Wrong read speed result for Samsung SSD 970 EVO Plus #48

Vascom · 2021-03-07T07:40:26Z

Linux-distro: Fedora 33
Desktop Environment (KDE/GNOME etc.): KDE
Qt Version: 5.15.2
KDiskMark Version: 2.2.0
FIO Version: fio-3.21

Description:

Strange speed result for Samsung SSD 970 EVO Plus (1TB and others).
By datasheet and in gnome-disk-utility I have ~3500MB/s sequential read but kdiskmark give me only ~1500MB/s.
And write speed ~2900MB/s. So read speed seems not relevant.

Steps To Reproduce:

Run test.

Can you help solve this bug?

tim77 · 2021-03-07T07:45:24Z

Same issue. When i've tested first time 5 months ago KDiskMark shows real speed - 3500MB/s:

Now:

JonMagon · 2021-03-07T08:27:24Z

Try to disable the flush cache option in Settings. This may not work as expected. If it helps, I will disable this option by default.

Vascom · 2021-03-07T08:45:45Z

Disabling flush cache option not helped.

JonMagon · 2021-03-07T08:52:40Z

Could you find the release the issue appeared the first time?

tim77 · 2021-03-07T08:54:46Z

As far i know @Vascom already tried this. I thought as well that this issue appears in some KDiskMark updates but seems like something else causing it.

JonMagon · 2021-03-07T08:56:19Z

I can't reproduce it, because I haven't such an SSD, but it's definitely a bug.

tim77 · 2021-03-07T08:59:36Z

Similar discussion on Reddit and the same disk. Doubt that this could help but a least seems like popular issue.

Vascom · 2021-03-07T08:59:54Z

May be we can give you some debug output?

JonMagon · 2021-03-07T10:12:52Z

Can you please check with 2.0.0 and 1.6.2?

mkdir build && cd build
cmake -D CMAKE_BUILD_TYPE=Release ..
cmake --build .

tim77 · 2021-03-07T20:16:42Z

Tested with 2.0.0 and 1.6.2 and 💯 the same issue. Maybe this the kernel bug or something? Very weird since my first test on the same OS and same SSD was OK.

JonMagon · 2021-03-07T21:28:22Z

Then it may be caused by separation loops inside kdiskmark. Could you please try also 1.6.0?

Vascom · 2021-03-08T05:30:40Z

The same with 1.6.0.

JonMagon · 2021-03-08T06:17:15Z

Then it can be really not kdiskmark issue.

JonMagon · 2021-03-09T09:36:58Z

What is the output of the commands below?

echo 3 | sudo tee /proc/sys/vm/drop_caches
fio --ioengine=libaio --direct=1 --randrepeat=0 --refill_buffers --end_fsync=1 --filename=test-fio.tmp --name=readjob --size=128M --bs=1m --rw=read --iodepth=8 --loops=5

echo 3 | sudo tee /proc/sys/vm/drop_caches
fio --ioengine=libaio --direct=1 --filename=test-fio.tmp --name=readjob --size=128M --bs=1m --rw=read --iodepth=8 --loops=5

Vascom · 2021-03-09T09:46:49Z

Last login: Tue Mar  9 12:26:53 2021 from 10.9.0.1
[vascom@milkyway ~]$ fio --ioengine=libaio --direct=1 --randrepeat=0 --refill_buffers --end_fsync=1 --filename=test-fio.tmp --name=readjob --size=128M --bs=1m --rw=read --iodepth=8 --loops=5
readjob: (g=0): rw=read, bs=(R) 1024KiB-1024KiB, (W) 1024KiB-1024KiB, (T) 1024KiB-1024KiB, ioengine=libaio, iodepth=8
fio-3.21
Starting 1 process
readjob: Laying out IO file (1 file / 128MiB)

readjob: (groupid=0, jobs=1): err= 0: pid=4886: Tue Mar  9 12:45:11 2021
  read: IOPS=1376, BW=1376MiB/s (1443MB/s)(640MiB/465msec)
    slat (usec): min=398, max=2304, avg=715.00, stdev=285.00
    clat (usec): min=313, max=14322, avg=4882.13, stdev=1889.90
     lat (usec): min=890, max=16629, avg=5598.28, stdev=2146.95
    clat percentiles (usec):
     |  1.00th=[ 1029],  5.00th=[ 4015], 10.00th=[ 4113], 20.00th=[ 4228],
     | 30.00th=[ 4293], 40.00th=[ 4359], 50.00th=[ 4490], 60.00th=[ 4621],
     | 70.00th=[ 4752], 80.00th=[ 4948], 90.00th=[ 5080], 95.00th=[11076],
     | 99.00th=[11731], 99.50th=[13042], 99.90th=[14353], 99.95th=[14353],
     | 99.99th=[14353]
  lat (usec)   : 500=0.78%, 1000=0.16%
  lat (msec)   : 2=1.41%, 4=2.50%, 10=88.59%, 20=6.56%
  cpu          : usr=1.29%, sys=81.72%, ctx=6, majf=0, minf=2059
  IO depths    : 1=0.8%, 2=1.6%, 4=3.1%, 8=94.5%, 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=99.2%, 8=0.8%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     issued rwts: total=640,0,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=8

