alex chen
2015-Jan-07 04:01 UTC
[Ocfs2-devel] [PATCH V2] ocfs2: add a mount option journal_async_commit on ocfs2 filesystem
Add a mount option to support JBD2 feature:
JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT. When this feature is opened,
journal commit block can be written to disk without waiting for
descriptor blocks, which can improve journal commit performance. This
option will enable 'journal_checksum' internally.
Using the fs_mark benchmark, using journal_async_commit shows a 50%
improvement, the files per second go up from 215.2 to 317.5.
test script:
fs_mark -d /mnt/ocfs2/ -s 10240 -n 1000
default:
FSUse% Count Size Files/sec App Overhead
0 1000 10240 215.2 17878
with journal_async_commit option:
FSUse% Count Size Files/sec App Overhead
0 1000 10240 317.5 17881
Signed-off-by: Alex Chen <alex.chen at huawei.com>
Signed-off-by: Weiwei Wang <wangww631 at huawei.comm>
Reviewed-by: Joseph Qi <joseph.qi at huawei.com>
---
Documentation/filesystems/ocfs2.txt | 4 ++++
fs/ocfs2/ocfs2.h | 2 ++
fs/ocfs2/super.c | 17 +++++++++++++++++
3 files changed, 23 insertions(+)
diff --git a/Documentation/filesystems/ocfs2.txt
b/Documentation/filesystems/ocfs2.txt
index 7618a28..28f8c08 100644
--- a/Documentation/filesystems/ocfs2.txt
+++ b/Documentation/filesystems/ocfs2.txt
@@ -100,3 +100,7 @@ coherency=full (*) Disallow concurrent O_DIRECT writes,
cluster inode
coherency=buffered Allow concurrent O_DIRECT writes without EX lock among
nodes, which gains high performance at risk of getting
stale data on other nodes.
+journal_async_commit Commit block can be written to disk without waiting
+ for descriptor blocks. If enabled older kernels cannot
+ mount the device. This will enable 'journal_checksum'
+ internally.
diff --git a/fs/ocfs2/ocfs2.h b/fs/ocfs2/ocfs2.h
index 131cc1c..7e39cd6 100644
--- a/fs/ocfs2/ocfs2.h
+++ b/fs/ocfs2/ocfs2.h
@@ -284,6 +284,8 @@ enum ocfs2_mount_options
writes */
OCFS2_MOUNT_HB_NONE = 1 << 13, /* No heartbeat */
OCFS2_MOUNT_HB_GLOBAL = 1 << 14, /* Global heartbeat */
+
+ OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT = 1 << 15, /* Journal Async Commit */
};
#define OCFS2_OSB_SOFT_RO 0x0001
diff --git a/fs/ocfs2/super.c b/fs/ocfs2/super.c
index 04ee6d1..cdf5e28 100644
--- a/fs/ocfs2/super.c
+++ b/fs/ocfs2/super.c
@@ -191,6 +191,7 @@ enum {
Opt_coherency_full,
Opt_resv_level,
Opt_dir_resv_level,
+ Opt_journal_async_commit,
Opt_err,
};
@@ -222,6 +223,7 @@ static const match_table_t tokens = {
{Opt_coherency_full, "coherency=full"},
{Opt_resv_level, "resv_level=%u"},
{Opt_dir_resv_level, "dir_resv_level=%u"},
+ {Opt_journal_async_commit, "journal_async_commit"},
{Opt_err, NULL}
};
@@ -1500,6 +1502,9 @@ static int ocfs2_parse_options(struct super_block *sb,
option < OCFS2_MAX_RESV_LEVEL)
mopt->dir_resv_level = option;
break;
+ case Opt_journal_async_commit:
+ mopt->mount_opt |= OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT;
+ break;
default:
mlog(ML_ERROR,
"Unrecognized mount option \"%s\" "
@@ -1606,6 +1611,9 @@ static int ocfs2_show_options(struct seq_file *s, struct
dentry *root)
if (osb->osb_dir_resv_level != osb->osb_resv_level)
seq_printf(s, ",dir_resv_level=%d", osb->osb_resv_level);
+ if (opts & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
+ seq_printf(s, ",journal_async_commit");
+
return 0;
}
@@ -2477,6 +2485,15 @@ static int ocfs2_check_volume(struct ocfs2_super *osb)
goto finally;
}
+ if (osb->s_mount_opt & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
+ jbd2_journal_set_features(osb->journal->j_journal,
+ JBD2_FEATURE_COMPAT_CHECKSUM, 0,
+ JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
+ else
+ jbd2_journal_clear_features(osb->journal->j_journal,
+ JBD2_FEATURE_COMPAT_CHECKSUM, 0,
+ JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
+
if (dirty) {
/* recover my local alloc if we didn't unmount cleanly. */
status = ocfs2_begin_local_alloc_recovery(osb,
--
1.8.4.3
Mark Fasheh
2015-Jan-16 23:21 UTC
[Ocfs2-devel] [PATCH V2] ocfs2: add a mount option journal_async_commit on ocfs2 filesystem
On Wed, Jan 07, 2015 at 12:01:41PM +0800, alex chen wrote:> Add a mount option to support JBD2 feature: > JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT. When this feature is opened, > journal commit block can be written to disk without waiting for > descriptor blocks, which can improve journal commit performance. This > option will enable 'journal_checksum' internally. > > Using the fs_mark benchmark, using journal_async_commit shows a 50% > improvement, the files per second go up from 215.2 to 317.5. > > test script: > fs_mark -d /mnt/ocfs2/ -s 10240 -n 1000 > > default: > FSUse% Count Size Files/sec App Overhead > 0 1000 10240 215.2 17878 > > with journal_async_commit option: > FSUse% Count Size Files/sec App Overhead > 0 1000 10240 317.5 17881 > > Signed-off-by: Alex Chen <alex.chen at huawei.com> > Signed-off-by: Weiwei Wang <wangww631 at huawei.comm> > Reviewed-by: Joseph Qi <joseph.qi at huawei.com>Reviewed-by: Mark Fasheh <mfasheh at suse.de> --Mark -- Mark Fasheh