FreeNAS vs unRAID vs OpenFiler vs ZFS

[black0ut]

I'm building a new fileserver, and I'm wondering what kind of redundancy is best.

I know this is only high-availability redundancy, I have backups elsewhere.

I currently have 3 1.5TB drives, but that will most certainly grow in the future. I don't want to have to purchase a hardware RAID card, and I dislike software RAID, so normal RAID is out.

This fileserver will be simply serving files over the network, may stream media in the future but that's unlikely. It will also serve as a print server. I just want to make sure I can keep working if a hard drive fails.

Also, I have significant Linux experience and am not afraid of learning something new if it means the end result will be better.

 

[Blue Fox]

One thing to point out. FreeNAS supports ZFS, so you can combine those options.

 

[Yonzie]

ZFS's RAIDZ pwns both driver RAID, other brands of software RAID and hardware RAID.
However, FreeNAS' ZFS implementation is lacking. Look at Nexenta or EON

Regarding expansion:
What you can't do:
* Expand the RAIDZ
What you can do:
* Add a new RAIDZ or single drive or mirrored drives.
* Add the new RAIDZ/drive/mirror to your existing Zpool, thereby seamlessly expanding your file system.

This means that by buying 3 drives at a time, you will have multiple RAIDZ groups of 3 drives, joined into one big file system. You will lose 2/3rds of the total space.

 

[jonnyjl]

This means that by buying 3 drives at a time, you will have multiple RAIDZ groups of 3 drives, joined into one big file system. You will lose 2/3rds of the total space.

I'm confused, why would he lose 2/3rds of total space?

 

[Yonzie]

RAIDZ is like RAID5. You get N-1 times the amount of space on one disk.
Provided the OP adds 3 drives at the start, 2/3rds of the total disk space will be available.
I assumed the OP might add 3 drives at a time, and just add to the zpool instead of starting over.

 

[jonnyjl]

RAIDZ is like RAID5. You get N-1 times the amount of space on one disk.
Provided the OP adds 3 drives at the start, 2/3rds of the total disk space will be available.
I assumed the OP might add 3 drives at a time, and just add to the zpool instead of starting over.

I thought you were saying he loses 2/3rds, when really he loses 1/3rd, if we go with the 3 drive raidz1 example.

Personally I plan on building a OpenSolaris ZFS server and adding raidz1 (4 drives) to the pool at time (building on a Norco 4220 case). Seems like a decent solution.

 

[black0ut]

Yeah, I've been reading up on all these software RAIDs, and the biggest pros and cons for each are:

ZFS Pros:
-Awesome data integrity, resolves silent bit rot, no RAID5 write hole
-Free!
-Backups are super easy (both full and incremental)
ZFS Cons:
-Adding more drives is inefficient, can't add to the same vdev, so every upgrade I have to set another drive as parity-> this means I have as many parity drives as times I upgraded

Unraid Pros:
-Sufficient data integrity, doesn't correct silent bit rot and has a write hole
-Very flexible upgrades, will always have only 1 parity drive, so very space efficient
Cons:
-Costs money, $100ish
-Very slow write (not a real big deal)
-No built in backup (not a big deal with rdiff-backup)
-Not as good zfs for data integrity, see above

It's essentially boiled down to those two technologies, because FreeNAS uses either the unRAID style XORing or the ZFS implementation (but bad), and Openfiler only supports traditional RAID.

ZFS looks really nice, but the whole "You have to upgrade at least 2+ disks at a time" thing is really a dealbreaker. They say they might have vdev expansion in the future, but who knows.... I'll probably go with unRAID and migrate to ZFS once it gets that feature.

Btw, anyone know ZFS's write performance? I've only a few scattered unsubstantiated reports that it's really really slow.

 

[Yonzie]

I thought you were saying he loses 2/3rds, when really he loses 1/3rd, if we go with the 3 drive raidz1 example.

Seems like you are correct. My bad. Sorry.

Btw, anyone know ZFS's write performance? I've only a few scattered unsubstantiated reports that it's really really slow.

"Really slow" would probably be when run on *BSD or similar under FUSE. Apparently you can expect write speeds around 10MB/s and about 40MB/s read... Nexenta, EON and OpenSolaris (duh) all use the OpenSolaris kernel. I don't have anectodal evidence of write speeds for these (I am still speccing out the new (ZFS) file server), but looking at the X4500, it better be fast

 

[OFaceSIG]

I just set up a unRAID server on an old single-core celeron microATX board I had lying around. I slapped in two hitachi ultrastars 1GBs and the mobo into an old case. Figured out the qwerks fairly quickly. Now all the machines in my house have a S: drive mapped to a folder on the unRAID and it's fully redundant. All I had to do was spring for two 50 dollar drives. Can't beat it.