NVME IO Latency 700~800ms, normal?

[nanankcornering]

I currently have a Dell R630 with a PCIE to M.2 Adapter running 990 PRO.

Quite overkill as it is limited by PCIE gen 3 x 4 speeds.

However, I noticed that I/O latency is averaging around 700-800ms, and it would go up to 900ms-1s during high io operations.

Is it normal? Anything I can do to reduce the latency?

IOPS data is collected from node_exporter in host server.

 

[_gabriel]

ZFS ?

 

[nanankcornering]

ZFS ?

no. ext4

 

[_gabriel]

What about VMs ?
Windows with Virtio SCSI single controller ? during benchmark ?

 

[LnxBil]

990 PRO

The general answer on the forum is: buy enterprise hardware if you want to have consistent and fast throughput. It does not make sense to buy server grade hardware an put in customer hardware that was build for a desktop machine and yes, that is everything samsung sells with their evo/pro line.

Besides that, have you checked with the magican if the NVMe is all right? What about a benchmark in another OS or without any workload / VMs running?