What if, instead of fiddling with the many ZFS topologies to ensure data resilience and all that jazz, I define two independent pools of comparable size, both stripe, I designate one of them as “backup” and the other one as “main,” and have a script back up data automatically once a week from the main pool to the backup pool? Does that look like something a responsible NAS designer might do?
No.
4 Likes
“data resilience and all that jazz”
-ResponsibleNAS
Why would you choose TrueNAS or ZFS in the first place? It assumes that you prioritize data integrity more than the casual user.
Why a script? Why not use the Replication Tasks, which leverages ZFS and snapshots?
1 Like
How would you guys structure the following disks: 4x4T M.2 NVMe + 2x8T M.2 NVMe? This will all be used for data (photos, videos, documents) storage–meaning that I already have one boot + apps pool. Ideally I’d like to have all this in one pool, each drive mirroring one drive of the same size, hence 16T overall.
The answer is in your question: A stripe of three 2-way mirrors, 2*4T, 2*4T, 2*8T.
To expand on @etorix’s answer…
You can only have a mirror of the 2x8TB NVMes, and since you should not mix vDev types in a pools data area, that means mirrors for everything.
But if you keep the 8TB NVMEs for another use, and buy a couple of matching 4TB NVMe cards you could do 6x4TB RAIDZ2 for a total of 16TB too.
1 Like
I’d be reasonably satisfied with raidz1 using SSDs, so a third 8TB for 16 TB from a 3-wide raidz1 is an option. 5*4 TB in raidz1 is an option.
But with exactly the described hardware, striped mirrors is the only reasonable solution.
Sorry - I should have explained even further that my comment about mirrors was an explanation for your design based on the specific hardware OP has.
One pool is not a backup strategy.
Maybe 4x4T in Z1 or Z2 in one system, replicating daily or weekly to one or two other systems in some other place with the 2 8Ts.
1 Like
personally, 1 backup locally & 1 backup off-site. Not just backup N forget it until the day you need but to test the backup every so often (meaning restoring it and confirm it good) - back in the day - I used par as part of the backup plan - just in case the backups got corrupted.
Alright, so I did the responsible thing and settled on the solution proposed by @etorix, namely a stripe of three 2-way mirrors, 2x4T, 2x4T, 2x8T. I take it that gives me better data resilience than any of the RaidZ solutions (at the expense of disk space), right? Can I pat myself on the back now?
Reasonably so.
The stripe of mirrors is not necessarily more resilient than raidz2/3 but throwing all drives in a raidz array would restrict the 8 TB to 4 TB and be less flexible for future evolutions.
1 Like