Summary: jfs had unkillable dbench processes. I've seen dbench have
some processes not exit (various filesystems) in recent 2.5.x.
This was the first time they were unkillable.
Every filesystem was best in at least one test. Filesystems
exhibit very different performance characteristics for some
workloads. For these benchmarks, xfs with non-default mkfs/mount
options is impressive.
Same LUN (hardware RAID 0) and partition used for each filesystem.
QLogic 2200 Fibre channel.
Quad P3 700 Xeon with 3.75 GB RAM.
System not rebooted between filesystem runs.
mkfs commands
mke2fs -q /dev/sdc1
mke2fs -q -j -J size=400 /dev/sdc1
yes "y" | mkreiserfs /dev/sdc1 >/tmp/mkr.out 2>&1
mkfs.xfs -l size=32768b -f /dev/sdc1
mount commands
mount -t ext2 -o defaults,noatime /dev/sdc1 /fs1
mount -t ext3 -o defaults,noatime,data=writeback /dev/sdc1 /fs1
mount -t reiserfs -o defaults,noatime,notail /dev/sdc1 /fs1
mount -t xfs -o logbufs=8,logbsize=32768,noatime /dev/sdc1 /fs1
dbench executed 5 times. Numbers are in MB/second.
dbench 64 processe Average High Low
2.5.70-ext2 254.37 255.86 252.54 MB/sec
2.5.70-xfs 143.46 144.41 142.73
2.5.70-ext3 123.77 127.68 115.50
2.5.70-reiserfs 104.46 118.11 93.33
dbench 192 processes Average High Low
2.5.70-ext2 218.44 241.18 196.96 MB/sec
2.5.70-xfs 114.43 115.53 113.20
2.5.70-ext3 101.56 115.24 91.38
2.5.70-reiserfs 70.44 74.01 66.81
bonnie++-1.02c with 8192 MB filesize
--------------------- Sequential Output ------------------
---- Per Char ----- ------ Block ----- ---- Rewrite ----
Kernel MB/sec %CPU Eff MB/sec %CPU Eff MB/sec %CPU Eff
2.5.70-ext2 9.42 99.0 9.51 69.42 53.0 131 16.02 17.0 94
2.5.70-xfs 9.34 99.0 9.44 62.81 51.3 122 16.02 18.3 87
2.5.70-ext3 8.64 95.0 9.10 59.80 70.0 85 16.15 18.0 90
2.5.70-reiserfs 9.06 95.3 9.51 49.64 65.0 76 16.59 21.0 79
-------- Sequential Input ---------- ----- Random -----
---- Per Char --- ----- Block ----- ----- Seeks -----
Kernel MB/sec %CPU Eff MB/sec %CPU Eff /sec %CPU Eff
2.5.70-ext2 9.28 98.7 9.4 27.06 19.7 138 488.1 3.00 16269
2.5.70-xfs 9.26 98.7 9.4 27.02 20.0 135 643.0 4.00 16076
2.5.70-ext3 9.33 99.0 9.4 27.01 19.3 140 457.7 3.00 15256
2.5.70-reiserfs 9.18 98.0 9.4 26.87 21.7 124 491.1 3.00 16371
bonnie++-1.02c with 65536 small files
reiserfs and xfs create small files faster than ext2 and ext3
but are slower at deleting them.
--------------- Sequential ----------
----- Create ----- ---- Delete ----
/sec %CPU Eff /sec %CPU Eff
2.5.70-reiserfs 7584 86.7 8751 2628 37.3 7038
2.5.70-xfs 1710 39.3 4347 2053 28.3 7247
2.5.70-ext2 150 99.0 151 60883 100.0 6088
2.5.70-ext3 119 95.0 126 26319 87.7 3002
Random small file create performance is similar to sequential.
Random small file deletes are a lot slower on ext2/ext3
compared to sequential deletes.
