[PATCH]block: make auto block plug flush threshold per-disk based

[PATCH]block: make auto block plug flush threshold per-disk based

[Shaohua Li]

Subject: block: make auto block plug flush threshold per-disk based

We do auto block plug flush to reduce latency, the threshold is 16
requests. This works well if the task is accessing one or two drives.
The problem is if the task is accessing a raid 0 device and the raid
disk number is big, say 8 or 16, 16/8 = 2 or 16/16=1, we will have
heavy lock contention.

This patch makes the threshold per-disk based. The latency should be
still ok accessing one or two drives. The setup with application
accessing a lot of drives in the meantime uaually is big machine,
avoiding lock contention is more important, because any contention
will actually increase latency.

Signed-off-by: Shaohua Li <shli@fusionio.com>
---
  block/blk-core.c |    3 ++-
  1 file changed, 2 insertions(+), 1 deletion(-)

Index: linux/block/blk-core.c
===================================================================
--- linux.orig/block/blk-core.c	2012-04-06 10:51:29.766932123 -0600
+++ linux/block/blk-core.c	2012-04-06 10:56:46.386929357 -0600
@@ -1277,7 +1277,8 @@
  	list_for_each_entry_reverse(rq, &plug->list, queuelist) {
  		int el_ret;

-		(*request_count)++;
+		if (rq->q == q)
+			(*request_count)++;

  		if (rq->q != q || !blk_rq_merge_ok(rq, bio))
  			continue;

Re: [PATCH]block: make auto block plug flush threshold per-disk based

[Jens Axboe]

On 2012-04-06 11:27, Shaohua Li wrote:
> We do auto block plug flush to reduce latency, the threshold is 16
> requests. This works well if the task is accessing one or two drives.
> The problem is if the task is accessing a raid 0 device and the raid
> disk number is big, say 8 or 16, 16/8 = 2 or 16/16=1, we will have
> heavy lock contention.
> 
> This patch makes the threshold per-disk based. The latency should be
> still ok accessing one or two drives. The setup with application
> accessing a lot of drives in the meantime uaually is big machine,
> avoiding lock contention is more important, because any contention
> will actually increase latency.

Thanks, applied. It should have been per-queue all along, so this
makes a lot of sense.

-- 
Jens Axboe

Re: [PATCH]block: make auto block plug flush threshold per-disk based

[Jeff Moyer]

Jens Axboe <axboe@kernel.dk> writes:

> On 2012-04-06 11:27, Shaohua Li wrote:
>> We do auto block plug flush to reduce latency, the threshold is 16
>> requests. This works well if the task is accessing one or two drives.
>> The problem is if the task is accessing a raid 0 device and the raid
>> disk number is big, say 8 or 16, 16/8 = 2 or 16/16=1, we will have
>> heavy lock contention.
>> 
>> This patch makes the threshold per-disk based. The latency should be
>> still ok accessing one or two drives. The setup with application
>> accessing a lot of drives in the meantime uaually is big machine,
>> avoiding lock contention is more important, because any contention
>> will actually increase latency.
>
> Thanks, applied. It should have been per-queue all along, so this
> makes a lot of sense.

Yeah, I had noticed this before when looking over the code.  Makes
sense.

Acked-by: Jeff Moyer <jmoyer@redhat.com>