5

I am experiencing very low IOPS with a SSD in my server. I noticed this when running a mysql database server, which performed very bad when there were many (~100 per second) updates to an InnoDB Database.

Here are the server specs:

SSD: Model=Samsung SSD 850 EVO 120GB, FwRev=EMT01B6Q

Server: HP proliant DL320e Gen8

CPU: Intel(R) Xeon(R) CPU E3-1270 v3 @ 3.50GHz

OS: Ubuntu 14.04.1 LTS

Kernel: Linux h119 3.13.0-44-generic #73-Ubuntu SMP Tue Dec 16 00:22:43 UTC 2014 x86_64 x86_64 x86_64 GNU/Linux

When checking the SSD while there's some load on it, I'm getting these results with iostat -kx 1 20

root@h119:~# iostat -kx 1 20
Linux 3.13.0-44-generic (h119)  02/19/2015      _x86_64_        (8 CPU)

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           0.14    0.00    0.08    1.06    0.00   98.72

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sda               5.14    14.75    0.80   17.68   378.51   258.59    68.94     0.22   12.03    1.76   12.50   7.90  14.60
sdb               0.00    19.89    0.00   18.44     0.05   636.16    68.99     0.26   14.32    0.43   14.33   8.22  15.17
md2               0.00     0.00    0.04   31.40     0.97   257.88    16.46     0.00    0.00    0.00    0.00   0.00   0.00
md1               0.00     0.00    0.00    0.00     0.00     0.00     7.93     0.00    0.00    0.00    0.00   0.00   0.00
md0               0.00     0.00    0.00    0.00     0.00     0.00     8.00     0.00    0.00    0.00    0.00   0.00   0.00

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           1.00    0.00    0.63    1.00    0.00   97.37

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sda               0.00    21.00    0.00   18.00     0.00   346.00    38.44     0.19   10.44    0.00   10.44   7.78  14.00
sdb               0.00    21.00    0.00   18.00     0.00   346.00    38.44     0.18   10.00    0.00   10.00   7.33  13.20
md2               0.00     0.00    0.00   36.00     0.00   356.00    19.78     0.00    0.00    0.00    0.00   0.00   0.00
md1               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00
md0               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           1.25    0.00    0.75    1.50    0.00   96.50

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sda               0.00    19.00    0.00   19.00     0.00   218.00    22.95     0.20    9.47    0.00    9.47   9.89  18.80
sdb               0.00    19.00    0.00   19.00     0.00   218.00    22.95     0.20    9.68    0.00    9.68  10.11  19.20
md2               0.00     0.00    0.00   37.00     0.00   236.00    12.76     0.00    0.00    0.00    0.00   0.00   0.00
md1               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00
md0               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           1.75    0.00    1.12    1.37    0.00   95.76

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sda               0.00    19.80    0.00   16.83     0.00   291.09    34.59     0.19    9.88    0.00    9.88   9.18  15.45
sdb               0.00    19.80    0.00   16.83     0.00   291.09    34.59     0.19    9.88    0.00    9.88   9.18  15.45
md2               0.00     0.00    0.00   35.64     0.00   336.63    18.89     0.00    0.00    0.00    0.00   0.00   0.00
md1               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00
md0               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           1.00    0.00    0.75    1.00    0.00   97.25

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sda               0.00    18.00    0.00   22.00     0.00   310.00    28.18     0.16    8.91    0.00    8.91   6.36  14.00
sdb               0.00    18.00    0.00   22.00     0.00   310.00    28.18     0.16    9.09    0.00    9.09   6.55  14.40
md2               0.00     0.00    0.00   32.00     0.00   228.00    14.25     0.00    0.00    0.00    0.00   0.00   0.00
md1               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00
md0               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           0.75    0.00    0.75    1.00    0.00   97.50

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sda               0.00    10.00    0.00   13.00     0.00   142.00    21.85     0.13    9.85    0.00    9.85   9.85  12.80
sdb               0.00    10.00    0.00   13.00     0.00   142.00    21.85     0.12    9.54    0.00    9.54   9.54  12.40
md2               0.00     0.00    0.00   19.00     0.00   140.00    14.74     0.00    0.00    0.00    0.00   0.00   0.00
md1               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00
md0               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           1.26    0.00    0.88    1.26    0.00   96.61

