Просмотр исходного текста страницы Ext4 vs XFS

== Copy kernel source (from SSD with warm cache) ==

HDD: WD Scorpio Black 2.5" 750GB 7200rpm

* xfs 1 thread: 12.348s
* xfs 4 threads: 65.883s
* ext4 1 thread: 7.662s
* ext4 4 threads: 33.876s

== FS-Mark 3.3, creating 1M files ==

* HDD: WD Scorpio Black 2.5" 750GB 7200rpm
* fs_mark is a write-only test and it does fsync(), so there should be no skew caused by page cache

<plot>
set xrange [1:64]
set logscale x
set xtics (1, 2, 4, 8, 16, 32, 64)
set yrange [0:80]
set xlabel 'threads'
set ylabel 'fil/s (more is better)'
set grid ytics
set style fill solid 1.0 noborder
set boxwidth 0.7 relative
plot 'xfs.dat' using 1:2 title 'XFS' with linespoints, 'ext4.dat' using 1:2 title 'ext4' with linespoints
DATASET xfs
1.0  26.6
2.0  31.9
4.0  42.4
8.0  52.0
16.0  54.4
32.0  53.9
64.0  54.6
ENDDATASET
DATASET ext4
1.0 23.7
2.0 34.6
4.0 50.8
8.0 69.5
16.0 75.2
32.0 76.8
64.0 79.4
ENDDATASET
</plot>

== sysbench random read/write 16K in 8M files ==

* HDD: WD Scorpio Black 2.5" 750GB 7200rpm
* sysbench was ran with O_DIRECT, so the page cache should also have no impact.
* It's not a filesystem benchmark at all! It tests disk performance because it holds ALL prepared files open during the test. It only shows us that XFS isn't slowing down the direct access to the underlying device (which is also good, of course)...
* Probably because of the above note, the results are totally same for 1x 1GB file, 128x 8MB files and 3072x 16KB files.

<plot>
set xrange [1:64]
set logscale x
set xtics (1, 2, 4, 8, 16, 32, 64)
set yrange [0:300]
set xlabel 'threads'
set ylabel 'requests/s (more is better)'
set xzeroaxis
set grid ytics
set style fill solid 1.0 noborder
set boxwidth 0.7 relative
plot 'xfs.dat' using 1:2 title 'XFS' with linespoints, 'ext4.dat' using 1:2 title 'ext4' with linespoints
DATASET xfs
1.0  148.13
2.0  160.01
4.0  193.00
8.0  205.62
16.0  208.25
32.0  203.88
64.0  204.35
ENDDATASET
DATASET ext4
1.0 150.32
2.0 163.29
4.0 197.65
8.0 205.95
16.0 206.99
32.0 215.62
64.0 213.57
ENDDATASET
</plot>

== sysbench random read/write 16K in 16K files ==

HDD: WD Scorpio Black 2.5" 750GB 7200rpm

<plot>
set xrange [1:64]
set logscale x
set xtics (1, 2, 4, 8, 16, 32, 64)
set yrange [0:300]
set xlabel 'threads'
set ylabel 'requests/s (more is better)'
set xzeroaxis
set grid ytics
set style fill solid 1.0 noborder
set boxwidth 0.7 relative
plot 'xfs.dat' using 1:2 title 'XFS' with linespoints, 'ext4.dat' using 1:2 title 'ext4' with linespoints
DATASET xfs
1.0  106.77
2.0  116.11
4.0  169.85
8.0  220.37
16.0  260.92
32.0  277.94
64.0  278.42
ENDDATASET
DATASET ext4
1.0  104.00
2.0  115.41
4.0  169.69
8.0  221.74
16.0  257.36
32.0  268.36
64.0  257.91
ENDDATASET
</plot>

== [http://sourceforge.net/projects/filebench/ filebench] fileserver, dirty_ratio=1% ==

* HDD: WD VelociRaptor WD6000HLHX, 10000rpm
* fileserver test is read whole file + append + write whole file test ran on 10000 files in X threads
* filebench fails to run fileserver test with O_DIRECT, so I tried to "disable" page cache using dirty_ratio=1% and ran tests like this:

<code-bash>
echo 1 > /proc/sys/vm/dirty_ratio
echo 0 > /proc/sys/vm/dirty_bytes
echo 0 > /proc/sys/kernel/randomize_va_space
for i in 1 2 4 8 16 32 50 64; do
    echo
    echo "== $i threads =="
    echo
    echo 1 > /proc/sys/vm/drop_caches
    sync
    filebench <<EOF
load fileserver
set \$dir=/media/sdd
set \$nthreads=$i
run 30
EOF
done
echo 20 > /proc/sys/vm/dirty_ratio
</code-bash>

<plot>
set xrange [1:64]
set logscale x
set xtics (1, 2, 4, 8, 16, 32, 50, 64)
set yrange [0:3200]
set xlabel 'threads'
set ylabel 'ops/s (more is better)'
set xzeroaxis
set grid ytics
set style fill solid 1.0 noborder
set boxwidth 0.7 relative
plot 'xfs.dat' using 1:2 title 'XFS' with linespoints, 'ext4.dat' using 1:2 title 'ext4' with linespoints
DATASET xfs
1.0  1893
2.0  1801
4.0  1778
8.0  1864
16.0  1720
32.0  1473
50.0  1210
64.0  1056
ENDDATASET
DATASET ext4
1.0  3075
2.0  2352
4.0  2334
8.0  2346
16.0  2022
32.0  1559
50.0  1317
64.0  1163
ENDDATASET
</plot>

== [http://sourceforge.net/projects/filebench/ filebench] fileserver, dirty_ratio=20% ==

* HDD: WD VelociRaptor WD6000HLHX, 10000rpm
* Same test but ran with default 20% dirty_ratio setting. It's clearly seen that the system was using page cache extensively (ext4 was permanently gaining an unreal result in the single-thread test...)

<plot>
set xrange [1:64]
set logscale x
set xtics (1, 2, 4, 8, 16, 32, 50, 64)
set yrange [0:20000]
set xlabel 'threads'
set ylabel 'ops/s (more is better)'
set xzeroaxis
set grid ytics
set style fill solid 1.0 noborder
set boxwidth 0.7 relative
plot 'xfs.dat' using 1:2 title 'XFS' with linespoints, 'ext4.dat' using 1:2 title 'ext4' with linespoints
DATASET xfs
1.0  7755
2.0  3796
4.0  3333
8.0  3320
16.0  3559
32.0  3653
50.0  2650
64.0  1671
ENDDATASET
DATASET ext4
1.0  16415
2.0  4136
4.0  4108
8.0  4026
16.0  3570
32.0  3147
50.0  2632
64.0  2778
ENDDATASET
</plot>