Extremely Slow SSD Write Speed in VM

[Dunuin]

Im aware and appreciate the recommendations but thats very overkill and expensive for my requirements. When you say spend a little more and get proper SSD's i can only see prices 3x that of what i mentioned for roughly the same size.

PM883 960GB is 74€.
870 EVO 1TB is 69€.
870 QVO 1TB is 57€.

Its never a good idea to buy any QLC SSD, even if you save 12€ per TB. And no point to buy a consumer TLC SSD if you can get a proper Enterprise grade TLC SSD for just 5€ more.

 

[BobhWasatch]

Host

root@proxmox-02:~# df /dev/SSD2TB Filesystem 1K-blocks Used Available Use% Mounted on udev 8122664 0 8122664 0% /dev

VM

docker@docker-02:~$ df /mnt/SSD2TB Filesystem 1K-blocks Used Available Use% Mounted on /dev/sdc 1856686824 236034864 1526263736 14% /mnt/SSD2TB

Wait a minute! This says you are running the test on the host in a udev filesystem, which is a ram-based filesystem not a real disk. Which would explain the unbelievable speeds (over 300K IOPS, 1.3 GB/s) you are seeing there and also why direct I/O isn't supported. You are basically testing the speed of the hosts RAM + udev filesystem overhead. Did you perhaps forget to actually mount the disk before testing?

Also, why do you test only read speed when your issue was about writing? Maybe fix your setup and then come back to argue the benefits of QLC.

ETA: FWIW I am using 2x Samsung EVO on an LVM setup and have not seen any particular performance issues. It is a lightly-loaded machine though.

 

[LnxBil]

And you also always try to benchmark "/dev/SSD2TB/test1.img". In the "/dev" folder you find block devices but you are trying to write to a file.

Wait a minute! This says you are running the test on the host in a udev filesystem,

That's why I asked in the first place.

 

[LnxBil]

PM883 960GB is 74€.

I bought 20x 960 GB SM863 (used of course) for 31€ each and in smaller batches for 39€ each.

 

[daNutz]

PM883 960GB is 74€.
870 EVO 1TB is 69€.
870 QVO 1TB is 57€.

Its never a good idea to buy any QLC SSD, even if you save 12€ per TB. And no point to buy a consumer TLC SSD if you can get a proper Enterprise grade TLC SSD for just 5€ more.

Nice prices, I’ll do some searching but prices are still more for me in the UK.

 

[daNutz]

I bought 20x 960 GB SM863 (used of course) for 31€ each and in smaller batches for 39€ each.

Nice! EBay?

 

[daNutz]

Wait a minute! This says you are running the test on the host in a udev filesystem, which is a ram-based filesystem not a real disk. Which would explain the unbelievable speeds (over 300K IOPS, 1.3 GB/s) you are seeing there and also why direct I/O isn't supported. You are basically testing the speed of the hosts RAM + udev filesystem overhead. Did you perhaps forget to actually mount the disk before testing?

Also, why do you test only read speed when your issue was about writing? Maybe fix your setup and then come back to argue the benefits of QLC.

ETA: FWIW I am using 2x Samsung EVO on an LVM setup and have not seen any particular performance issues. It is a lightly-loaded machine though.

Can you or @Dunuin or @LnxBil provide the correct parameters to test the drive correctly?

 

[Dunuin]

Nice prices, I’ll do some searching but prices are still more for me in the UK.

On amazon.co.uk it's 75 GBP for the PM883 960GB and 70 GBP for the 870 EVO 1TB. So still no big difference in price.

 

[Dunuin]

Can you or @Dunuin or @LnxBil provide the correct parameters to test the drive correctly?

There isn't a simple command as write benchmarks are destructive when done on the block level. And for doing them on file level you will have to provide a partition and filesystem, mount it (and make sure no VM with passthrough is running) and then do some sync write benchmarks. Depending on your filesystem you might also need to disable stuff like caching for benchmarking the SSDs read performance.
Without knowing your system setup it's hard to tell what to do.