I noticed that docker containers were slow on writing to mapped volumes, the dd inside the docker was 20x faster. First i thought it was something wrong with docker but then i did a dd test on the server

 

root@Tower:/mnt# dd if=/dev/zero of=/mnt/disk4/DST/test.dat bs=1024 count=100000
100000+0 records in
100000+0 records out
102400000 bytes (102 MB, 98 MiB) copied, 0.182794 s, 560 MB/s

root@Tower:/mnt# dd if=/dev/zero of=/mnt/cache/DST/test.dat bs=1024 count=100000
100000+0 records in
100000+0 records out
102400000 bytes (102 MB, 98 MiB) copied, 0.442347 s, 231 MB/s

root@Tower:/mnt# dd if=/dev/zero of=/mnt/user/DST/test.dat bs=1024 count=100000
100000+0 records in
100000+0 records out
102400000 bytes (102 MB, 98 MiB) copied, 7.74579 s, 13.2 MB/s

root@Tower:/mnt# dd if=/mnt/disk4/DST/test.dat of=/dev/null bs=1024
100000+0 records in
100000+0 records out
102400000 bytes (102 MB, 98 MiB) copied, 0.0804936 s, 1.3 GB/s

root@Tower:/mnt# dd if=/mnt/cache/DST/test.dat of=/dev/null bs=1024
100000+0 records in
100000+0 records out
102400000 bytes (102 MB, 98 MiB) copied, 0.0693544 s, 1.5 GB/s

root@Tower:/mnt# dd if=/mnt/user/DST/test.dat of=/dev/null bs=1024
100000+0 records in
100000+0 records out
102400000 bytes (102 MB, 98 MiB) copied, 2.76295 s, 37.1 MB/s

I have an old KINGSTONE savage SSD as cache and form these test it looks that it has slower write speed than a hdd, i can live with that, but the /mnt/user speed is extremely slow. I saw that there are other topics that complain that they have slow speeds in /mnt/user but it was at most 50% slower not like this.

 

Can anyone help with this?

Edited January 29, 20206 yr by vik2or

Here is the diag data

tower-diagnostics-20200129-1733.zip

Your test is heavily flawed.

• /mnt/disk4/DST/test.dat and /mnt/cache/DST/test.dat are duplicated files under /mnt/user/DST/test.dat. You then do an additional write test to /mnt/user/DST/test.dat. You just introduced an unrealistic variable to the test.
• You haven't made any attempt to remove the effect of RAM cache from the test. There is no SATA SSD / HDD that is capable of 1.3GB/s and 1.4GB/s read speed.
• Block size of 1k (1024) is equivalent to random IO. 100k of 1k files write to a parity-protected array at 13.2MB/s is completely NOT unreasonable.

 

Let's simplify the test.

Take 3 large (multiple GB) files, copy them using cp command and then watch the speed reported on the Unraid Main page of the GUI.

Ok, copying 2Gb+ files to hdd i get around 55MB/s and to ssd cache, 230MB/s and copying to /mnt/user i get similar numbers (if is no cache or cache preferd ).

So i was just fooling myself into thinking i had a problem but i was using dd incorrectly.