Pack our DDT ZAPs a bit denser.

The DDT is really inefficient on 4k and up vdevs, because it always
allocates 4k blocks, and while compression could save us somewhat
at ashift 9, that stops being true.

So let's change the default to 32 KiB, which seems like a reasonable
compromise between improved space savings and inflated write sizes
for DDT updates.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rich Ercolani <rincebrain@gmail.com>
Closes #14654
This commit is contained in:
Rich Ercolani
2023-06-30 12:42:02 -04:00
committed by GitHub
parent 61ab05cac7
commit 2b10e32561
2 changed files with 20 additions and 3 deletions
+10 -3
View File
@@ -31,8 +31,8 @@
#include <sys/zap.h>
#include <sys/dmu_tx.h>
static const int ddt_zap_leaf_blockshift = 12;
static const int ddt_zap_indirect_blockshift = 12;
static unsigned int ddt_zap_default_bs = 15;
static unsigned int ddt_zap_default_ibs = 15;
static int
ddt_zap_create(objset_t *os, uint64_t *objectp, dmu_tx_t *tx, boolean_t prehash)
@@ -43,7 +43,7 @@ ddt_zap_create(objset_t *os, uint64_t *objectp, dmu_tx_t *tx, boolean_t prehash)
flags |= ZAP_FLAG_PRE_HASHED_KEY;
*objectp = zap_create_flags(os, 0, flags, DMU_OT_DDT_ZAP,
ddt_zap_leaf_blockshift, ddt_zap_indirect_blockshift,
ddt_zap_default_bs, ddt_zap_default_ibs,
DMU_OT_NONE, 0, tx);
return (*objectp == 0 ? SET_ERROR(ENOTSUP) : 0);
@@ -166,3 +166,10 @@ const ddt_ops_t ddt_zap_ops = {
ddt_zap_walk,
ddt_zap_count,
};
/* BEGIN CSTYLED */
ZFS_MODULE_PARAM(zfs_dedup, , ddt_zap_default_bs, UINT, ZMOD_RW,
"DDT ZAP leaf blockshift");
ZFS_MODULE_PARAM(zfs_dedup, , ddt_zap_default_ibs, UINT, ZMOD_RW,
"DDT ZAP indirect blockshift");
/* END CSTYLED */