------------- Random --------------
----- Create ---- ---- Delete ----
/sec %CPU Eff /sec %CPU Eff
2.5.70-reiserfs 6666 79.3 8402 956 17.0 5622
2.5.70-xfs 1701 35.7 4770 1277 20.0 6385
2.5.70-ext2 152 99.0 154 534 99.0 540
2.5.70-ext3 119 95.0 125 434 91.3 475
tiobench-0.3.3
Unit information
================
File size = 8192 megabytes
Blk Size = 4096 bytes
Rate = megabytes per second
CPU% = percentage of CPU used during the test
Latency = milliseconds
Lat% = percent of requests that took longer than X seconds
CPU Eff = Rate divided by CPU% - throughput per cpu load
Sequential Reads
Num Avg Maximum Lat% Lat% CPU
Kernel Thr Rate (CPU%) Latency Latency >2s >10s Eff
------------------ --- ------------------------------------------------------------
2.5.70-ext2 1 28.92 13.54% 0.402 670.39 0.00000 0.00000 214
2.5.70-ext3 1 28.91 13.51% 0.402 696.75 0.00000 0.00000 214
2.5.70-xfs 1 28.59 13.71% 0.407 738.63 0.00000 0.00000 208
2.5.70-reiserfs 1 28.42 15.07% 0.410 73.45 0.00000 0.00000 188
2.5.70-xfs 8 55.60 28.80% 1.616 757.41 0.00000 0.00000 193
2.5.70-ext2 8 36.13 18.05% 2.578 967.76 0.00000 0.00000 200
2.5.70-reiserfs 8 31.26 18.04% 2.973 265.99 0.00000 0.00000 173
2.5.70-ext3 8 17.27 11.90% 5.278 3321.04 0.00000 0.00000 145
2.5.70-xfs 16 50.52 25.77% 3.532 877.59 0.00000 0.00000 196
2.5.70-ext2 16 30.83 15.31% 6.012 1149.26 0.00000 0.00000 201
2.5.70-reiserfs 16 26.98 15.44% 6.884 594.97 0.00000 0.00000 175
2.5.70-ext3 16 12.15 8.84% 14.936 5434.13 0.00033 0.00000 137
2.5.70-xfs 32 49.41 25.84% 7.477 1154.73 0.00000 0.00000 191
2.5.70-ext2 32 28.38 13.84% 13.084 1523.90 0.00000 0.00000 205
2.5.70-reiserfs 32 27.90 15.32% 13.309 1337.08 0.00000 0.00000 182
2.5.70-ext3 32 8.47 7.83% 40.867 15191.64 0.07029 0.00000 108
2.5.70-xfs 64 46.24 24.20% 15.655 1618.93 0.00000 0.00000 191
2.5.70-ext2 64 27.78 9.18% 25.852 29261.64 0.09457 0.00005 303
2.5.70-reiserfs 64 27.07 9.41% 26.668 36274.53 0.15678 0.00053 287
2.5.70-ext3 64 8.18 7.95% 78.515 32956.01 1.30109 0.00024 103
2.5.70-xfs 128 47.28 24.68% 30.299 8088.06 0.00048 0.00000 192
2.5.70-ext2 128 31.29 11.40% 40.825 76550.19 0.34548 0.03205 274
2.5.70-reiserfs 128 27.95 10.24% 46.841 88666.98 0.50650 0.04029 273
2.5.70-ext3 128 8.07 10.04% 147.272 55941.86 2.85859 0.04143 80
2.5.70-xfs 256 47.72 24.96% 52.069 185880.04 0.21720 0.07501 191
2.5.70-ext2 256 34.82 13.95% 66.450 134454.22 0.57445 0.13957 249
2.5.70-reiserfs 256 26.39 10.16% 98.556 129873.62 1.11609 0.23409 260
2.5.70-ext3 256 8.08 15.67% 285.410 108616.57 3.94969 0.69022 52
Random Reads
Num Avg Maximum Lat% Lat% CPU
Kernel Thr Rate (CPU%) Latency Latency >2s >10s Eff
------------------ --- ------------------------------------------------------------
2.5.70-xfs 1 1.04 0.95% 11.229 61.99 0.00000 0.00000 110
2.5.70-reiserfs 1 0.96 1.01% 12.233 109.56 0.00000 0.00000 95
2.5.70-ext3 1 0.90 0.90% 13.025 78.39 0.00000 0.00000 100
2.5.70-ext2 1 0.84 0.80% 13.987 69.70 0.00000 0.00000 105
2.5.70-xfs 8 5.23 4.56% 16.663 109.98 0.00000 0.00000 115