Run status group 0 (all jobs):
   READ: bw=1376MiB/s (1443MB/s), 1376MiB/s-1376MiB/s (1443MB/s-1443MB/s), io=640MiB (671MB), run=465-465msec
[vascom@milkyway ~]$

tim77 · 2021-03-09T09:47:45Z

readjob: (g=0): rw=read, bs=(R) 1024KiB-1024KiB, (W) 1024KiB-1024KiB, (T) 1024KiB-1024KiB, ioengine=libaio, iodepth=8
fio-3.21
Starting 1 process

readjob: (groupid=0, jobs=1): err= 0: pid=279803: Tue Mar  9 11:46:33 2021
  read: IOPS=1649, BW=1649MiB/s (1730MB/s)(640MiB/388msec)
    slat (usec): min=331, max=4245, avg=601.14, stdev=252.80
    clat (usec): min=237, max=10631, avg=4064.16, stdev=1420.99
     lat (usec): min=731, max=14878, avg=4665.47, stdev=1630.65
    clat percentiles (usec):
     |  1.00th=[  775],  5.00th=[ 3458], 10.00th=[ 3556], 20.00th=[ 3589],
     | 30.00th=[ 3621], 40.00th=[ 3654], 50.00th=[ 3687], 60.00th=[ 3720],
     | 70.00th=[ 3752], 80.00th=[ 3851], 90.00th=[ 5669], 95.00th=[ 7767],
     | 99.00th=[ 9241], 99.50th=[ 9503], 99.90th=[10683], 99.95th=[10683],
     | 99.99th=[10683]
  lat (usec)   : 250=0.47%, 500=0.31%, 750=0.16%, 1000=0.62%
  lat (msec)   : 2=1.56%, 4=80.94%, 10=15.62%, 20=0.31%
  cpu          : usr=0.52%, sys=67.96%, ctx=27, majf=0, minf=2059
  IO depths    : 1=0.8%, 2=1.6%, 4=3.1%, 8=94.5%, 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=99.2%, 8=0.8%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     issued rwts: total=640,0,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=8

Run status group 0 (all jobs):
   READ: bw=1649MiB/s (1730MB/s), 1649MiB/s-1649MiB/s (1730MB/s-1730MB/s), io=640MiB (671MB), run=388-388msec

readjob: (g=0): rw=read, bs=(R) 1024KiB-1024KiB, (W) 1024KiB-1024KiB, (T) 1024KiB-1024KiB, ioengine=libaio, iodepth=8
fio-3.21
Starting 1 process

readjob: (groupid=0, jobs=1): err= 0: pid=279979: Tue Mar  9 11:47:23 2021
  read: IOPS=1649, BW=1649MiB/s (1730MB/s)(640MiB/388msec)
    slat (usec): min=335, max=4083, avg=601.38, stdev=233.62
    clat (usec): min=219, max=8998, avg=4076.16, stdev=1331.32
     lat (usec): min=717, max=13082, avg=4677.74, stdev=1517.70
    clat percentiles (usec):
     |  1.00th=[  758],  5.00th=[ 3359], 10.00th=[ 3523], 20.00th=[ 3589],
     | 30.00th=[ 3621], 40.00th=[ 3654], 50.00th=[ 3687], 60.00th=[ 3752],
     | 70.00th=[ 3785], 80.00th=[ 3884], 90.00th=[ 6783], 95.00th=[ 7046],
     | 99.00th=[ 7767], 99.50th=[ 7963], 99.90th=[ 8979], 99.95th=[ 8979],
     | 99.99th=[ 8979]
  lat (usec)   : 250=0.62%, 500=0.16%, 750=0.16%, 1000=0.62%
  lat (msec)   : 2=1.56%, 4=80.00%, 10=16.88%
  cpu          : usr=0.00%, sys=68.73%, ctx=28, majf=0, minf=2060
  IO depths    : 1=0.8%, 2=1.6%, 4=3.1%, 8=94.5%, 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=99.2%, 8=0.8%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     issued rwts: total=640,0,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=8

Run status group 0 (all jobs):
   READ: bw=1649MiB/s (1730MB/s), 1649MiB/s-1649MiB/s (1730MB/s-1730MB/s), io=640MiB (671MB), run=388-388msec

JonMagon · 2021-03-09T10:24:19Z

I would guess that this is the real reading performance.
What if test the speed with dd? Copy the all output.

dd if=/dev/zero of=test-fio.tmp bs=1M count=128
echo 3 | sudo tee /proc/sys/vm/drop_caches
dd if=test-fio.tmp of=/dev/null bs=1M

tim77 · 2021-03-09T10:26:16Z

echo 3 | sudo tee /proc/sys/vm/drop_caches
dd if=test-fio.tmp of=/dev/null bs=1M
128+0 записей получено
128+0 записей отправлено
134217728 байт (134 MB, 128 MiB) скопирован, 0,0916864 s, 1,5 GB/s
3
128+0 записей получено
128+0 записей отправлено
134217728 байт (134 MB, 128 MiB) скопирован, 0,0393467 s, 3,4 GB/s

JonMagon · 2021-03-09T12:56:46Z

No idea, something wrong with fio.

