From: Wu Fengguang <fengguang.wu@intel.com>
To: Andrew Morton <akpm@linux-foundation.org>
Cc: Jan Kara <jack@suse.cz>, Dave Chinner <david@fromorbit.com>,
Wu Fengguang <fengguang.wu@intel.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <Trond.Myklebust@netapp.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: linux-mm <linux-mm@kvack.org>
Cc: <linux-fsdevel@vger.kernel.org>
Cc: LKML <linux-kernel@vger.kernel.org>
Subject: [PATCH 31/47] writeback: increase min pause time on concurrent dirtiers
Date: Mon, 13 Dec 2010 14:43:20 +0800 [thread overview]
Message-ID: <20101213064840.799225309@intel.com> (raw)
In-Reply-To: 20101213064249.648862451@intel.com
[-- Attachment #1: writeback-min-pause-time-for-concurrent-dirtiers.patch --]
[-- Type: text/plain, Size: 2104 bytes --]
Target for >60ms pause time when there are 100+ heavy dirtiers per bdi.
(will average around 100ms given 200ms max pause time)
It's OK for 1 dd task doing 100MB/s to be throttle paused 100 times per
second. However when there are 100 tasks writing to the same disk,
That sums up to 100*100 balance_dirty_pages() calls per second and may
lead to massive cacheline bouncing on accessing the global page states
in NUMA machines. Even in single socket boxes, we easily see >10% CPU
time reduction by increasing the pause time.
CC: Dave Chinner <david@fromorbit.com>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
---
mm/page-writeback.c | 23 ++++++++++++++++++++++-
1 file changed, 22 insertions(+), 1 deletion(-)
--- linux-next.orig/mm/page-writeback.c 2010-12-09 12:24:45.000000000 +0800
+++ linux-next/mm/page-writeback.c 2010-12-09 12:24:47.000000000 +0800
@@ -666,6 +666,27 @@ static unsigned long max_pause(unsigned
}
/*
+ * Scale up pause time for concurrent dirtiers in order to reduce CPU overheads.
+ * But ensure reasonably large [min_pause, max_pause] range size, so that
+ * nr_dirtied_pause (and hence future pause time) can stay reasonably stable.
+ */
+static unsigned long min_pause(struct backing_dev_info *bdi,
+ unsigned long max)
+{
+ unsigned long hi = ilog2(bdi->write_bandwidth);
+ unsigned long lo = ilog2(bdi->throttle_bandwidth);
+ unsigned long t;
+
+ if (lo >= hi)
+ return 1;
+
+ /* (N * 10ms) on 2^N concurrent tasks */
+ t = (hi - lo) * (10 * HZ) / 1024;
+
+ return clamp_val(t, 1, max / 2);
+}
+
+/*
* balance_dirty_pages() must be called by processes which are generating dirty
* data. It looks at the number of dirty pages in the machine and will force
* the caller to perform writeback if the system is over `vm_dirty_ratio'.
@@ -833,7 +854,7 @@ pause:
if (pause == 0 && nr_dirty < background_thresh)
current->nr_dirtied_pause = ratelimit_pages(bdi);
- else if (pause == 1)
+ else if (pause <= min_pause(bdi, pause_max))
current->nr_dirtied_pause += current->nr_dirtied_pause / 32 + 1;
else if (pause >= pause_max)
/*
WARNING: multiple messages have this Message-ID (diff)
From: Wu Fengguang <fengguang.wu@intel.com>
To: Andrew Morton <akpm@linux-foundation.org>
Cc: Jan Kara <jack@suse.cz>, Dave Chinner <david@fromorbit.com>,
Wu Fengguang <fengguang.wu@intel.com>,
Christoph Hellwig <hch@lst.de>,
Trond Myklebust <Trond.Myklebust@netapp.com>,
Theodore Ts'o <tytso@mit.edu>,
Chris Mason <chris.mason@oracle.com>,
Peter Zijlstra <a.p.zijlstra@chello.nl>,
Mel Gorman <mel@csn.ul.ie>, Rik van Riel <riel@redhat.com>,
KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>,
Greg Thelen <gthelen@google.com>,
Minchan Kim <minchan.kim@gmail.com>,
linux-mm <linux-mm@kvack.org>,
linux-fsdevel@vger.kernel.org,
LKML <linux-kernel@vger.kernel.org>
Subject: [PATCH 31/47] writeback: increase min pause time on concurrent dirtiers
Date: Mon, 13 Dec 2010 14:43:20 +0800 [thread overview]
Message-ID: <20101213064840.799225309@intel.com> (raw)
In-Reply-To: 20101213064249.648862451@intel.com
[-- Attachment #1: writeback-min-pause-time-for-concurrent-dirtiers.patch --]
[-- Type: text/plain, Size: 2104 bytes --]
Target for >60ms pause time when there are 100+ heavy dirtiers per bdi.
(will average around 100ms given 200ms max pause time)
It's OK for 1 dd task doing 100MB/s to be throttle paused 100 times per
second. However when there are 100 tasks writing to the same disk,
That sums up to 100*100 balance_dirty_pages() calls per second and may
lead to massive cacheline bouncing on accessing the global page states
in NUMA machines. Even in single socket boxes, we easily see >10% CPU
time reduction by increasing the pause time.
CC: Dave Chinner <david@fromorbit.com>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
---
mm/page-writeback.c | 23 ++++++++++++++++++++++-
1 file changed, 22 insertions(+), 1 deletion(-)
--- linux-next.orig/mm/page-writeback.c 2010-12-09 12:24:45.000000000 +0800
+++ linux-next/mm/page-writeback.c 2010-12-09 12:24:47.000000000 +0800
@@ -666,6 +666,27 @@ static unsigned long max_pause(unsigned
}
/*
+ * Scale up pause time for concurrent dirtiers in order to reduce CPU overheads.
