c/mirror_zfs
1
0
Fork 0
mirror of https://git.proxmox.com/git/mirror_zfs.git synced 2024-11-18 10:21:01 +03:00
Code Issues Packages Projects Releases Wiki Activity

Move write aggregation memory copy out of vq_lock

Browse Source 
Memory copy is too heavy operation to do under the congested lock.
Moving it out reduces congestion by many times to almost invisible.
Since the original zio removed from the queue, and the child zio is
not executed yet, I don't see why would the copy need protection.
My guess it just remained like this from the time when lock was not
dropped here, which was added later to fix lock ordering issue.

Multi-threaded sequential write tests with both HDD and SSD pools
with ZVOL block sizes of 4KB, 16KB, 64KB and 128KB all show major
reduction of lock congestion, saving from 15% to 35% of CPU time
and increasing throughput from 10% to 40%.

Reviewed-by: Richard Yao <ryao@gentoo.org>
Reviewed-by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by:  Alexander Motin <mav@FreeBSD.org>
Closes #8890
This commit is contained in:
Alexander Motin 2019-06-13 16:08:24 -04:00 committed by Tony Hutter
parent 19cebf0518
commit cab7d856ea
1 changed files with 12 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
module/zfs/vdev_queue.c
View File

@ -709,6 +709,18 @@ vdev_queue_aggregate(vdev_queue_t *vq, zio_t *zio)
do {
dio = nio;
nio = AVL_NEXT(t, dio);
zio_add_child(dio, aio);
vdev_queue_io_remove(vq, dio);
} while (dio != last);
/*
* We need to drop the vdev queue's lock during zio_execute() to
* avoid a deadlock that we could encounter due to lock order
* reversal between vq_lock and io_lock in zio_change_priority().
* Use the dropped lock to do memory copy without congestion.
*/
mutex_exit(&vq->vq_lock);
while ((dio = zio_walk_parents(aio, &zl)) != NULL) {
ASSERT3U(dio->io_type, ==, aio->io_type);
if (dio->io_flags & ZIO_FLAG_NODATA) {
@ -720,16 +732,6 @@ vdev_queue_aggregate(vdev_queue_t *vq, zio_t *zio)
dio->io_offset - aio->io_offset, 0, dio->io_size);
}
zio_add_child(dio, aio);
vdev_queue_io_remove(vq, dio);
} while (dio != last);
/*
* We need to drop the vdev queue's lock to avoid a deadlock that we
* could encounter since this I/O will complete immediately.
*/
mutex_exit(&vq->vq_lock);
while ((dio = zio_walk_parents(aio, &zl)) != NULL) {
zio_vdev_io_bypass(dio);
zio_execute(dio);
}