I was investigating a performance bug that looked like dd to tmpfs
had regressed. The bulk of the problem turned out to be a difference
in Kconfig but it got me looking at the unnecessary overhead in tmpfs,
mark_page_accessed and parts of the allocator. This series is the result.
The patches themselves have details of the performance results but here
are some of the results based on ext4. This is the result of dd'ing to
a file multiple times on tmpfs
loopdd Throughput
3.15.0-rc3 3.15.0-rc3
vanilla lockpage-v2r33
Min 4096.0000 ( 0.00%) 3891.2000 ( -5.00%)
Mean 4840.1067 ( 0.00%) 5154.1333 ( 6.49%)
TrimMean 4867.6571 ( 0.00%) 5204.1143 ( 6.91%)
Stddev 160.6807 ( 0.00%) 275.1917 ( 71.27%)
Max 5017.6000 ( 0.00%) 5324.8000 ( 6.12%)
loopdd elapsed time
3.15.0-rc3 3.15.0-rc3
vanilla lockpage-v2r33
Min elapsed 0.4100 ( 0.00%) 0.3900 ( 4.88%)
Mean elapsed 0.4780 ( 0.00%) 0.4203 ( 12.06%)
TrimMean elapsed 0.4796 ( 0.00%) 0.4179 ( 12.88%)
Stddev elapsed 0.0353 ( 0.00%) 0.0379 ( -7.23%)
Max elapsed 0.5100 ( 0.00%) 0.4800 ( 5.88%)
This table shows the latency in usecs of accessing ext4-backed
mappings of various sizes
lat_mmap
3.15.0-rc3 3.15.0-rc3
vanilla lockpage-v2r33
Procs 107M 557.0000 ( 0.00%) 544.0000 ( 2.33%)
Procs 214M 1150.0000 ( 0.00%) 1058.0000 ( 8.00%)
Procs 322M 1897.0000 ( 0.00%) 1554.0000 ( 18.08%)
Procs 429M 2188.0000 ( 0.00%) 2652.0000 (-21.21%)
Procs 536M 2622.0000 ( 0.00%) 2473.0000 ( 5.68%)
Procs 644M 3065.0000 ( 0.00%) 2486.0000 ( 18.89%)
Procs 751M 3400.0000 ( 0.00%) 3012.0000 ( 11.41%)
Procs 859M 3996.0000 ( 0.00%) 3926.0000 ( 1.75%)
Procs 966M 4646.0000 ( 0.00%) 3763.0000 ( 19.01%)
Procs 1073M 4981.0000 ( 0.00%) 4154.0000 ( 16.60%)
Procs 1181M 5419.0000 ( 0.00%) 5152.0000 ( 4.93%)
Procs 1288M 5553.0000 ( 0.00%) 5538.0000 ( 0.27%)
Procs 1395M 5841.0000 ( 0.00%) 5730.0000 ( 1.90%)
Procs 1503M 6225.0000 ( 0.00%) 5981.0000 ( 3.92%)
Procs 1610M 6558.0000 ( 0.00%) 6332.0000 ( 3.45%)
Procs 1717M 7130.0000 ( 0.00%) 6741.0000 ( 5.46%)
Procs 1825M 9394.0000 ( 0.00%) 8483.0000 ( 9.70%)
Procs 1932M 8056.0000 ( 0.00%) 9427.0000 (-17.02%)
Procs 2040M 8463.0000 ( 0.00%) 9030.0000 ( -6.70%)
Procs 2147M 9014.0000 ( 0.00%) 8608.0000 ( 4.50%)
In general the system CPU overhead is lower.
arch/tile/mm/homecache.c | 2 +-
fs/btrfs/extent_io.c | 11 +-
fs/btrfs/file.c | 5 +-
fs/buffer.c | 7 +-
fs/ext4/mballoc.c | 14 +-
fs/f2fs/checkpoint.c | 3 -
fs/f2fs/node.c | 2 -
fs/fuse/dev.c | 2 +-
fs/fuse/file.c | 2 -
fs/gfs2/aops.c | 1 -
fs/gfs2/meta_io.c | 4 +-
fs/ntfs/attrib.c | 1 -
fs/ntfs/file.c | 1 -
include/linux/cpuset.h | 31 +++++
include/linux/gfp.h | 4 +-
include/linux/mmzone.h | 22 ++-
include/linux/page-flags.h | 18 +++
include/linux/pageblock-flags.h | 34 ++++-
include/linux/pagemap.h | 115 ++++++++++++++--
include/linux/swap.h | 9 +-
kernel/cpuset.c | 8 +-
kernel/sched/wait.c | 3 +-
mm/filemap.c | 292 ++++++++++++++++++++++------------------
mm/page_alloc.c | 226 ++++++++++++++++++-------------
mm/shmem.c | 8 +-
mm/swap.c | 17 ++-
mm/swap_state.c | 2 +-
mm/vmscan.c | 6 +-
28 files changed, 556 insertions(+), 294 deletions(-)
--
1.8.4.5
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
