Commit da023618 authored by Christoph Hellwig's avatar Christoph Hellwig Committed by David Sterba

btrfs: submit IO synchronously for fast checksum implementations

Most modern hardware supports very fast accelerated crc32c calculation.
If that is supported the CPU overhead of the checksum calculation is
very limited, and offloading the calculation to special worker threads
has a lot of overhead for no gain.

E.g. on an Intel Optane device is actually very much slows down even
1M buffered writes with fio:

Unpatched:

write: IOPS=3316, BW=3316MiB/s (3477MB/s)(200GiB/61757msec); 0 zone resets

With synchronous CRCs:

write: IOPS=4882, BW=4882MiB/s (5119MB/s)(200GiB/41948msec); 0 zone resets

With a lot of variation during the unpatched run going down as low as
1100MB/s, while the synchronous CRC version has about the same peak write
speed but much lower dips, and fewer kworkers churning around.
Both tests had fio saturated at 100% CPU.

(thanks to Jens Axboe via Chris Mason for the benchmarking)
Reviewed-by: default avatarChris Mason <clm@fb.com>
Reviewed-by: default avatarJohannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: default avatarChristoph Hellwig <hch@lst.de>
Reviewed-by: default avatarDavid Sterba <dsterba@suse.com>
Signed-off-by: default avatarDavid Sterba <dsterba@suse.com>
parent adbe7e38
......@@ -574,6 +574,10 @@ static void run_one_async_free(struct btrfs_work *work)
static bool should_async_write(struct btrfs_bio *bbio)
{
/* Submit synchronously if the checksum implementation is fast. */
if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &bbio->fs_info->flags))
return false;
/*
* If the I/O is not issued by fsync and friends, (->sync_writers != 0),
* then try to defer the submission to a workqueue to parallelize the
......@@ -582,19 +586,9 @@ static bool should_async_write(struct btrfs_bio *bbio)
if (atomic_read(&bbio->inode->sync_writers))
return false;
/*
* Submit metadata writes synchronously if the checksum implementation
* is fast, or we are on a zoned device that wants I/O to be submitted
* in order.
*/
if (bbio->bio.bi_opf & REQ_META) {
struct btrfs_fs_info *fs_info = bbio->fs_info;
if (btrfs_is_zoned(fs_info))
return false;
if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
return false;
}
/* Zoned devices require I/O to be submitted in order. */
if ((bbio->bio.bi_opf & REQ_META) && btrfs_is_zoned(bbio->fs_info))
return false;
return true;
}
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment