RAIDZ level for backup server?

[Kevin Horton]

I am planning the vdev scheme for a remote backup server. My current main system will be repurposed to be the backup server, and I will build a new, more capable, main system. The intended backup system can support 8 disks, which will be 4TB WD Reds. Normally, I'd be happy with RAIDZ2, but given that it is offsite, and I spend quite a bit of time traveling on business, I cannot be sure to be able to quickly replace a failed disk. The system may have to run with a failed disk for three or more weeks before I could replace it. I am not happy with the potential single disk redundancy for a long period.

I am looking at two options to assure a robust, reliable system even after one disk has failed:

8 disks in RAIDZ3, or
7 disks in RAIDZ2 + one spare disk. I will have admin access via VPN, and could remotely replace a failed disk with the spare disk if required.

I have no experience with RAIDZ3. If a disk fails, does it work about as well as a RAIDZ2 system? Are there any downsides to running a RAIDZ3 system for a long period with one failed disk?

I am leaning towards RAIDZ3, but would appreciate any comments.

Of course, perhaps I should keep in mind that this is a backup server, and the world won't end if it fails, as all the data already exists in one or two additional places. Maybe I should use 8 disks in RAIDZ2, to allow more room for growth in the size of the backups. I've thrown out any thoughts of RAIDZ1, as the network connection at the backup site is not fast, and it would take a very long period to copy the data back on to the server if I ever lose the backup pool. I would have to bring the system back to the main site for a few days to reload the data.

 

[MatthewSteinhoff]

I wouldn't hesitate to run RAIDZ2 on a secondary server even if I couldn't get to it for a month. For data on the secondary server to be needed and unavailable the following would need to happen...

1. Primary server loses first tier of redundancy (RAIDZ2 -> RAIDZ1, effectively).
2. Primary server loses second tier of redundancy (RAIDZ1 -> Running on Empty) .
3. Primary server loses data before redundancy restored (RoE -> DEAD).
4. Backup server loses first tier of redundancy (RAIDZ2 -> RAIDZ1).
5. Backup server loses second tier of redundancy (RAIDZ1 -> RoE).
6. Backup server loses data before redundancy restored (RoE -> DEAD).

I'm a pessimist by nature and I don't see that happening. If you began to have trouble with the primary, I suspect you'd fix it before you lost the data. If something catastrophic happened at the primary site, you'd be at the backup sight relocating the data to the primary site and could manage any redundancy losses on the secondary quickly.

Cheers,
Matt

 

[Spearfoot]

I concur with @MatthewSteinhoff that RAIDZ2 will suit your purposes just fine.

 

[Kevin Horton]

I agree that it is extremely unlikely that failure of the primary server + failure of the backup server would occur in the same period. I look at the off-site backup server more as the primary mitigation against other major events such as fire, theft, etc. But, those events are relatively rare as well, so perhaps RAIDZ2 is the right answer. I'm currently leaning towards 8 disks in RAIDZ2. When I out grow that, I'll decide probably replace the whole server rather than do the one disk at a time upgrade dance.

Thanks for your thoughts guys.

 

[depasseg]

If it makes you feel any better, my offsite is 6 disk * 6TB RAIDZ1. I also have an onsite backup though.