lkml.org 
[lkml]   [2014]   [Jun]   [18]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Subject[PATCH 0/4] Improve sequential read throughput
Date
IO performance since 3.0 has been a mixed bag. In many respects we are
better and in some we are worse and one of those places is sequential read
performance, particularly for higher numbers of threads. This is visible
in a number of benchmarks but tiobench has been the one I looked at the
closest despite its age.

3.16.0-rc1 3.16.0-rc1 3.0.0
vanilla patch-series vanilla
Mean SeqRead-MB/sec-1 121.88 ( 0.00%) 133.84 ( 9.81%) 134.59 ( 10.42%)
Mean SeqRead-MB/sec-2 101.99 ( 0.00%) 115.01 ( 12.77%) 122.59 ( 20.20%)
Mean SeqRead-MB/sec-4 97.42 ( 0.00%) 108.40 ( 11.27%) 114.78 ( 17.82%)
Mean SeqRead-MB/sec-8 83.39 ( 0.00%) 97.50 ( 16.92%) 100.14 ( 20.09%)
Mean SeqRead-MB/sec-16 68.90 ( 0.00%) 82.14 ( 19.22%) 81.64 ( 18.50%)

The impact on the other operations is negligible. Note that 3.0-vanilla is
still far better but bringing the patch series further in line would involve
increasing the CFQ target latency higher and there should be better options.
This series is a major improvement on 3.16-rc1-vanilla at least so worth
sending out to a larger audience for comment.

block/cfq-iosched.c | 2 +-
include/linux/mmzone.h | 9 +++
include/linux/writeback.h | 1 +
include/trace/events/pagemap.h | 16 ++--
mm/internal.h | 1 +
mm/mm_init.c | 5 +-
mm/page-writeback.c | 15 ++--
mm/page_alloc.c | 176 ++++++++++++++++++++++++++---------------
mm/swap.c | 4 +-
9 files changed, 144 insertions(+), 85 deletions(-)

--
1.8.4.5



\
 
 \ /
  Last update: 2014-06-18 11:01    [W:0.110 / U:0.572 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site