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sda               0.00    34.00    0.00   12.00     0.00   333.00    55.50     0.12   10.00    0.00   10.00  10.33  12.40
sdb               0.00    34.00    0.00   12.00     0.00   333.00    55.50     0.12   10.00    0.00   10.00  10.33  12.40
md2               0.00     0.00    0.00   45.00     0.00   352.00    15.64     0.00    0.00    0.00    0.00   0.00   0.00
md1               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00
md0               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           1.37    0.00    0.87    1.50    0.00   96.26

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sda               0.00    17.00    0.00   28.00     0.00   713.00    50.93     0.28   10.00    0.00   10.00   5.86  16.40
sdb               0.00    17.00    0.00   28.00     0.00   713.00    50.93     0.27    9.86    0.00    9.86   5.71  16.00
md2               0.00     0.00    0.00   43.00     0.00   692.00    32.19     0.00    0.00    0.00    0.00   0.00   0.00
md1               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00
md0               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           1.50    0.00    0.75    1.38    0.00   96.37

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sda               0.00    35.00    0.00   20.00     0.00   361.00    36.10     0.18    9.20    0.00    9.20   8.00  16.00
sdb               0.00    35.00    0.00   20.00     0.00   361.00    36.10     0.18    9.20    0.00    9.20   8.00  16.00
md2               0.00     0.00    0.00   53.00     0.00   360.00    13.58     0.00    0.00    0.00    0.00   0.00   0.00
md1               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00
md0               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           1.25    0.00    1.12    1.12    0.00   96.50

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sda               0.00    20.00    0.00   11.00     0.00   193.00    35.09     0.11    9.82    0.00    9.82   9.82  10.80
sdb               0.00    20.00    0.00   11.00     0.00   193.00    35.09     0.11   10.18    0.00   10.18  10.18  11.20
md2               0.00     0.00    0.00   29.00     0.00   192.00    13.24     0.00    0.00    0.00    0.00   0.00   0.00
md1               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00
md0               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           1.00    0.00    0.50    0.62    0.00   97.88

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sda               0.00    12.00    0.00    9.00     0.00   137.00    30.44     0.09    9.78    0.00    9.78   8.44   7.60
sdb               0.00    12.00    0.00    9.00     0.00   137.00    30.44     0.09   10.22    0.00   10.22   8.89   8.00
md2               0.00     0.00    0.00   19.00     0.00   136.00    14.32     0.00    0.00    0.00    0.00   0.00   0.00
md1               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00
md0               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           1.13    0.00    0.63    0.63    0.00   97.62

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sda               0.00    15.00    0.00    8.00     0.00   181.00    45.25     0.08   10.00    0.00   10.00  10.00   8.00
sdb               0.00    15.00    0.00    8.00     0.00   181.00    45.25     0.07    9.00    0.00    9.00   9.00   7.20
md2               0.00     0.00    0.00   25.00     0.00   244.00    19.52     0.00    0.00    0.00    0.00   0.00   0.00
md1               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00
md0               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           0.88    0.00    0.63    0.88    0.00   97.60

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sda               0.00    28.00    0.00   15.00     0.00   301.00    40.13     0.14    9.33    0.00    9.33   6.93  10.40
sdb               0.00    28.00    0.00   15.00     0.00   301.00    40.13     0.14    9.33    0.00    9.33   6.93  10.40
md2               0.00     0.00    0.00   38.00     0.00   236.00    12.42     0.00    0.00    0.00    0.00   0.00   0.00
md1               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00
md0               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           0.75    0.00    1.25    0.50    0.00   97.50

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sda               0.00    15.00    0.00    7.00     0.00   177.00    50.57     0.08    9.71    0.00    9.71  10.86   7.60
sdb               0.00    15.00    0.00    7.00     0.00   177.00    50.57     0.07    9.14    0.00    9.14  10.29   7.20
md2               0.00     0.00    0.00   22.00     0.00   188.00    17.09     0.00    0.00    0.00    0.00   0.00   0.00
md1               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00
md0               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           0.76    0.00    0.38    1.13    0.00   97.73

