Raidz-1 parity disks?

Question

I am a bit confused about how to configure my zfs pool. What is the exact difference between adding 8 disks to a raidz1 pool or adding 2x4 to the pool with 2 raidz1 groups. Example

zpool create testPool raidz1 d1 d2 d3 d4 raidz1 d5 d6 d7 d8

zpool status shows the same amount of usable disk space but when running zfs list testPool the one with the 2 raidz groups shows 1/2 of the usable diskspace of the one with all disks in one group.

Does this method define 2 parity disks? How are the number of parity disks defined? If I create the following pool with raidz2 it shows pretty much the same usable space.

zpool create testPool raidz2 d1 d2 d3 d4 d5 d6 d7 d8

Any hints on this?

Thanks a lot

Answer 1

If you're familiar with traditional RAID, basically you're comparing RAID6 (double parity) to RAID 50 (striped raid5 sets). The 8 disk RAID-z2 setup is probably the better bet as then any two of the 8 disks to fail without losing all your data, while with a striped pair of 4disk raidz1 sets can only tolerate one disk failure per set (failure of two disks in the same set and you loose everything).

The 2x4disk RAIDz1 sets may offer increased IOPS in some situations as smaller reads/writes might be serviced by one half or the other, whereas every IO hits every disk in the 8disk raidz2 setup, but if performance is your primary concern you should definitely think about mirroring your disks instead (ala RAID 10).

Answer 2

Your first method defines two "virtual devices", with each one having its own parity disk. The second method creates one virtual device two parity disks. Both will have the same amount of usable space, and if there's no controller bottlenecks anywhere then I'd guess they'd have the same performance.

Answer 3

In the both examples you give 2 disks for the redundancy. So the usable disk space will be the equal.

A first one creates raidz striping (actually raid50), it's more fast because the systems tries to balance a load between first and a second group of disks. Each group has a redundancy of 1 device, so if you have a failure on the 2 drives and both are from the different groups, you'll not lose your data. If 2 drives fail and they both are from the same group, your data will be corrupted.

A second one creates actually raid6. It's less fast but you'll keep your data if ANY 2 drives fail.