+ * But ensure reasonably large [min_pause, max_pause] range size, so that
+ * nr_dirtied_pause (and hence future pause time) can stay reasonably stable.
+ */
+static unsigned long min_pause(struct backing_dev_info *bdi,
+ unsigned long max)
+{
+ unsigned long hi = ilog2(bdi->write_bandwidth);
+ unsigned long lo = ilog2(bdi->throttle_bandwidth);
+ unsigned long t;
+
+ if (lo >= hi)
+ return 1;
+
+ /* (N * 10ms) on 2^N concurrent tasks */
+ t = (hi - lo) * (10 * HZ) / 1024;
+
+ return clamp_val(t, 1, max / 2);
+}
+
+/*
* balance_dirty_pages() must be called by processes which are generating dirty
* data. It looks at the number of dirty pages in the machine and will force
* the caller to perform writeback if the system is over `vm_dirty_ratio'.
@@ -833,7 +854,7 @@ pause:
if (pause == 0 && nr_dirty < background_thresh)
current->nr_dirtied_pause = ratelimit_pages(bdi);
- else if (pause == 1)
+ else if (pause <= min_pause(bdi, pause_max))
current->nr_dirtied_pause += current->nr_dirtied_pause / 32 + 1;
else if (pause >= pause_max)
/*
next prev parent reply other threads:[~2010-12-13 6:49 UTC|newest]
Thread overview: 143+ messages / expand[flat|nested] mbox.gz Atom feed top
2010-12-13 6:42 [PATCH 00/47] IO-less dirty throttling v3 Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:42 ` [PATCH 01/47] writeback: enabling gate limit for light dirtied bdi Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:42 ` [PATCH 02/47] writeback: safety margin for bdi stat error Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:42 ` [PATCH 03/47] writeback: IO-less balance_dirty_pages() Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:42 ` [PATCH 04/47] writeback: consolidate variable names in balance_dirty_pages() Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:42 ` [PATCH 05/47] writeback: per-task rate limit on balance_dirty_pages() Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:42 ` [PATCH 06/47] writeback: prevent duplicate balance_dirty_pages_ratelimited() calls Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:42 ` [PATCH 07/47] writeback: account per-bdi accumulated written pages Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:42 ` [PATCH 08/47] writeback: bdi write bandwidth estimation Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:42 ` [PATCH 09/47] writeback: show bdi write bandwidth in debugfs Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:42 ` [PATCH 10/47] writeback: quit throttling when bdi dirty pages dropped low Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:42 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 11/47] writeback: reduce per-bdi dirty threshold ramp up time Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 12/47] writeback: make reasonable gap between the dirty/background thresholds Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 13/47] writeback: scale down max throttle bandwidth on concurrent dirtiers Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 14/47] writeback: add trace event for balance_dirty_pages() Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 15/47] writeback: make nr_to_write a per-file limit Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 16/47] writeback: make-nr_to_write-a-per-file-limit fix Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 17/47] writeback: do uninterruptible sleep in balance_dirty_pages() Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 18/47] writeback: move BDI_WRITTEN accounting into __bdi_writeout_inc() Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 19/47] writeback: fix increasement of nr_dirtied_pause Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 20/47] writeback: use do_div in bw calculation Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 21/47] writeback: prevent divide error on tiny HZ Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 22/47] writeback: prevent bandwidth calculation overflow Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 23/47] writeback: spinlock protected bdi bandwidth update Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 24/47] writeback: increase pause time on concurrent dirtiers Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 25/47] writeback: make it easier to break from a dirty exceeded bdi Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 26/47] writeback: start background writeback earlier Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 27/47] writeback: user space think time compensation Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 28/47] writeback: bdi base throttle bandwidth Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 29/47] writeback: smoothed bdi dirty pages Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 30/47] writeback: adapt max balance pause time to memory size Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang [this message]
2010-12-13 6:43 ` [PATCH 31/47] writeback: increase min pause time on concurrent dirtiers Wu Fengguang
2010-12-13 6:43 ` [PATCH 32/47] writeback: extend balance_dirty_pages() trace event Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 33/47] writeback: trace global dirty page states Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 34/47] writeback: trace writeback_single_inode() Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 35/47] writeback: scale IO chunk size up to device bandwidth Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 36/47] btrfs: dont call balance_dirty_pages_ratelimited() on already dirty pages Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 37/47] btrfs: lower the dirty balacing rate limit Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 38/47] btrfs: wait on too many nr_async_bios Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 39/47] nfs: livelock prevention is now done in VFS Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 40/47] nfs: writeback pages wait queue Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 41/47] nfs: in-commit pages accounting and " Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 42/47] nfs: heuristics to avoid commit Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 43/47] nfs: dont change wbc->nr_to_write in write_inode() Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 44/47] nfs: limit the range of commits Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 45/47] nfs: adapt congestion threshold to dirty threshold Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 46/47] nfs: trace nfs_commit_unstable_pages() Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` [PATCH 47/47] nfs: trace nfs_commit_release() Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 6:43 ` Wu Fengguang
2010-12-13 11:27 ` [PATCH 00/47] IO-less dirty throttling v3 Peter Zijlstra
2010-12-13 11:27 ` Peter Zijlstra
2010-12-13 11:49 ` Wu Fengguang
2010-12-13 11:49 ` Wu Fengguang
2010-12-13 12:38 ` Peter Zijlstra
2010-12-13 12:38 ` Peter Zijlstra
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20101213064840.799225309@intel.com \
--to=fengguang.wu@intel.com \
--cc=akpm@linux-foundation.org \
--cc=david@fromorbit.com \
--cc=jack@suse.cz \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
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.