What is the reason behind low performance of ssd drives?

Question

linux

The server with the database (mysql) was running out of space, so it was decided to move the database to a new server with more capacious disks.
Server A configuration: 2 ssds of 240 GB, software raid 1
https://www.hetzner.de/de/hosting/produkte_rootser...
pastebin.com/GVBu1eLJ
When migrating the database to another server (B), I encountered poor database performance .
Server B configuration: 2 500 GB ssds, software raid 1:
https://www.hetzner.de/de/hosting/produkte_rootser...
Suspicion falls on ssh disks.
After the failure with the new server, it was decided to take another server (C) on nvme disks, but the result was the same.
Server configuration C: 2 ssd 512 GB, software raid 1:
https://www.hetzner.de/de/hosting/produkte_rootser...
pastebin.com/eSLMtgnF
All benchmarks below are for server with nvme (C) since server B is no longer available, however it showed similar results across all benchmarks.
Benchmarks
wa: i.imgur.com/wJCP4eL.png (purple - new server (C) which currently works only as a slave from the master, blue - server A)
A: pastebin.com/5LfzCqUu
C: pastebin.com/JSJiGqhN
According to userbenchmark, server C should outperform A in all respects.
ssd.userbenchmark.com/SpeedTest/3641/INTEL-SSDSC2CW240A3
ssd.userbenchmark.com/SpeedTest/30300/NVMe-SAMSUNG-MZVKV512
What could be the problem? OS settings or is it in the disks themselves?

Answer 1

this is more of a jamb in OS than in disks, they work awesome under Windows, under linux most likely something is wrong with the drivers

Answer 2

the only place where SSD is inferior to HDD in speed is sequential reading in small blocks, especially at 4k, the difference can be up to 3 times in favor of hard.
"at the request of the sectarians, we beat their sacred cow with a vengeance"

Answer 3

How many did not touch the hetzner's server, they always have problems with disks.
Don't think that they gave you a new server, maybe it's a server that was beaten to death by another client, and your server B was given to another client.
Look at the stats for io, maybe something is loading your disk. In my last case, it was a raid sync that was running 100% of the time (couldn't complete due to large discrepancies), and was eating 99% of the io. As a result, the discs braked like crazy.

Answer 4

Let me guess. You have disks with a 512b physical block on your strobe server. And the new ones (NVMe for sure) come with a 4kb block. Did you do alignment when creating the file system? Try to break handles and normally create partitions. And run tests in the rescue system. You can also create partitions there.

Answer 5

Well, first of all, the first server is not a xeon, but it also has 1 core running faster, so it's all right.
As for slow work, dear, can my.cnf be sent to the studio?