Hardware RAID6, or ZFS RAID-Z2 on JBOD?

[JPorter]

I'm getting ready to build a SAS/6G box for expandable network storage for my company, and trying to decide between traditional hardware raid (Adaptec 6805) and advanced software raid using ZFS on a fast SAS 6g HBA.

FreeBSD, OpenIndiana, and Nexentastor are all good options for ZFS... but what is the real-world reliability of current ZFS implementations in practice?

Any opinions?

 

[Hydaral]

Are you looking for actual case studies? What is your measurement for reliability? A quick search came up with this:

http://www.cs.wisc.edu/wind/Publications/zfs-corruption-fast10.pdf

Reliability would depend heavily on the hardware you are using.

 

[eon]

I would also look into striped mirror RAID-10 for better ZFS, IOPs performance compared to RAID-Z2. Yes, the performance comes at the cost of burning disks.

I'm also a bit biased, but I would go with ZFS.
-If your hardware dies the pool can be taken to a different machine with different controller.
-Any ZFS OS (with equal or higher pool and filesystem version) will be able to access your pool. So if the OS disk dies it's not that big a deal either
-You get enterprise features like snapshots, writable clones and in-band compression. You also get Deduplication!!!
-Clone and transfer datasets on the fly as fast as your network permits.

Hope that helps

"minutes to build, seconds to boot, a lifetime to store" - EON
http://eonstorage.blogspot.com/
http://sites.google.com/site/eonstorage/

 

[JPorter]

The reason that I ask is that several commenters on StorageMojo claim to have lost entire arrays in ZFS, unrecoverably, on more than one occasion. I'm trying to figure out whether they are real people or paid FUD shills.

Comments 18, 19, 20 by "Dave" and "John" on this article: http://storagemojo.com/2009/05/20/btrfs-vs-zfs-omg/

 

[00Roush]

It is certainly possible to lose a ZFS array, especially if you are just doing a JBOD array like John was. But from my point of view the chances of losing a ZFS array are probably about the same or less than most other filesystems. Realize that any RAID 0 or JBOD arrays on any OS or filesystem is prone to failure. That is why there is RAID 1, 5, 6, and in the case of ZFS RAID-Z3 (triple parity). One of the nice benefits of ZFS is it actually does error checking each time data is accessed and verifies the data is what it should be. Whereas most filesystems just blindly read or write from/to the disk without actually checking if the data is correct.

Overall I would say ZFS can be considered just as reliable as any other filesystem. But I am by no means an expert on filesystems.

00Roush

 

[Hydaral]

There isn't much detail to "John's" comment on that site so we don't know whether he tried an import to try to get the data back or what actually wen't wrong, maybe it was a mainboard controller or expansion card problem.

Also, JBOD, with hotspares??

 

[JPorter]

The thing that blew my mind is that these guys are claiming to have lost arrays entirely without any hardware failure. That's precisely what ZFS is supposed to be deeply resistant to, with its integral checksumming. RAID-Z, even more so.

Has anyone else ever heard of "mysterious" total data loss in ZFS in this way? It sounds implausible.