This patchset is against one of project ideas, RBtree lock contention:
"Btrfs uses a number of rbtrees to index in-memory data structures.
Some of these are dominated by reads, and the lock contention from searching
them is showing up in profiles. We need to look into an RCU and sequence
counter combination to allow lockless reads."
The goal is to use RCU, but we take it as a long term one, and instead we use
rwlock until we find a mature rcu structure for lockless read.
So what we need to do is to make the code RCU friendly, and the idea mainly
comes from Chris Mason:
Quoted:
"I think the extent_state code can be much more RCU friendly if we separate
the operations on the tree from operations on the individual state.
In general, we can gain a lot of performance if we are able to reduce
the write locks taken at endio time. Especially for reads, these are
critical."
If we read a 10M file with pagecache,
$ dd if=/mnt/btrfs/foobar of=/dev/null bs=1M
without this patchset,
lock for read : lock for write ~= 2:1
with this patchset,
lock for read : lock for write ~= 200:1
I've run through xfstests, and no bugs jump out by then.
MORE TESTS ARE WELCOME!
Liu Bo (5):
Btrfs: use radix tree for checksum
Btrfs: merge adjacent states as much as possible
Btrfs: use large extent range for read and its endio
Btrfs: apply rwlock for extent state
Btrfs: avoid splitting state when truncating pagecache
fs/btrfs/extent_io.c | 768 +++++++++++++++++++++++++++++++++----------
fs/btrfs/extent_io.h | 5 +-
fs/btrfs/free-space-cache.c | 5 +
fs/btrfs/inode.c | 22 +-
4 files changed, 614 insertions(+), 186 deletions(-)
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
