3-way mirror ?

[anankeo]

For added availability I'm considering building a RAID0 pool on top of several 3 way-RAID1 vdevs built with 4TB hard drives.
However I'm wondering :
- Will the read performances be better compared to a 2-way mirror ?
- What if a second drive fails in the same vdev during the reconstruction ? Does ZFS handle this transparently (ie uses the remaining healthy drive for reconstruction without downtime) ?
- How does ZFS reconstruct a failed raid1 vdev if it has 2 available mirror drives ? Will it use both drives, reducing by half the access on the mirrored drives, or will it use only one drive ?
Thanks ;)

 

[Arwen]

For added availability I'm considering building a RAID0 pool on top of several 3 way-RAID1 vdevs built with 4TB hard drives.
However I'm wondering :
- Will the read performances be better compared to a 2-way mirror ?
...

Yes, barring other bottlenecks like network or PCIe lanes to the disk controller chip.

...
- What if a second drive fails in the same vdev during the reconstruction ? Does ZFS handle this transparently (ie uses the remaining healthy drive for reconstruction without downtime) ?
...

Yes.

...
- How does ZFS reconstruct a failed raid1 vdev if it has 2 available mirror drives ? Will it use both drives, reducing by half the access on the mirrored drives, or will it use only one drive ?
Thanks ;)

Sorry, don't know.

 

[Robert Trevellyan]

I'm considering building a RAID0 pool on top of several 3 way-RAID1 vdevs

You can't do nested RAID levels with ZFS. Pools are made of vdevs, and vdevs are made of physical drives. Redundancy is present at the vdev level.

EDIT: now I see you're talking about striped vdevs. OK, that's how ZFS works, carry on ...

(but using proper ZFS terminology would help to avoid confusion)

 

[anankeo]

Yes, barring other bottlenecks like network or PCIe lanes to the disk controller chip.

Yes.

Sorry, don't know.

Thanks Arwen, I guess this is something I can try and see if both drives are working during the reconstruction.

 

[anankeo]

You can't do nested RAID levels with ZFS. Pools are made of vdevs, and vdevs are made of physical drives. Redundancy is present at the vdev level.

EDIT: now I see you're talking about striped vdevs. OK, that's how ZFS works, carry on ...

(but using proper ZFS terminology would help to avoid confusion)

Hehe sorry for the confusion