Btrfs: introduce qgroup_ulist to avoid frequently allocating/freeing ulist
PATCH 1/1 Btrfs: introduce qgroup_ulist to avoid frequently allocating/freeing ulist
When doing qgroup accounting, we call ulist_alloc()/ulist_free() every time when we want to walk qgroup tree.
By introducing 'qgroup_ulist', we only need to call ulist_alloc()/ulist_free() once. This reduce some sys time to allocate memory, see the measurements below
fsstress -p 4 -n 10000 -d $dir
With this patch:
real 0m50.153s user 0m0.081s sys 0m6.294s
real 0m51.113s user 0m0.092s sys 0m6.220s
real 0m52.610s user 0m0.096s sys 0m6.125s avg 6.213
-----------------------------------------------------
Without the patch:
real 0m54.825s user 0m0.061s sys 0m10.665s
real 1m6.401s user 0m0.089s sys 0m11.218s
real 1m13.768s user 0m0.087s sys 0m10.665s avg 10.849
we can see the sys time reduce ~43%.
Signed-off-by Wang Shilong <[email protected]>
--- fs/btrfs/ctree.h | 6 ++++ fs/btrfs/disk-io.c | 1 + fs/btrfs/qgroup.c | 70 ++++++++++++++++++++++++++------------------------- 3 files changed, 43 insertions(+), 34 deletions(-)
fs/btrfs/ctree.h +6 -0
@@ -1594,6 +1594,12 @@ struct btrfs_fs_info {struct rb_root qgroup_tree;spinlock_t qgroup_lock; +/* +* used to avoid frequently calling ulist_alloc()/ulist_free() +* when doing qgroup accounting, it must be protected by qgroup_lock. +*/ +struct ulist *qgroup_ulist; +/* protect user change for quota operations */ struct mutex qgroup_ioctl_lock;
fs/btrfs/disk-io.c +1 -0
@@ -2267,6 +2267,7 @@ int open_ctree(struct super_block *sb, fs_info->qgroup_seq =1 ; fs_info->quota_enabled =0 ; fs_info->pending_quota_state =0 ; + fs_info->qgroup_ulist =NULL ;mutex_init (&fs_info->qgroup_rescan_lock);btrfs_init_free_cluster (&fs_info->meta_alloc_cluster);
fs/btrfs/qgroup.c +36 -34
195 lines not on this page · raw message
Signature · 1 line
1.7.7.6