2

I just got 3 one terabyte HDDs to put into my small NAS box.

I plan on raiding these together with RAID-Z (ZFS, if I'm not mistaken). I've done a lot of research, but there doesn't seem to be very much good information for my questions.
So here they are:

  • How easy is it to expand/contract the number of drives?
    I am currently adding two drives, but am planning on adding more in the future (also getting a sata card for more ports), how easy is this todo?
  • Does it hurt performance to expand/contract multiple times?
  • What Read/Write speed should I expect? I currently have a one gigabit home network humming along smoothly (so I'm getting 130 MB across), will read/write speeds easily keep up?
  • How stable is this on Ubuntu (11.04 x64, server)? It seems like a bit of a hack to get it on ubuntu (though it looks easy), would this be a stable solution?
  • How close do I need to pay attention to the dev community? I know many open source projects I use I need to have one ear on the dev community forums, another in the mailing list, etc. Would this product be similar. I'm really hoping for a setup and forget solution, but would this really be the case?
  • What is the chance I will loose my data because of a malfunction with this software? (I can understand if drives die, just looking at this software's possibility of fubaring my data)
Oliver Salzburg
  • 4,635
  • 17
  • 55
  • 82
Jess
  • 389
  • 1
  • 6
  • 16

2 Answers2

3

About RAID-Z on Ubuntu: You will likely get relatively poor performance with ZFS on any Linux, as there are currently only userspace implementations available for Linux. There where reports about a working kernel mode port last year from India, but this project apparently vanished.

I would consider ZFS on Linux highly experimental and don't trust any of my data on it. This is unfortunate, as ZFS is a really nice filesystem, but I wouldn't bet a cent on Oracle changing the ZFS license so it can be ported without problems.

Sven
  • 98,649
  • 14
  • 180
  • 226
  • Thank you, that was what I was afraid of, now back to square one on finding a solution. – Jess Jul 08 '11 at 20:36
  • Well, Linux md raid is pretty much a standard solution, and with modern file systems like ext4 or XFS you should do really well with only a few TB of data. Do you have any specific requirements that are not met by these technologies? – Sven Jul 08 '11 at 20:39
  • @mazzzzz Why not just use md/lvm for software raid like almost everyone else on Linux? It's tried and true. – MDMarra Jul 08 '11 at 20:41
  • This is the first raid I am building (for a local file server), so any all advice is welcome. The main problem was finding a good system for expanding the raid (with 1tb hds), and allowing for a disk failure (would like to be able to account for two, but that's a lot of storage space to loose). – Jess Jul 08 '11 at 21:11
  • I'd probably just use OpenSolaris (OpenIndiana) – Tom O'Connor Jul 08 '11 at 21:24
  • The problem is that I locked myself into ubuntu when I configured everything (mail server, user systems, file download programs), I would hate to reconfigure all of this, the only thing I have left is the RAID (because I didn't have the drives when I started this project). – Jess Jul 08 '11 at 21:43
  • You will find tons of tutorials about using md RAID and LVM via Google and many questions already asked here with great answers. If you have any specific questions after that, please feel free to ask them here. – Sven Jul 08 '11 at 21:59
  • Although I haven't tested it, there is a native Linux port (beta) of ZFS available here http://zfsonlinux.org/ – jlliagre Jul 08 '11 at 22:23
  • @jiliagre: This is what I would consider experimental. They haven't even a POSIX layer yet in a released version, so I wouldn't put it into production. Hopefully, it's important enough for LLNL that they continue working on it. – Sven Jul 08 '11 at 22:42
  • Yes. I know and already wrote the native port is a beta release. I was just commenting about your statements "there are currently only userspace implementations available for Linux" and "There where reports about a working kernel mode port last year from India, but this project apparently vanished". The QT Infotech release was based on the LLNL port. – jlliagre Jul 08 '11 at 23:56
3
* How easy is it to expand/contract the number of drives?

You can't add or remove drives from a RAIDZ unless if you do a full backup/restore.

* I am currently adding two drives

RAIDZ requires at least three disks (or partitions), two disks implies either mirroring or a non redundant configuration.

* Does it hurt performance to expand/contract multiple times?

Not applicable.

* What is the chance I will loose my data because of a malfunction with this software? (I can understand if drives die, just looking at this software's possibility of fubaring my data)

If a single RAIDZ drive die, you loose nothing. ZFS is pretty solid by design. I'm using it on Solaris (native) but also on Ubuntu (fuse) and never had any issue. I didn't test yet the native port available at zfsonlinux.org.

jlliagre
  • 8,861
  • 18
  • 36
  • The backup/restore for a size adjustment is a deal breaker (even though zfs offers a nice snapshot system) for me, considering I plan to expand size indefinitely. – Jess Jul 08 '11 at 23:22
  • You certainly can expand a ZFS pool indefinitely, the restriction I'm talking about is a RAIDZ one, not a pool one. – jlliagre Jul 08 '11 at 23:32
  • Hm, if I read correctly, a pool is a group of drives, and you can expand a pool indefinitely? – Jess Jul 09 '11 at 00:12
  • You can expand a pool indefinitely. A pool is a group of vdevs. A vdev can be a group of one or more files, partitions, whole disks, or combination of them. Then, to "expand you raidz", just add another raidz to its pool and that's it. Alternatively, you can also replace one after the other the disks used by a raidz by larger ones and finally, the pool size will grow. – jlliagre Jul 09 '11 at 00:19