JonMagon · 2021-03-09T13:08:10Z

By the way, @tim77 haven't you changed partitions since the results was correct? Maybe alignment is wrong...

tim77 · 2021-03-09T13:26:35Z

By the way, @tim77 haven't you changed partitions since the results was correct? Maybe alignment is wrong...

Indeed. Partition was changed since first time. BTW i've also wondered about some another issue in BTRFS and i assumed is alignment can be the cause of it or not https://pagure.io/fedora-btrfs/project/issue/36#comment-701576

Vascom · 2021-03-09T20:09:45Z

I just found that it is seems not kdiskmark problem.
At 5.9.16 kernel I see full speed. But on 5.10-5.11.5 still half speed.

JonMagon · 2021-03-10T13:00:14Z

I think you should open an issue in the fio repository, I can't resolve it.

tim77 · 2021-03-10T13:04:35Z

In same time gnome-disks shows close to speed which rated by datasheet, but also clearly shown that it's half at beginning:

But i am definitely not trying to say that this is KDiskMark bug, no. But for end user this could not obvious.

JonMagon · 2021-03-10T13:06:46Z

I understand perfectly, but I can't fix it myself. It's one thing if it never worked right, it's quite another thing if it suddenly broke down, and with only one (?) device at that.

tim77 · 2021-03-10T13:09:01Z

Yep. We will try to file a bug into kernel and fio. Worth to try. At least maybe we can learn something new and maybe this is a feature.

Vascom added bug Something isn't working unconfirmed labels Mar 7, 2021

JonMagon added confirmed and removed unconfirmed labels Mar 7, 2021

JonMagon added the help wanted Extra attention is needed label Mar 9, 2021

JonMagon removed the help wanted Extra attention is needed label Mar 10, 2021

tim77 mentioned this issue Mar 10, 2021

Wrong read speed result for Samsung SSD 970 EVO Plus axboe/fio#1202

Closed

1 task

JonMagon added the related to fio label Mar 15, 2021

JonMagon pinned this issue Aug 23, 2022

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Wrong read speed result for Samsung SSD 970 EVO Plus #48

Wrong read speed result for Samsung SSD 970 EVO Plus #48

Vascom commented Mar 7, 2021

tim77 commented Mar 7, 2021

JonMagon commented Mar 7, 2021

Vascom commented Mar 7, 2021

JonMagon commented Mar 7, 2021

tim77 commented Mar 7, 2021

JonMagon commented Mar 7, 2021

tim77 commented Mar 7, 2021

Vascom commented Mar 7, 2021 •

edited

JonMagon commented Mar 7, 2021 •

edited

tim77 commented Mar 7, 2021

JonMagon commented Mar 7, 2021

Vascom commented Mar 8, 2021

JonMagon commented Mar 8, 2021

JonMagon commented Mar 9, 2021 •

edited

Vascom commented Mar 9, 2021

tim77 commented Mar 9, 2021

JonMagon commented Mar 9, 2021

tim77 commented Mar 9, 2021

JonMagon commented Mar 9, 2021

JonMagon commented Mar 9, 2021

tim77 commented Mar 9, 2021

Vascom commented Mar 9, 2021

JonMagon commented Mar 10, 2021

tim77 commented Mar 10, 2021

JonMagon commented Mar 10, 2021 •

edited

tim77 commented Mar 10, 2021

Wrong read speed result for Samsung SSD 970 EVO Plus #48

Wrong read speed result for Samsung SSD 970 EVO Plus #48

Comments

Vascom commented Mar 7, 2021

Description:

Steps To Reproduce:

tim77 commented Mar 7, 2021

JonMagon commented Mar 7, 2021

Vascom commented Mar 7, 2021

JonMagon commented Mar 7, 2021

tim77 commented Mar 7, 2021

JonMagon commented Mar 7, 2021

tim77 commented Mar 7, 2021

Vascom commented Mar 7, 2021 • edited

JonMagon commented Mar 7, 2021 • edited

tim77 commented Mar 7, 2021

JonMagon commented Mar 7, 2021

Vascom commented Mar 8, 2021

JonMagon commented Mar 8, 2021

JonMagon commented Mar 9, 2021 • edited

Vascom commented Mar 9, 2021

tim77 commented Mar 9, 2021

JonMagon commented Mar 9, 2021

tim77 commented Mar 9, 2021

JonMagon commented Mar 9, 2021

JonMagon commented Mar 9, 2021

tim77 commented Mar 9, 2021

Vascom commented Mar 9, 2021

JonMagon commented Mar 10, 2021

tim77 commented Mar 10, 2021

JonMagon commented Mar 10, 2021 • edited

tim77 commented Mar 10, 2021

Vascom commented Mar 7, 2021 •

edited

JonMagon commented Mar 7, 2021 •

edited

JonMagon commented Mar 9, 2021 •

edited

JonMagon commented Mar 10, 2021 •

edited