Proxmox VE ZFS Benchmark with NVMe

[koalillo]

Opened https://bugzilla.proxmox.com/show_bug.cgi?id=4385 .

 

[Dunuin]

The benchmarks were only using block devices, so recordsize isn't used at all, as recordsize is only for datasets on file level. For block devices (physical SSDs or Zvols (virtual disks of VMs)) the volblocksize is used instead of the recordisze.

 

[Curt Hall]

To optimize performance in hyper-converged deployments with Proxmox VE and ZFS storage, the appropriate hardware setup is essential. This benchmark presents a possible setup and its resulting performance, with the intention of supporting Proxmox users in making better decisions.

Download PDF
https://www.proxmox.com/en/downloads/item/proxmox-ve-zfs-benchmark-2020
__________________
Best regards,

Martin Maurer
Proxmox VE project leader

I would like to test my existing sytem (has data already), so I need a NONdestructive (read only) test. The tests lists in the article look like read/write and destroy's existing data.

 

[Dunuin]

I would like to test my existing sytem (has data already), so I need a NONdestructive (read only) test. The tests lists in the article look like read/write and destroy's existing data.

Page 11 "4.1 – fio command for Proxmox VE node" is using a empty zvol. So yes, it is destructive but thats why you create a new zvol first that you can delete after the test.

Page 11 "4.2 - fio command for “VM Performance (Windows)”" and "4.3 - fio command for “VM Performance (Linux)”" are for running inside a VM Vm. So you could create a new VM for testing and destroy it later.

But the test on page 3 that is writing directly to the SSD is indeed destructive. (And great to see the warnign there...often this is missing and people type in code without understanding what they are doing and then wiping their disks...)