| Message ID | 20190318191031.11066-1-idryomov@gmail.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | rbd: set io_min, io_opt and discard_granularity to alloc_size | expand |
On Mon, Mar 18, 2019 at 3:10 PM Ilya Dryomov <idryomov@gmail.com> wrote: > > Now that we have alloc_size that controls our discard behavior, it > doesn't make sense to have these set to object (set) size. alloc_size > defaults to 64k, but because discard_granularity is likely 4M, only > ranges that are equal to or bigger than 4M can be considered during > fstrim. A smaller io_min is also more likely to be met, resulting in > fewer deferred writes on bluestore OSDs. > > Signed-off-by: Ilya Dryomov <idryomov@gmail.com> > --- > drivers/block/rbd.c | 8 ++++---- > 1 file changed, 4 insertions(+), 4 deletions(-) > > diff --git a/drivers/block/rbd.c b/drivers/block/rbd.c > index a32d5b54d59e..e037f1ab2fde 100644 > --- a/drivers/block/rbd.c > +++ b/drivers/block/rbd.c > @@ -834,7 +834,7 @@ static int parse_rbd_opts_token(char *c, void *private) > pctx->opts->queue_depth = intval; > break; > case Opt_alloc_size: > - if (intval < 1) { > + if (intval < SECTOR_SIZE) { > pr_err("alloc_size out of range\n"); > return -EINVAL; > } > @@ -4204,12 +4204,12 @@ static int rbd_init_disk(struct rbd_device *rbd_dev) > q->limits.max_sectors = queue_max_hw_sectors(q); > blk_queue_max_segments(q, USHRT_MAX); > blk_queue_max_segment_size(q, UINT_MAX); > - blk_queue_io_min(q, objset_bytes); > - blk_queue_io_opt(q, objset_bytes); > + blk_queue_io_min(q, rbd_dev->opts->alloc_size); > + blk_queue_io_opt(q, rbd_dev->opts->alloc_size); > > if (rbd_dev->opts->trim) { > blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); > - q->limits.discard_granularity = objset_bytes; > + q->limits.discard_granularity = rbd_dev->opts->alloc_size; > blk_queue_max_discard_sectors(q, objset_bytes >> SECTOR_SHIFT); > blk_queue_max_write_zeroes_sectors(q, objset_bytes >> SECTOR_SHIFT); > } > -- > 2.19.2 > Reviewed-by: Jason Dillaman <dillaman@redhat.com>
diff --git a/drivers/block/rbd.c b/drivers/block/rbd.c index a32d5b54d59e..e037f1ab2fde 100644 --- a/drivers/block/rbd.c +++ b/drivers/block/rbd.c @@ -834,7 +834,7 @@ static int parse_rbd_opts_token(char *c, void *private) pctx->opts->queue_depth = intval; break; case Opt_alloc_size: - if (intval < 1) { + if (intval < SECTOR_SIZE) { pr_err("alloc_size out of range\n"); return -EINVAL; } @@ -4204,12 +4204,12 @@ static int rbd_init_disk(struct rbd_device *rbd_dev) q->limits.max_sectors = queue_max_hw_sectors(q); blk_queue_max_segments(q, USHRT_MAX); blk_queue_max_segment_size(q, UINT_MAX); - blk_queue_io_min(q, objset_bytes); - blk_queue_io_opt(q, objset_bytes); + blk_queue_io_min(q, rbd_dev->opts->alloc_size); + blk_queue_io_opt(q, rbd_dev->opts->alloc_size); if (rbd_dev->opts->trim) { blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); - q->limits.discard_granularity = objset_bytes; + q->limits.discard_granularity = rbd_dev->opts->alloc_size; blk_queue_max_discard_sectors(q, objset_bytes >> SECTOR_SHIFT); blk_queue_max_write_zeroes_sectors(q, objset_bytes >> SECTOR_SHIFT); }
Now that we have alloc_size that controls our discard behavior, it doesn't make sense to have these set to object (set) size. alloc_size defaults to 64k, but because discard_granularity is likely 4M, only ranges that are equal to or bigger than 4M can be considered during fstrim. A smaller io_min is also more likely to be met, resulting in fewer deferred writes on bluestore OSDs. Signed-off-by: Ilya Dryomov <idryomov@gmail.com> --- drivers/block/rbd.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)