NAS storage: "bare-metal" or virtual disks?

[giovanesaggio]

Hi everyone,
just a simple question: let's assume you have your shining server with an array of disks (zfs raid, btrfs, mdraid, hardware raid, whatever you like) and you need to setup a NAS-like service. Would you rather pass a folder to a container (in charge of sharing the data over the network using SMB/NFS/whatever) to let it write directly to the disk or would you create a virtual disk and mount it in the container? What are the pros and the cons?

Thanks everyone, have a nice day!
Luca

 

[LnxBil]

With ZFS, I'd go with LX(C) containers and use bind-mounting to get the dedicated nas dataset(s) inside (plenty of people using it like that). The term "virtual disk" is a bit stretched here, because in ZFS it's just a dataset (file-storage) but it's the way to go. It does not matter if you create the disk inside of PVE or outside and bind mount it, the underlying technique is the same - bindmount a zfs dataset. Each dataset should also be the root of a samba share, so that the snapshots can be used for Shadow Copy.

If you would want to use a special NAS distribution like True/Free NAS, OMV, <yournamehere> ... then the answer varies:
People using passthrough for the disks, whereas I would not do such a thing. Your PVE is a virtualization host, let's it virtualize stuff and just use those virtualized stuff inside of the guests. If your NAS solution is based on Linux, it may run inside of a LX(C) container, which is good.

Passthrough has in general the problem that you need disks for PVE itself and for the passthrough, whereas if you virtualize everything, you only need disks for all, which is assuming you have the same number of disks available in both scenarios always faster and you have more space available ... and easier to setup and keep track of.