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sda               0.00    17.00    0.00   17.00     0.00   205.00    24.12     0.16    9.65    0.00    9.65   7.06  12.00
sdb               0.00    17.00    0.00   17.00     0.00   205.00    24.12     0.16    9.65    0.00    9.65   7.06  12.00
md2               0.00     0.00    0.00   33.00     0.00   196.00    11.88     0.00    0.00    0.00    0.00   0.00   0.00
md1               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00
md0               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           1.38    0.00    0.63    0.88    0.00   97.11

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sda               0.00    19.00    0.00   11.00     0.00   245.00    44.55     0.10    9.45    0.00    9.45   9.45  10.40
sdb               0.00    19.00    0.00   11.00     0.00   245.00    44.55     0.10    9.45    0.00    9.45   9.45  10.40
md2               0.00     0.00    0.00   28.00     0.00   244.00    17.43     0.00    0.00    0.00    0.00   0.00   0.00
md1               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00
md0               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           0.76    0.00    0.50    1.39    0.00   97.35

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sda               0.00    28.00    0.00   20.00     0.00   305.00    30.50     0.19    9.40    0.00    9.40   8.60  17.20
sdb               0.00    28.00    0.00   20.00     0.00   305.00    30.50     0.19    9.40    0.00    9.40   8.60  17.20
md2               0.00     0.00    0.00   47.00     0.00   304.00    12.94     0.00    0.00    0.00    0.00   0.00   0.00
md1               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00
md0               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           1.01    0.00    0.63    1.13    0.00   97.23

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sda               0.00    28.00    0.00   16.00     0.00   357.00    44.62     0.15    9.75    0.00    9.75   8.50  13.60
sdb               0.00    28.00    0.00   16.00     0.00   357.00    44.62     0.15    9.50    0.00    9.50   8.25  13.20
md2               0.00     0.00    0.00   44.00     0.00   356.00    16.18     0.00    0.00    0.00    0.00   0.00   0.00
md1               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00
md0               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           1.13    0.00    0.75    1.50    0.00   96.62

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sda               0.00    19.00    0.00   22.00     0.00   297.00    27.00     0.19    8.91    0.00    8.91   5.82  12.80
sdb               0.00    19.00    0.00   22.00     0.00   297.00    27.00     0.22   10.36    0.00   10.36   6.36  14.00
md2               0.00     0.00    0.00   40.00     0.00   296.00    14.80     0.00    0.00    0.00    0.00   0.00   0.00
md1               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00
md0               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           1.13    0.00    0.75    1.13    0.00   96.99

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sda               0.00    23.00    0.00   18.00     0.00   313.00    34.78     0.19   10.22    0.00   10.22   7.78  14.00
sdb               0.00    23.00    0.00   18.00     0.00   313.00    34.78     0.19   10.22    0.00   10.22   7.78  14.00
md2               0.00     0.00    0.00   39.00     0.00   312.00    16.00     0.00    0.00    0.00    0.00   0.00   0.00
md1               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00
md0               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00    0.00    0.00   0.00   0.00

The following result is with fio --rw=write --name=test --size=20M --direct=1

root@h119:~# fio --rw=write --name=test --size=20M --direct=1
test: (g=0): rw=write, bs=4K-4K/4K-4K/4K-4K, ioengine=sync, iodepth=1
fio-2.1.3
Starting 1 process
test: Laying out IO file(s) (1 file(s) / 20MB)
Jobs: 1 (f=1): [W] [100.0% done] [0KB/404KB/0KB /s] [0/101/0 iops] [eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=11288: Sat Feb 21 19:59:52 2015
  write: io=20480KB, bw=415022B/s, iops=101, runt= 50531msec
    clat (usec): min=9674, max=21523, avg=9867.58, stdev=615.15
     lat (usec): min=9675, max=21524, avg=9867.76, stdev=615.16
    clat percentiles (usec):
     |  1.00th=[ 9792],  5.00th=[ 9792], 10.00th=[ 9792], 20.00th=[ 9792],
     | 30.00th=[ 9792], 40.00th=[ 9792], 50.00th=[ 9792], 60.00th=[ 9792],
     | 70.00th=[ 9792], 80.00th=[ 9792], 90.00th=[ 9920], 95.00th=[10048],
     | 99.00th=[12480], 99.50th=[12864], 99.90th=[18304], 99.95th=[18560],
     | 99.99th=[21632]
    bw (KB  /s): min=  380, max=  411, per=100.00%, avg=405.33, stdev= 5.11
    lat (msec) : 10=90.66%, 20=9.32%, 50=0.02%
  cpu          : usr=0.03%, sys=0.19%, ctx=5138, majf=0, minf=26
  IO depths    : 1=100.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=64=0.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     issued    : total=r=0/w=5120/d=0, short=r=0/w=0/d=0

