Verify jobs - Terrible IO performance

[Dunuin]

blocksize is the same thing as recordsize???

No, recordsize is used for datasets. Volblocksize is used for zvol. When you set the "Block size" for a ZFS Storage in the PVE webUI it will only define what volblocksize will be used for new zvols.

After creating the pool I will

zfs set special_small_blocks=4K zfs
zfs set compression=lz4 zfs
zfs create zfs/PBS
zfs set recordsize=1M zfs/PBS

Sounds ok for PBS. But I would also use zfs set relatime=on zfs so not every read will cause an additional write.

then create the following I will mount for VM data

zfs create zfs/VMDATA
zfs set recordsize=16K zfs/VMDATA

That won't help. VMs are using zvols and zvols won't use recordsize. And the volblocksize isn't inherited, you will have to defne it for each virtual disk you create. And as usually PVE is creating the virtual disks for you, you should set the "Block Size" for the ZFS Storage in PVE to 16K, so PVE will create new virtual disks with a volblocksize of 16K.

do I use wwn-0x5000cca28f6d86a0 or scsi-35000cca28f6d86a0

Both will work. What you prefer.

 

[mow]

But I would also use zfs set relatime=on zfs so not every read will cause an additional write.

Isn't relatime the default?

 

[Dunuin]

I don't know if they changed it meanwhile, but the last few times I installed PVE I quite sure it wasn't.

 

[mow]

I don't know if they changed it meanwhile, but the last few times I installed PVE I quite sure it wasn't.

You're right, it's off by default. Something that really should be changed, considering that for ext4 on linux, the default is to use relatime since about 20 years ...