[PATCH -mmotm 0/5] memcg: per cgroup dirty limit (v6)

[Andrea Righi]

Control the maximum amount of dirty pages a cgroup can have at any given time.

Per cgroup dirty limit is like fixing the max amount of dirty (hard to reclaim)
page cache used by any cgroup. So, in case of multiple cgroup writers, they
will not be able to consume more than their designated share of dirty pages and
will be forced to perform write-out if they cross that limit.

The overall design is the following:

 - account dirty pages per cgroup
 - limit the number of dirty pages via memory.dirty_ratio / memory.dirty_bytes
   and memory.dirty_background_ratio / memory.dirty_background_bytes in
   cgroupfs
 - start to write-out (background or actively) when the cgroup limits are
   exceeded

This feature is supposed to be strictly connected to any underlying IO
controller implementation, so we can stop increasing dirty pages in VM layer
and enforce a write-out before any cgroup will consume the global amount of
dirty pages defined by the /proc/sys/vm/dirty_ratio|dirty_bytes and
/proc/sys/vm/dirty_background_ratio|dirty_background_bytes limits.

Changelog (v5 -> v6)
~~~~~~~~~~~~~~~~~~~~~~
 * always disable/enable IRQs at lock/unlock_page_cgroup(): this allows to drop
   the previous complicated locking scheme in favor of a simpler locking, even
   if this obviously adds some overhead (see results below)
 * drop FUSE and NILFS2 dirty pages accounting for now (this depends on
   charging bounce pages per cgroup)

Results
~~~~~~~
I ran some tests using a kernel build (2.6.33 x86_64_defconfig) on a
Intel Core 2 @ 1.2GHz as testcase using different kernels:
 - mmotm "vanilla"
 - mmotm with cgroup-dirty-memory using the previous "complex" locking scheme
   (my previous patchset + the fixes reported by Kame-san and Daisuke-san)
 - mmotm with cgroup-dirty-memory using the simple locking scheme
   (lock_page_cgroup() with IRQs disabled)

Following the results:
<before>
 - mmotm "vanilla", root  cgroup:			11m51.983s
 - mmotm "vanilla", child cgroup:			11m56.596s

<after>
 - mmotm, "complex" locking scheme, root  cgroup:	11m53.037s
 - mmotm, "complex" locking scheme, child cgroup:	11m57.896s

 - mmotm, lock_page_cgroup+irq_disabled, root  cgroup:	12m5.499s
 - mmotm, lock_page_cgroup+irq_disabled, child cgroup:	12m9.920s

With the "complex" locking solution, the overhead introduced by the
cgroup dirty memory accounting is minimal (0.14%), compared with the overhead
introduced by the lock_page_cgroup+irq_disabled solution (1.90%).

The performance overhead is not so huge in both solutions, but the impact on
performance is even more reduced using a complicated solution...

Maybe we can go ahead with the simplest implementation for now and start to
think to an alternative implementation of the page_cgroup locking and
charge/uncharge of pages.

If someone is interested or want to repeat the tests (maybe on a bigger
machine) I can post also the other version of the patchset. Just let me know.

-Andrea

 Documentation/cgroups/memory.txt |   36 +++
 fs/nfs/write.c                   |    4 +
 include/linux/memcontrol.h       |   87 +++++++-
 include/linux/page_cgroup.h      |   42 ++++-
 include/linux/writeback.h        |    2 -
 mm/filemap.c                     |    1 +
 mm/memcontrol.c                  |  475 +++++++++++++++++++++++++++++++++-----
 mm/page-writeback.c              |  215 +++++++++++-------
 mm/rmap.c                        |    4 +-
 mm/truncate.c                    |    1 +
 10 files changed, 722 insertions(+), 145 deletions(-)
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/