policy_type = {
 	.create = smq_create
 };
 
+static struct dm_cache_policy_type default_policy_type = {
+	.name = "default",
+	.version = {1, 0, 0},
+	.hint_size = 4,
+	.owner = THIS_MODULE,
+	.create = smq_create,
+	.real = &smq_policy_type
+};
+
 static int __init smq_init(void)
 {
 	int r;
@@ -1758,12 +1767,20 @@ static int __init smq_init(void)
 		return -ENOMEM;
 	}
 
+	r = dm_cache_policy_register(&default_policy_type);
+	if (r) {
+		DMERR("register failed (as default) %d", r);
+		dm_cache_policy_unregister(&smq_policy_type);
+		return -ENOMEM;
+	}
+
 	return 0;
 }
 
 static void __exit smq_exit(void)
 {
 	dm_cache_policy_unregister(&smq_policy_type);
+	dm_cache_policy_unregister(&default_policy_type);
 }
 
 module_init(smq_init);