All of lore.kernel.org
 help / color / mirror / Atom feed
* [patch 2/2]block: recursive merge requests
@ 2011-12-16  1:52 Shaohua Li
  2011-12-16 12:59 ` Jens Axboe
  2011-12-16 19:34 ` Christoph Hellwig
  0 siblings, 2 replies; 4+ messages in thread
From: Shaohua Li @ 2011-12-16  1:52 UTC (permalink / raw)
  To: Jens Axboe; +Cc: lkml

In my workload, thread 1 accesses a, a+2, ..., thread 2 accesses a+1,
a+3,.... When the requests are flushed to queue, a and a+1 are merged
to (a, a+1), a+2 and a+3 too to (a+2, a+3), but (a, a+1) and (a+2, a+3)
aren't merged.
With recursive merge below, the workload throughput gets improved 20%
and context switch drops 60%.

Signed-off-by: Shaohua Li <shaohua.li@intel.com>
---
 block/elevator.c |   16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)

Index: linux/block/elevator.c
===================================================================
--- linux.orig/block/elevator.c	2011-12-16 09:08:50.000000000 +0800
+++ linux/block/elevator.c	2011-12-16 09:08:58.000000000 +0800
@@ -521,6 +521,7 @@ static bool elv_attempt_insert_merge(str
 				     struct request *rq)
 {
 	struct request *__rq;
+	bool ret;
 
 	if (blk_queue_nomerges(q))
 		return false;
@@ -534,14 +535,21 @@ static bool elv_attempt_insert_merge(str
 	if (blk_queue_noxmerges(q))
 		return false;
 
+	ret = false;
 	/*
 	 * See if our hash lookup can find a potential backmerge.
 	 */
-	__rq = elv_rqhash_find(q, blk_rq_pos(rq));
-	if (__rq && blk_attempt_req_merge(q, __rq, rq))
-		return true;
+	while (1) {
+		__rq = elv_rqhash_find(q, blk_rq_pos(rq));
+		if (!__rq || !blk_attempt_req_merge(q, __rq, rq))
+			break;
+
+		/* The merged request could be merged with others, try again */
+		ret = true;
+		rq = __rq;
+	}
 
-	return false;
+	return ret;
 }
 
 void elv_merged_request(struct request_queue *q, struct request *rq, int type)



^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [patch 2/2]block: recursive merge requests
  2011-12-16  1:52 [patch 2/2]block: recursive merge requests Shaohua Li
@ 2011-12-16 12:59 ` Jens Axboe
  2011-12-16 19:34 ` Christoph Hellwig
  1 sibling, 0 replies; 4+ messages in thread
From: Jens Axboe @ 2011-12-16 12:59 UTC (permalink / raw)
  To: Shaohua Li; +Cc: lkml

On 2011-12-16 02:52, Shaohua Li wrote:
> In my workload, thread 1 accesses a, a+2, ..., thread 2 accesses a+1,
> a+3,.... When the requests are flushed to queue, a and a+1 are merged
> to (a, a+1), a+2 and a+3 too to (a+2, a+3), but (a, a+1) and (a+2, a+3)
> aren't merged.
> With recursive merge below, the workload throughput gets improved 20%
> and context switch drops 60%.

Interesting, I didn't consider that corner case. Hard to argue with the
numbers, this might positively impact apps using posix aio for instance.
Applied to for-3.3/core.

-- 
Jens Axboe


^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [patch 2/2]block: recursive merge requests
  2011-12-16  1:52 [patch 2/2]block: recursive merge requests Shaohua Li
  2011-12-16 12:59 ` Jens Axboe
@ 2011-12-16 19:34 ` Christoph Hellwig
  2011-12-16 20:16   ` Jens Axboe
  1 sibling, 1 reply; 4+ messages in thread
From: Christoph Hellwig @ 2011-12-16 19:34 UTC (permalink / raw)
  To: Shaohua Li; +Cc: Jens Axboe, lkml

On Fri, Dec 16, 2011 at 09:52:43AM +0800, Shaohua Li wrote:
> With recursive merge below, the workload throughput gets improved 20%
> and context switch drops 60%.

What workload and what filesystem?


^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [patch 2/2]block: recursive merge requests
  2011-12-16 19:34 ` Christoph Hellwig
@ 2011-12-16 20:16   ` Jens Axboe
  0 siblings, 0 replies; 4+ messages in thread
From: Jens Axboe @ 2011-12-16 20:16 UTC (permalink / raw)
  To: Christoph Hellwig; +Cc: Shaohua Li, lkml

On 2011-12-16 20:34, Christoph Hellwig wrote:
> On Fri, Dec 16, 2011 at 09:52:43AM +0800, Shaohua Li wrote:
>> With recursive merge below, the workload throughput gets improved 20%
>> and context switch drops 60%.
> 
> What workload and what filesystem?

It shouldn't really matter for the quoted case, where you have 2
processes issuing interleaved reads.

-- 
Jens Axboe


^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2011-12-16 20:16 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2011-12-16  1:52 [patch 2/2]block: recursive merge requests Shaohua Li
2011-12-16 12:59 ` Jens Axboe
2011-12-16 19:34 ` Christoph Hellwig
2011-12-16 20:16   ` Jens Axboe

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.