This patch series adds support for fast commits which is a simplified
version of the scheme proposed by Park and Shin, in their paper,
"iJournaling: Fine-Grained Journaling for Improving the Latency of
Fsync System Call"[1]. The basic idea of fast commits is to make JBD2
give the client file system an opportunity to perform a faster
commit. Only if the file system cannot perform such a commit
operation, then JBD2 should fall back to traditional commits.
Because JBD2 operates at block granularity, for every file system
metadata update it commits all the changed blocks to the journal at
commit time. This is inefficient because updates to some blocks that
JBD2 commits are derivable from some other blocks. For example, if a
new extent is added to an inode, then corresponding updates to the
inode table, the block bitmap, the group descriptor and the superblock
can be derived based on just the extent information and the
corresponding inode information. So, if we take this relationship
between blocks into account and replay the journalled blocks smartly,
we could increase performance of file system commits significantly.
Fast commits introduced in this patch has two main contributions:
(1) Making JBD2 fast commit aware, so that clients of JBD2 can
implement fast commits
(2) Add support in ext4 to use JBD2's new interfaces and implement
fast commits
Testing
-------
e2fsprogs was updated to set fast commit feature flag and to ignore
fast commit blocks during e2fsck.
https://github.com/harshadjs/e2fsprogs.git
After applying all the patches in this series, following runs of
xfstests were performed:
- kvm-xfstest.sh -g log -c 4k
- kvm-xfstests.sh smoke
All the log tests were successful and smoke tests didn't introduce any
additional failures.
Performance Evaluation
----------------------
In order to evaluate fast commit performance we used fs_mark
benchmark. We updated fs_mark benchmark to send fsync() calls after
every write operation.
https://github.com/harshadjs/fs_mark.git
Following are the results that we got:
Write performance measured in MB/s with 4 parallel threads file sizes
(X) vs write unit sizes (Y).
Without Fast Commit:
|-----+------+------+------|
| | 32k | 128k | 256k |
|-----+------+------+------|
| 4k | 0.27 | 0.25 | 0.24 |
| 8k | 0.45 | 0.51 | 0.46 |
| 32k | 2.15 | 2.23 | 2.28 |
|-----+------+------+------|
With Fast Commit:
|-----+------+------+------|
| | 32k | 128k | 256k |
|-----+------+------+------|
| 4k | 0.74 | 1.42 | 1.94 |
| 8k | 1.52 | 1.88 | 2.48 |
| 32k | 1.8 | 4.29 | 7.38 |
|-----+------+------+------|
On an average, fast commits increased file system write performance by
280% on modified fs_mark benchmark.
Harshad Shirwadkar(13):
docs: Add fast commit documentation
ext4: fast-commit recovery path changes
ext4: fast-commit commit path changes
ext4: fast-commit commit range tracking
ext4: track changed files for fast commit
ext4: add fields that are needed to track changed files
jbd2: fast-commit recovery path changes
jbd2: fast-commit commit path new APIs
jbd2: fast-commit commit path changes
jbd2: fast commit setup and enable
jbd2: add fast commit fields to journal_s structure
ext4: add handling for extended mount options
ext4: add support fast commit
Documentation/filesystems/ext4/journal.rst | 78 ++
Documentation/filesystems/journalling.rst | 15
fs/ext4/acl.c | 1
fs/ext4/balloc.c | 7
fs/ext4/ext4.h | 87 +++
fs/ext4/ext4_jbd2.c | 92 +++
fs/ext4/ext4_jbd2.h | 29 +
fs/ext4/extents.c | 44 +
fs/ext4/fsync.c | 2
fs/ext4/ialloc.c | 1
fs/ext4/inline.c | 17
fs/ext4/inode.c | 62 +-
fs/ext4/ioctl.c | 3
fs/ext4/mballoc.c | 83 ++
fs/ext4/mballoc.h | 2
fs/ext4/migrate.c | 1
fs/ext4/namei.c | 14
fs/ext4/super.c | 538 ++++++++++++++++++-
fs/ext4/xattr.c | 1
fs/jbd2/checkpoint.c | 2
fs/jbd2/commit.c | 85 ++-
fs/jbd2/journal.c | 230 +++++++-
fs/jbd2/recovery.c | 70 ++
fs/jbd2/transaction.c | 6
fs/ocfs2/alloc.c | 2
fs/ocfs2/journal.c | 4
fs/ocfs2/super.c | 2
include/linux/jbd2.h | 106 +++
include/trace/events/ext4.h | 59 ++
include/trace/events/jbd2.h | 9
30 files changed, 1561 insertions(+), 91 deletions(-)
--
2.23.0.rc1.153.gdeed80330f-goog