2.5.70-ext3 8 5.07 4.62% 16.711 105.20 0.00000 0.00000 110
2.5.70-reiserfs 8 4.86 5.46% 17.939 121.22 0.00000 0.00000 89
2.5.70-ext2 8 4.31 4.46% 20.156 135.14 0.00000 0.00000 97
2.5.70-ext3 16 5.46 4.40% 29.942 166.29 0.00000 0.00000 124
2.5.70-xfs 16 5.46 4.30% 32.314 181.00 0.00000 0.00000 127
2.5.70-reiserfs 16 4.80 4.71% 36.803 208.73 0.00000 0.00000 102
2.5.70-ext2 16 4.48 3.85% 39.028 201.53 0.00000 0.00000 116
2.5.70-xfs 32 5.58 5.34% 61.650 208.50 0.00000 0.00000 104
2.5.70-ext3 32 5.50 4.81% 55.804 225.78 0.00000 0.00000 114
2.5.70-reiserfs 32 4.78 4.85% 72.547 336.00 0.00000 0.00000 98
2.5.70-ext2 32 4.51 4.24% 78.365 350.57 0.00000 0.00000 106
2.5.70-ext3 64 5.16 4.54% 109.937 401.81 0.00000 0.00000 114
2.5.70-xfs 64 4.89 4.31% 133.968 384.25 0.00000 0.00000 113
2.5.70-ext2 64 4.87 4.50% 139.364 631.17 0.00000 0.00000 108
2.5.70-reiserfs 64 4.70 5.19% 139.623 676.06 0.00000 0.00000 90
2.5.70-ext3 128 5.49 5.19% 203.859 700.59 0.00000 0.00000 106
2.5.70-reiserfs 128 5.25 5.59% 220.590 1291.70 0.00000 0.00000 94
2.5.70-ext2 128 5.23 5.39% 222.819 1656.22 0.00000 0.00000 97
2.5.70-xfs 128 5.12 4.83% 244.213 693.75 0.00000 0.00000 106
2.5.70-ext3 256 5.66 5.96% 340.087 5869.11 0.00000 0.00000 95
2.5.70-reiserfs 256 5.39 6.49% 366.114 6565.36 0.39063 0.00000 83
2.5.70-ext2 256 5.30 5.57% 358.209 6453.99 0.44271 0.00000 95
2.5.70-xfs 256 4.91 4.87% 397.934 6717.78 1.01562 0.00000 101
Sequential Writes
xfs max latency starts is higher than other filesystems when thread count is low,
when thread count is high, xfs max latency is lowest.
Num Avg Maximum Lat% Lat% CPU
Kernel Thr Rate (CPU%) Latency Latency >2s >10s Eff
------------------ --- ------------------------------------------------------------
2.5.70-ext2 1 56.39 41.44% 0.170 2238.18 0.00000 0.00000 136
2.5.70-xfs 1 56.14 46.50% 0.175 69039.98 0.00030 0.00030 121
2.5.70-reiserfs 1 54.77 90.98% 0.202 1959.03 0.00000 0.00000 60
2.5.70-ext3 1 52.89 61.79% 0.210 15213.72 0.00114 0.00000 86
2.5.70-xfs 8 58.01 70.52% 1.090 148705.27 0.00535 0.00315 82
2.5.70-ext2 8 33.64 31.92% 2.221 32241.37 0.03910 0.00020 105
2.5.70-reiserfs 8 26.10 105.9% 3.147 99959.42 0.02638 0.00497 25
2.5.70-ext3 8 3.37 5.67% 21.474 225029.19 0.08302 0.05707 59
2.5.70-xfs 16 57.48 78.49% 2.146 194628.15 0.00901 0.00525 73
2.5.70-ext2 16 32.77 33.64% 4.280 67553.19 0.05822 0.00529 97
2.5.70-reiserfs 16 26.16 110.4% 5.933 112234.69 0.04840 0.00963 24
2.5.70-ext3 16 2.39 4.40% 62.682 511062.66 0.16084 0.09003 54
2.5.70-xfs 32 57.69 83.55% 4.459 236439.33 0.02103 0.01159 69
2.5.70-ext2 32 28.81 32.52% 9.324 151405.98 0.08040 0.03066 89
2.5.70-reiserfs 32 28.47 106.6% 9.921 108081.18 0.10381 0.02303 27
2.5.70-ext3 32 2.11 4.89% 135.246 948909.66 0.29182 0.13547 43
2.5.70-xfs 64 57.46 84.46% 8.238 272236.51 0.04845 0.01850 68
2.5.70-ext2 64 28.86 36.76% 18.376 219132.56 0.13365 0.06003 78
2.5.70-reiserfs 64 26.48 104.7% 20.644 221498.65 0.13495 0.06079 25
2.5.70-ext3 64 1.99 5.29% 276.918 1696330.83 0.50163 0.24386 38
2.5.70-xfs 128 57.18 85.25% 16.545 276182.86 0.11464 0.04244 67
