[OpenIndiana-discuss] Puzzling behaviour in zpools

[Jim Klimov]

On 2013-03-04 20:54, Jan Owoc wrote:
>> For the 4 TB disks there is only 1 12 disk vdev so 10 data disks, 2 parity disks, not 4.
> Your zpool status showed 2x 6-disk vdevs...

This brings up an interesting question to contemplate: with the same
capacity but doubled performance one can make 2x6-disk raidz1 sets
(possibly even more performance than double due to simpler XOR math).
Smaller disk sets would also use less IOPS and less overall read
bandwidth to recover (resilver) a disk after a fault, though they
would also have less redundancy.

On the other hand, calculations showed that with huge disks (over 3Tb)
it may be worthwhile to use raidz3 just to protect against failures
of additional drives while one is being resilvered - the operation
is bound to take considerable time (maybe days) during which the
pool's defenses are lowered.

https://blogs.oracle.com/ahl/entry/triple_parity_raid_z

Did you think and consider the trade-offs like these? If yes, what
was your educated conclusion and why?

//Jim