Run status group 0 (all jobs):
  WRITE: io=20480KB, aggrb=405KB/s, minb=405KB/s, maxb=405KB/s, mint=50531msec, maxt=50531msec

Disk stats (read/write):
    md2: ios=0/5580, merge=0/0, ticks=0/0, in_queue=0, util=0.00%, aggrios=0/5379, aggrmerge=0/191, aggrticks=0/52640, aggrin_queue=52640, aggrutil=99.00%
  sda: ios=0/5379, merge=0/191, ticks=0/52508, in_queue=52508, util=98.63%
  sdb: ios=0/5379, merge=0/191, ticks=0/52772, in_queue=52772, util=99.00%

As you can see, the IOPS are at 101, and I can only write 404kB/sec. Also, there's not much fluctuation on the IOPS while the tool is running, it's always between 101 and 103.

I also trust the results of fio, as my database gets slow when it hits about 100 updates/inserts per second.

With an identical server (hard- and software) and no RAID, I get the exactly same results, so this can't be due to the software-RAID:

root@h073:~# fio --rw=write --name=test --size=20M --direct=1
test: (g=0): rw=write, bs=4K-4K/4K-4K/4K-4K, ioengine=sync, iodepth=1
fio-2.1.3
Starting 1 process
Jobs: 1 (f=1): [W] [100.0% done] [0KB/408KB/0KB /s] [0/102/0 iops] [eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=15147: Sat Feb 21 20:52:14 2015
  write: io=20480KB, bw=418492B/s, iops=102, runt= 50112msec
    clat (usec): min=1504, max=17337, avg=9785.74, stdev=417.14
     lat (usec): min=1504, max=17337, avg=9785.93, stdev=417.13
    clat percentiles (usec):
     |  1.00th=[ 9792],  5.00th=[ 9792], 10.00th=[ 9792], 20.00th=[ 9792],
     | 30.00th=[ 9792], 40.00th=[ 9792], 50.00th=[ 9792], 60.00th=[ 9792],
     | 70.00th=[ 9792], 80.00th=[ 9792], 90.00th=[ 9792], 95.00th=[ 9792],
     | 99.00th=[12480], 99.50th=[12608], 99.90th=[13120], 99.95th=[15040],
     | 99.99th=[17280]
    bw (KB  /s): min=  400, max=  416, per=100.00%, avg=408.75, stdev= 2.69
    lat (msec) : 2=0.04%, 4=0.04%, 10=98.14%, 20=1.78%
  cpu          : usr=0.04%, sys=0.12%, ctx=5132, majf=0, minf=27
  IO depths    : 1=100.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=64=0.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     issued    : total=r=0/w=5120/d=0, short=r=0/w=0/d=0

Run status group 0 (all jobs):
  WRITE: io=20480KB, aggrb=408KB/s, minb=408KB/s, maxb=408KB/s, mint=50112msec, maxt=50112msec

Disk stats (read/write):
  sda: ios=0/5132, merge=0/43, ticks=0/50048, in_queue=50048, util=99.52%

What is remarkable is that the speed is very constant at 404kB/sec at 101-103 IOPS.