2.5.70-ext2 128 31.83 40.35% 29.891 406660.86 0.20223 0.08163 79
2.5.70-reiserfs 128 26.53 114.8% 38.856 306137.57 0.29292 0.10090 23
2.5.70-ext3 128 2.09 6.28% 492.454 3187897.55 0.88648 0.53048 33
2.5.70-xfs 256 56.39 83.96% 29.725 351180.99 0.16737 0.07796 67
2.5.70-ext2 256 35.58 49.97% 51.891 493133.54 0.30092 0.13971 71
2.5.70-reiserfs 256 21.33 161.4% 97.778 504849.92 0.82383 0.21262 13
2.5.70-ext3 256 2.32 11.65% 830.694 4474318.91 1.66059 1.01834 20
Random Writes
Num Avg Maximum Lat% Lat% CPU
Kernel Thr Rate (CPU%) Latency Latency >2s >10s Eff
------------------ --- ------------------------------------------------------------
2.5.70-xfs 1 3.91 3.74% 0.076 1.55 0.00000 0.00000 105
2.5.70-ext2 1 3.30 3.06% 0.622 57.66 0.00000 0.00000 108
2.5.70-reiserfs 1 2.83 3.22% 1.218 430.83 0.00000 0.00000 88
2.5.70-ext3 1 2.06 2.26% 2.764 457.66 0.00000 0.00000 91
2.5.70-ext2 8 3.79 6.12% 1.377 76.30 0.00000 0.00000 62
2.5.70-reiserfs 8 3.56 3.81% 3.918 1119.27 0.00000 0.00000 93
2.5.70-ext3 8 3.47 6.20% 3.609 114.94 0.00000 0.00000 56
2.5.70-xfs 8 3.26 5.46% 0.708 1864.91 0.00000 0.00000 60
2.5.70-ext2 16 3.79 4.20% 2.223 410.09 0.00000 0.00000 90
2.5.70-reiserfs 16 3.62 3.82% 7.063 1291.33 0.00000 0.00000 95
2.5.70-xfs 16 3.49 7.84% 0.454 1143.64 0.00000 0.00000 44
2.5.70-ext3 16 3.43 5.51% 6.481 835.05 0.00000 0.00000 62
2.5.70-xfs 32 3.77 6.66% 0.349 448.57 0.00000 0.00000 57
2.5.70-reiserfs 32 3.76 3.74% 7.401 3053.12 0.02500 0.00000 100
2.5.70-ext2 32 3.71 4.37% 2.026 296.68 0.00000 0.00000 85
2.5.70-ext3 32 3.59 7.92% 7.573 907.51 0.00000 0.00000 45
2.5.70-ext2 64 3.81 5.23% 2.277 387.93 0.00000 0.00000 73
2.5.70-reiserfs 64 3.78 6.02% 5.438 866.49 0.00000 0.00000 63
2.5.70-xfs 64 3.74 7.29% 0.222 184.15 0.00000 0.00000 51
2.5.70-ext3 64 3.56 8.74% 10.799 1276.71 0.00000 0.00000 41
2.5.70-ext2 128 3.86 6.83% 2.844 4862.76 0.02520 0.00000 56
2.5.70-reiserfs 128 3.85 9.40% 4.324 2293.99 0.00000 0.00000 41
2.5.70-ext3 128 3.68 12.35% 10.807 873.82 0.00000 0.00000 30
2.5.70-xfs 128 3.63 6.93% 2.272 7922.55 0.07560 0.00000 52
2.5.70-ext2 256 3.62 6.43% 6.602 8693.57 0.05208 0.00000 56
2.5.70-ext3 256 3.56 16.74% 39.608 1447.47 0.00000 0.00000 21
2.5.70-reiserfs 256 3.47 6.57% 11.627 3941.93 0.00000 0.00000 53
2.5.70-xfs 256 3.42 6.56% 2.040 3681.95 0.02604 0.00000 52
--
Randy Hron
http://home.earthlink.net/~rwhron/kernel/bigbox.html
[email protected] writes:
> --------------- Sequential ----------
> ----- Create ----- ---- Delete ----
> /sec %CPU Eff /sec %CPU Eff
> 2.5.70-reiserfs 7584 86.7 8751 2628 37.3 7038
> 2.5.70-xfs 1710 39.3 4347 2053 28.3 7247
> 2.5.70-ext2 150 99.0 151 60883 100.0 6088
> 2.5.70-ext3 119 95.0 126 26319 87.7 3002
It's quite surprising that reiserfs is so slow at deletion. In my
normal experience reiserfs rm -rf is much faster than anything else
(e.g. with a big rm -rf on an ext2 you have a chance to ctrl-c still,
on reiserfs no such chance; XFS is really slow at this). Perhaps this
is some 2.5 regression? Do you have 2.4 comparison numbers?
