zfs setup question

Question

Currently I have a linux storage box and server with 4x750gb harddrives in raid-5 with ext3. I have ordered 3x1.5tb disks to upgrade this. Here is my planned upgrade:

Backup:

1. Format the 1.5 tb disks
2. Copy all data from the raid-5 disks to the 1.5tb disks
3. Destroy the raid-5 array.

New setup:

1. Create a VirtualBox system and install Nexenta (OpenSolaris + ubuntu) on it.
2. Create a zfs pool with zraid1 with the 4 750gb disks.
3. Copy from 1.5tb disks to the virtualbox zfs pool
4. Format the 1.5tb disks.
5. Replace 3 off the 750gb disks with 1.5tb disks.
6. Reuse the 750gb disks elsewhere.

The reason I wish to use one 750gb disk is since I can't grow the disk count in a raidz array, and this gives me the option off replacing that disk later for an extra 750gb storage.

Would the ZFS performance be good running through virtualbox? Or will the performance overhead be too large? Will I get 1.5tb+1.5tb+750gb storage on the zraid? Or just 750gbx3 until all disks are 1.5tb?

Answer 1

This seems like a very inconvenient way to do it.

• Use ZFS on FUSE rather than running it through a virtual machine, the performance will not be good in any case, but I'd definately try not to have the file system inside a virtual machine
• AFAIK you can actually increase the number of disks in a running raid using zfs
• Using three 1.5TB disks and one 750GB disk in an array would limit the usable sizes of all disks to the smallest one. There is some research being done on having raid over different disk sizes, but it's not usable now

I would rather create a new raid array of the 1.5TB disks (either md raid or raidz via fuse) in linux, and expand it with single disks as the need arises.

Answer 2

You can't grow the disc count, that is correct, but if you have 3x1.5tb and one 750gb drive you will waste 3x750 gb. Your pool will look like it contains 4x750 gigs. Once you replace that last drive, you will be able to get 4x1.5tb.

I have ran zfs in a virtual machine as well (solaris & opensolaris) but the performance is not good. Maybe it can be better but at the end of the day, file IO is the weakest link of a virtual system.

Take into consideration that zfs is very memory hungry as well (certain features are turned off by default if you have less than 2 gigs of ram) and it is optimized for 64-bit systems.

I suggest you install your zfs-capable OS on the actual hardware, and use virtualbox to virtualize your current linux setup if you still need that for something.