Proxmox file transfer performance

[patrick.pesegodinski]

Hello.

The throughput on the Windows Server 2022 VM is not in line with CrystalDisk tests on the same VM.

Could it be a Windows limitation or some configuration error?

- Hardware: DELL R410 Server with Intel 4520 SSD for PROXMOX in ZFS. ZFS pool in RAIDZ1 with 04 960GB KINGSTON DC600M SSD.

- VM configuration:


- Tests:

- Copy:



- CrystalDisk:

 

[_gabriel]

CrystalDiskMark with 1GB size only test your cache, (first small Windows cache itself) then ZFS ARC cache...

 

[patrick.pesegodinski]

CrystalDiskMark with 1GB size only test your cache, (first small Windows cache itself) then ZFS ARC cache...

Isn't the speed number real then?

 

[_gabriel]

ZFS will indeed tank performance, because of its write amplication ( and checksum , integrity ....).
+ ZFS RAID-Z1 (it's not RAID5 ...) isn't the best choice for performance,
using ZFS striped mirror (like RAID10) is recommanded for VMs storage, it should surely double your Windows copy file test ...

 

[patrick.pesegodinski]

ZFS will indeed tank performance, because of its write amplication ( and checksum , integrity ....).
+ ZFS RAID-Z1 (it's not RAID5 ...) isn't the best choice for performance,
using ZFS striped mirror (like RAID10) is recommanded for VMs storage, it should surely double your Windows copy file test ...

This RAID10?

 

[_gabriel]

indeed, I didn't remember PVE call it directly RAID10.

 

[patrick.pesegodinski]

indeed, I didn't remember PVE call it directly RAID10.

I created RAID10 and the transfer rate and the test on Crystal showed the same results.

 

[_gabriel]

try disabling vdisk writecache
what about speed with an Ubuntu VM ?

EDIT: if it's vanilla pve 8.2, try upgrade.
try kernel 6.5 too.

 

[patrick.pesegodinski]

try disabling vdisk writecache
what about speed with an Ubuntu VM ?

EDIT: if it's vanilla pve 8.2, try upgrade.
try kernel 6.5 too.

I will test with Ubuntu VM.

Proxmox is 8.2.4 version and kernel is 6.8.8.3.

 

[_gabriel]

don't forget to reset to Default None write cache in vDisk of the VM , as ZFS already write cache, double write cache can be worst.