-Andi
> It's quite surprising that reiserfs is so slow at deletion. In my
> normal experience reiserfs rm -rf is much faster than anything else
> (e.g. with a big rm -rf on an ext2 you have a chance to ctrl-c still,
> on reiserfs no such chance; XFS is really slow at this). Perhaps this
> is some 2.5 regression? Do you have 2.4 comparison numbers?
Maybe the other filesystems are just catching up :)
My experience is reiserfs is amazingly fast at rm -rf.
Here is bonnie++ small file benchmark on reiserfs with more kernels.
A couple of notes. You see the number of files was reduced recently.
Also the reiserfs notail option was removed based on a suggestion from
Hans to benefit bigger file benchmarks.
--------------- Sequential ---------
----- Create ----- ---- Delete ----
files /sec %CPU Eff /sec %CPU Eff
2.4.19-rmap13c 131072 3565 40.7 8766 2212 33.3 6635
2.4.20-jam2 131072 3702 43.3 8543 2148 31.3 6855
2.4.21-pre4-ac3 131072 3372 40.3 8360 2187 31.3 6980
2.4.21-pre4aa1 131072 3612 43.7 8273 2141 31.0 6905
2.5.68 131072 2935 37.3 7861 1787 25.7 6963
2.5.68-mm2 131072 3031 38.3 7906 1776 26.3 6743
2.5.68-mjb2 65536 7652 86.7 8830 4027 56.7 7105
2.5.69 65536 7884 90.3 8727 3244 45.7 7102
2.5.69-bk1 65536 7694 88.0 8743 3419 48.3 7073
2.5.69-mm5 65536 7585 87.0 8719 3538 50.3 7029
2.5.70 65536 7584 86.7 8751 2628 37.3 7038
2.5.69 was about 20% faster than 2.5.70 on sequential file deletes
on reiserfs.
I haven't benchmarked any 2.4 kernels with 65536 files and
tails yet.
--
Randy Hron
http://home.earthlink.net/~rwhron/kernel/bigbox.html
You should expect V3 to be slow at quad CPU smp benchmarks because
balancing is giant locked. V4 fixes this with fine grained locking, we
hope to release at Linux Tag in July. Fortunately quad CPU boxes will
not be economical compared to dual CPU boxes until V4 has been out for a
while....;-) V4 CPU usage and general performance has gotten a lot
better, we need to put a new snapshot on our website.....
Andi Kleen wrote:
>[email protected] writes:
>
>
>> --------------- Sequential ----------
>> ----- Create ----- ---- Delete ----
>> /sec %CPU Eff /sec %CPU Eff
>>2.5.70-reiserfs 7584 86.7 8751 2628 37.3 7038
>>2.5.70-xfs 1710 39.3 4347 2053 28.3 7247
>>2.5.70-ext2 150 99.0 151 60883 100.0 6088
>>2.5.70-ext3 119 95.0 126 26319 87.7 3002
>>
>>
>
>It's quite surprising that reiserfs is so slow at deletion. In my
>normal experience reiserfs rm -rf is much faster than anything else
>(e.g. with a big rm -rf on an ext2 you have a chance to ctrl-c still,
>on reiserfs no such chance; XFS is really slow at this). Perhaps this
>is some 2.5 regression? Do you have 2.4 comparison numbers?
>
>-Andi
>
>-
>To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
>the body of a message to [email protected]
>More majordomo info at http://vger.kernel.org/majordomo-info.html
>Please read the FAQ at http://www.tux.org/lkml/
>
>
>
>
--
Hans