Copying files, however, is very fast and fulfills my expectations to a SSD:

root@h119:~# dd if=randomfile of=randomfile2
2097152+0 records in
2097152+0 records out
1073741824 bytes (1.1 GB) copied, 2.27684 s, 472 MB/s

So it seems only 4k random writes are slow as hell. If there's some more information you need, please let me know and I'll update it. Thanks in advance for your answer.

Update: The issue doesn't persist any more with a non-HP server. So it must have something to do how HP servers access the SSD. On the other hardware I get the following results, which are perfectly fine:

root@ca286:~# fio --rw=write --name=test --size=40M --direct=1
test: (g=0): rw=write, bs=4K-4K/4K-4K/4K-4K, ioengine=sync, iodepth=1
fio-2.1.3
Starting 1 process
test: Laying out IO file(s) (1 file(s) / 40MB)

test: (groupid=0, jobs=1): err= 0: pid=1285: Wed Feb 25 20:47:25 2015
  write: io=40960KB, bw=130032KB/s, iops=32507, runt=   315msec
    clat (usec): min=24, max=1588, avg=30.34, stdev=15.59
     lat (usec): min=24, max=1589, avg=30.39, stdev=15.60
    clat percentiles (usec):
     |  1.00th=[   26],  5.00th=[   29], 10.00th=[   29], 20.00th=[   30],
     | 30.00th=[   30], 40.00th=[   30], 50.00th=[   30], 60.00th=[   30],
     | 70.00th=[   30], 80.00th=[   31], 90.00th=[   31], 95.00th=[   31],
     | 99.00th=[   33], 99.50th=[   54], 99.90th=[   56], 99.95th=[   56],
     | 99.99th=[  114]
    lat (usec) : 50=99.30%, 100=0.68%, 250=0.01%
    lat (msec) : 2=0.01%
  cpu          : usr=2.55%, sys=18.47%, ctx=10247, majf=0, minf=26
  IO depths    : 1=100.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=64=0.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     issued    : total=r=0/w=10240/d=0, short=r=0/w=0/d=0

Run status group 0 (all jobs):
  WRITE: io=40960KB, aggrb=130031KB/s, minb=130031KB/s, maxb=130031KB/s, mint=315msec, maxt=315msec

Disk stats (read/write):
  sda: ios=0/4806, merge=0/0, ticks=0/120, in_queue=120, util=48.19%
  • 1
    Why are you using a consumer drive in a HP server - this site is for professional sysadmins as we make very clear when you sign up - and we'd never do this, it goes against everything HP state and support. You're unlikely to get anyone looking at your question because of this. – Chopper3 Feb 21 '15 at 20:20
  • 2
    @Chopper3 not everyone has $1,100 per disk that HP wants for SSDs. I have a bunch of 850 Evo Pro's and get around 250MB/sec 4k random writes and 900MB/sec sequential reads in similar hardware (no caching), so something is definitely not right, and there's no guarantee that forking out the thousands of dollars for the same drive with a HP sticker on them will make any difference. – Mark Henderson Feb 21 '15 at 21:26
  • 1
    Which RAID controller is in use here? Does your server have a Smart Array controller in it? – ewwhite Feb 23 '15 at 2:29
  • @ewwhite one server has software RAID, the other has no RAID configured at all. In both configurations the performance is the same. Also, there's no Smart Array in these servers. – Hook Feb 23 '15 at 21:09
1

We made the same observation at our tests with SSD in Gen7, but I guess the difference to Gen8 is not very big.

If you connect the SSD directly to the Board, the Performance is as you would expect ( the ProLiant servers have only one port for that where the DVD Drive is connected at default).

If you connect the SSD via the Raid Controller, no matter if you configure no RAID (Jbod) it slows you down, as long as the data has to go through the Controller the Performance is shitty. We guess it has to do with the controller not build for drives that fast.

If you buy HP SSDs and put them into such servers, your performance gets better( we guess because of better communication between Controller and Disk), but nowhere near the advertised performance(~20% of that).

So the ProLiants Gen7/8 SAS Controller, are simply not made for SSDs. It would be interesting if this bottleneck still exists with Gen9, because that way you can only connect one disk via SlimSATA to a Gen7/Gen8(and that only via putting the disk into the chassis and not in the slots)

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.