I’ve recently had issues with my 6x4T RAIDZ2 setup. Their net storage is 15T. The drives have some miles (at least age) and may need replacement. Two possible options:
replace all 6 drives with new 4T drives and keep RAIDZ2
replace RAIDZ2 with 2 x 24T mirror array
My system is lightly used and has 1Gb network.
The cost is about the same. Upgrading to 6 x 8T would be about 60% costly.
I have 2 extra SAS/SATA ports so data migration to the mirror would be much easier. It would also reduce power consumption and noise. The retired drives could be used sparingly for backing up other systems.
AIUI, raidz should have better throughput, and mirror should have better (read) iops [1]. If you only have a 1Gbps network then throughput doesn’t really matter, as it can be saturated by a single hdd.
I too would go with Mirrors. With the size of HDDs being larger than many SOHO user’s needs, it make a bit of sense to use Mirrors.
Just keep in mind that for a 2 way Mirror vDev, a single disk lost, takes away all redundancy. So I would have a plan in place to immediately replace any failing disk. Either with a cold spare, or backup HDD as suggested below.
One benefit of Mirrors, especially single vDevs, (though this works with multiple Mirror vDev pools as well), is using split Mirror for backups. Their is a special ZFS command to gracefully detach a device from each Mirror vDev, to create a new pool.
Now using split Mirror backups don’t support incremental re-syncs, so using ZFS Send & Receive locally for single HDD backups might still be preferred. However, in either case, having 1 or 2 same sized HDDs for backups has the added benefit that you could immediately put 1 in place in case of failure in your main pool.
Last, there are some people, (me included), that would prefer to use 2 separate vendors for such a task. Like Seagate and Western Digital, both of which make larger HDDs. This way you might avoid similar life span of the HDDs. Of course, test the drives before putting them in production.
