Reducing Single Points of Failure (SPoF) in Redundant Storage

[Miklos]

I've never seen one before, do you know if it's a common practice? I would think that doing a RAID Z2 of mirrors would be better -- you'd get faster resilver times because you're rebuilding mirrors instead of a whole RAID Z2 in the event of a drive failure.

You loose a lot of space that way, it's more economical to stripe z2 or z3. As long as the disks are the same size I doubt you get longer resilvering times as its the same amount of writes.

[vrod]

On 1/4/2016 at 3:42 AM, wpirobotbuilder said:

I've never seen one before, do you know if it's a common practice? I would think that doing a RAID Z2 of mirrors would be better -- you'd get faster resilver times because you're rebuilding mirrors instead of a whole RAID Z2 in the event of a drive failure.

Looks like you are thinking conventional RAID logic. In ZFS we are dealing with vdevs. If you have 2x6 RAIDZ2 vdevs you can either stripe them or mirror them, however there will be no change in rebuild times because a vdev is its own. I wouldn't make a mirror out if 2 vdevs. You usually make the mirrored, smaller vdevs first, then stripe or mirror them. For example, if you want to make RAID10 or similar, you make several small 2-disk mirror vdevs and then throw them into pool, being the "big vdev". Right now I run a 6x6TB raidz2 and if I needed more space, I would have to make another of the same vdev and then add it to the pool (striping).

[Dlog]

Instead of Raid 60 across multiple controllers why not a redundant controller device with a replicated copy of it's data on another device.

Look at a NetApp cluster of 2x FAS 2552s, The two devices each have two controlelrs (servers) built into each that work together and are cross wired for SAS. If one of the servers were to restart, fail etc the other takes over until it's partner returns. On the second FAS 2552 you could configure a SnapVault to take copys of the data from the first FAS and in the event of a failure on the first simply take over it's job.

 

I'm expaining this poorly I think.

FAS 2552 - 1

[{server 1 with disks}{server 2 with disks}] SAS between all wired together                    

(super important volume one on Virtual interface with ip 10.0.0.10, if Server 1 fails server 2 takes the Virtual interface and access the disks through the shared SAS cabling)

 

2x 10 Gbit High Speed connection each (2x switches interconnected with 4x interconnects, one conection per switch for redundancy) for inter cluster talk (internal data transfer)

 

FAS 2552 -2

[{server with disks}{server with disks} ]SAS between all wired together

(Snap mirror/vault/metro...whatever of data from super important volume on FAS 2552 -1, if FAS 2552-1 fails, FAS2552-2 can provide the version of the data it has with the same IP as the virtual LIF from FAS2552-1) when you fail back you can reverse update the original volume to get data back to current.

 

Again I'm terrible at explaining so here.

 

Edit:FFS sorry for the necro. to much work in this post to just delete and hide embarrassment.

[Pickles von Brine]

Doesn't NetApp use something called Waffle for how their file system/drives are setup to be redundent @Dlog? 

[Hugh54321]

So what you are saying is I should buy lots of intel nucs

[r00tb33r]

The gasket between the chair and the keyboard.