ZFS Slowness in iostats %util

[arkayde]

Hello everyone, next I have a problem and I imagine that it is the SSD that are weak, but for now if someone can help me solve it, I would appreciate it.
I have a ZFS with two SSD
crucial bx500
One mirroring the other. On my server I have 112GB of memory and I left 40gb just for ZFS.
Everything was working fine, there are 3 virtual machines in that pool. Being two Linux and one Windows.
From a few days ago, everything got extremely slow, the bank, the web, and Widows. After turning off the windows machine, everything goes back to normal with the system that uses the bank.
I used the iostats tool and noticed that the two SSDs that make ZFS the %util option reaches almost 100 and after a while it goes back to 2%.
The one that is at 100% is when everything is slow, I believe it's the SSDs that weren't made for that, but if anyone has something I can test, or something I can fix for now I'd be very grateful.

Just out of curiosity, after I run this command on the Windows machine:
Optimize-Volume -DriveLetter C -ReTrim -Verbose

The %util that was at 100% went to 2% but it stayed like that for a couple of hours and then everything came back again.

 

[Dunuin]

crucial bx500

The newer ones are QLC SSDs, so terrible write performance. And ideally you want a SSD with power-loss protection as otherwise the SSD can't cache sync writes (and ZFS is doing a lot of sync writes).

 

[arkayde]

The newer ones are QLC SSDs, so terrible write performance. And ideally you want a SSD with power-loss protection as otherwise the SSD can't cache sync writes (and ZFS is doing a lot of sync writes).

Is the only solution to replace the SSD? Is there nothing you can do right now? From what I'm watching with the windows vm turned off, everything is working fine, is there anything I can do on the vm to improve? This type scsi single

 

[Dunuin]

A slow SSD is a slow SSD, no matter how much you try to improve performance in software. You might be able to get a bit more performance by disabling atime or forbidding sync writes (wouldn't recommend that) or leaving most of the SSD unused, but it still will be a slow SSD.
With SSDs you get what you pay for.

PVE ZFS Benchmark paper FAQ page 8:

Q: Can I use consumer or pro-sumer SSDs, as these are much cheaper than enterprise-class SSDs?
A: No. Never. These SSDs wont provide the required performance, reliability or endurance. See the fio results from before and/or run your own fio tests.

Your best bet would be to use HW raid or mdraid instead of ZFS in case you are not willing to get new SSDs.

 

[arkayde]

A slow SSD is a slow SSD, no matter how much you try to improve performance in software. You might be able to get a bit more performance by disabling atime or forbidding sync writes (wouldn't recommend that) or leaving most of the SSD unused, but it still will be a slow SSD.
With SSDs you get what you pay for.

PVE ZFS Benchmark paper FAQ page 8:


Your best bet would be to use HW raid or mdraid instead of ZFS in case you are not willing to get new SSDs.

Well then, the solution would be to change the SSDs, I need 2 of 2TB to do the ZFS, do you think this one here is of good quality for servers?
Kingston SEDC500M/1920G 1.92TB​