Archive-Team
/
zfs
2
0
Fork 0
Code Issues Pull Requests Packages Projects Releases 103 Wiki Activity

Remove fastwrite mutex 

The fast write mutex is intended to protect accounting, but it is
redundant because all accounting is performed through atomic operations.
It also serializes all metaslab IO behind a mutex, which introduces a
theoretical scaling regression that the Illumos developers did not like
when we showed this to them. Removing it makes the selection of the
metaslab_group lock free as it is on Illumos. The selection is not quite
the same without the lock because the loop races with IO completions,
but any imbalances caused by this are likely to be corrected by
subsequent metaslab group selections.

Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #3643
This commit is contained in:
Richard Yao 2015-07-28 10:22:56 -04:00 committed by Brian Behlendorf
parent c96c36fa22
commit b10695c8f1
2 changed files with 0 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
include/sys/metaslab_impl.h
View File

@ -69,7 +69,6 @@ struct metaslab_class {
uint64_t mc_space; /* total space (alloc + free) */ uint64_t mc_space; /* total space (alloc + free) */
uint64_t mc_dspace; /* total deflated space */ uint64_t mc_dspace; /* total deflated space */
uint64_t mc_histogram[RANGE_TREE_HISTOGRAM_SIZE]; uint64_t mc_histogram[RANGE_TREE_HISTOGRAM_SIZE];
kmutex_t mc_fastwrite_lock;
}; };
/* /*

9
module/zfs/metaslab.c
View File

@ -198,7 +198,6 @@ metaslab_class_create(spa_t *spa, metaslab_ops_t *ops)
mc->mc_spa = spa; mc->mc_spa = spa;
mc->mc_rotor = NULL; mc->mc_rotor = NULL;
mc->mc_ops = ops; mc->mc_ops = ops;
mutex_init(&mc->mc_fastwrite_lock, NULL, MUTEX_DEFAULT, NULL);
return (mc); return (mc);
} }
@ -212,7 +211,6 @@ metaslab_class_destroy(metaslab_class_t *mc)
ASSERT(mc->mc_space == 0); ASSERT(mc->mc_space == 0);
ASSERT(mc->mc_dspace == 0); ASSERT(mc->mc_dspace == 0);
mutex_destroy(&mc->mc_fastwrite_lock);
kmem_free(mc, sizeof (metaslab_class_t)); kmem_free(mc, sizeof (metaslab_class_t));
} }
@ -2214,9 +2212,6 @@ metaslab_alloc_dva(spa_t *spa, metaslab_class_t *mc, uint64_t psize,
if (psize >= metaslab_gang_bang && (ddi_get_lbolt() & 3) == 0) if (psize >= metaslab_gang_bang && (ddi_get_lbolt() & 3) == 0)
return (SET_ERROR(ENOSPC)); return (SET_ERROR(ENOSPC));
if (flags & METASLAB_FASTWRITE)
mutex_enter(&mc->mc_fastwrite_lock);
/* /*
* Start at the rotor and loop through all mgs until we find something. * Start at the rotor and loop through all mgs until we find something.
* Note that there's no locking on mc_rotor or mc_aliquot because * Note that there's no locking on mc_rotor or mc_aliquot because
@ -2401,7 +2396,6 @@ top:
if (flags & METASLAB_FASTWRITE) { if (flags & METASLAB_FASTWRITE) {
atomic_add_64(&vd->vdev_pending_fastwrite, atomic_add_64(&vd->vdev_pending_fastwrite,
psize); psize);
mutex_exit(&mc->mc_fastwrite_lock);
} }
return (0); return (0);
@ -2425,9 +2419,6 @@ next:
bzero(&dva[d], sizeof (dva_t)); bzero(&dva[d], sizeof (dva_t));
if (flags & METASLAB_FASTWRITE)
mutex_exit(&mc->mc_fastwrite_lock);
return (SET_ERROR(ENOSPC)); return (SET_ERROR(ENOSPC));
} }