OpenZFS 9188 - increase size of dbuf cache to reduce indirect block decompression

With compressed ARC (bug 6950) we use up to 25% of our CPU to decompress
indirect blocks, under a workload of random cached reads. To reduce this
decompression cost, we would like to increase the size of the dbuf cache so
that more indirect blocks can be stored uncompressed.

If we are caching entire large files of recordsize=8K, the indirect blocks
use 1/64th as much memory as the data blocks (assuming they have the same
compression ratio). We suggest making the dbuf cache be 1/32nd of all memory,
so that in this scenario we should be able to keep all the indirect blocks
decompressed in the dbuf cache. (We want it to be more than the 1/64th that
the indirect blocks would use because we need to cache other stuff in the dbuf
cache as well.)

In real world workloads, this won't help as dramatically as the example above,
but we think it's still worth it because the risk of decreasing performance is
low. The potential negative performance impact is that we will be slightly
reducing the size of the ARC (by ~3%).

Porting Notes:
* Added modules options to zfs-module-parameters.5 man page.
* Preserved scaling based on target ARC size rather than max ARC size.

Authored by: George Wilson <[email protected]>
Reviewed by: Dan Kimmel <[email protected]>
Reviewed by: Prashanth Sreenivasa <[email protected]>
Reviewed by: Paul Dagnelie <[email protected]>
Reviewed-by: Richard Elling <[email protected]>
Reviewed-by: George Melikov <[email protected]>
Reviewed-by: Giuseppe Di Natale <[email protected]>
Ported-by: Brian Behlendorf <[email protected]>

OpenZFS-issue: https://www.illumos.org/issues/9188
OpenZFS-commit: https://github.com/openzfs/openzfs/pull/564
Upstream bug: DLPX-46942
Closes #7273 

1 files changed, 15 insertions, 11 deletions

diff --git a/module/zfs/dbuf.c b/module/zfs/dbuf.c
index 51cb0c982..7e3f0a713 100644
--- a/module/zfs/dbuf.c
+++ b/module/zfs/dbuf.c
@@ -186,10 +186,10 @@ static boolean_t dbuf_evict_thread_exit;
  */
 static multilist_t *dbuf_cache;
 static refcount_t dbuf_cache_size;
-unsigned long  dbuf_cache_max_bytes = 100 * 1024 * 1024;
+unsigned long dbuf_cache_max_bytes = 0;
 
-/* Cap the size of the dbuf cache to log2 fraction of arc size. */
-int dbuf_cache_max_shift = 5;
+/* Set the default size of the dbuf cache to log2 fraction of arc size. */
+int dbuf_cache_shift = 5;
 
 /*
  * The dbuf cache uses a three-stage eviction policy:
@@ -561,7 +561,7 @@ static inline unsigned long
 dbuf_cache_target_bytes(void)
 {
 	return MIN(dbuf_cache_max_bytes,
-	    arc_target_bytes() >> dbuf_cache_max_shift);
+	    arc_target_bytes() >> dbuf_cache_shift);
 }
 
 static inline uint64_t
@@ -787,11 +787,15 @@ retry:
 	dbuf_stats_init(h);
 
 	/*
-	 * Setup the parameters for the dbuf cache. We cap the size of the
-	 * dbuf cache to 1/32nd (default) of the size of the ARC.
+	 * Setup the parameters for the dbuf cache. We set the size of the
+	 * dbuf cache to 1/32nd (default) of the target size of the ARC. If
+	 * the value has been specified as a module option and it's not
+	 * greater than the target size of the ARC, then we honor that value.
 	 */
-	dbuf_cache_max_bytes = MIN(dbuf_cache_max_bytes,
-	    arc_target_bytes() >> dbuf_cache_max_shift);
+	if (dbuf_cache_max_bytes == 0 ||
+	    dbuf_cache_max_bytes >= arc_target_bytes()) {
+		dbuf_cache_max_bytes = arc_target_bytes() >> dbuf_cache_shift;
+	}
 
 	/*
 	 * All entries are queued via taskq_dispatch_ent(), so min/maxalloc
@@ -4216,8 +4220,8 @@ MODULE_PARM_DESC(dbuf_cache_lowater_pct,
 	"Percentage below dbuf_cache_max_bytes when the evict thread stops "
 	"evicting dbufs.");
 
-module_param(dbuf_cache_max_shift, int, 0644);
-MODULE_PARM_DESC(dbuf_cache_max_shift,
-	"Cap the size of the dbuf cache to a log2 fraction of arc size.");
+module_param(dbuf_cache_shift, int, 0644);
+MODULE_PARM_DESC(dbuf_cache_shift,
+	"Set the size of the dbuf cache to a log2 fraction of arc size.");
 /* END CSTYLED */
 #endif