Is this a good pool configuration?

[ash77]

I'm building a new pool and trying to plan the configuration. My server has 24 bays and will be used for data archiving and home media streaming. I'm optimizing for capacity and redundancy, and I'm not very concerned about speed. I also expect to be doing a lot more reading from the pool than writing.

I am using 12TB HGST Ultrastar HUH721212ALE600 drives. I can use 3 RAID-Z2 vdevs of 8 drives each or 2 RAID-Z3 vdevs of 12 drives each. Both configurations obviously have the same number of parity drives. The first configuration can only lose a maximum of any 2 drives without data loss, but the second configuration can lose a maximum of any 3 drives without data loss. If a favorable set of drives fail, then either can lose up to 6 drives without data loss. I understand RAID-Z3 has a higher cost for parity calculation, but I would expect that to only matter when writing to the drive (which I expect to do relatively rarely and I don't really care much in this case) and when recovering data due to either partial or complete drive failure in which case I will replace the failing drive. Resilvers would slower as well.

I think for my particular use case the 2 RAID-Z3 vdevs of 12 drives each is the better option.

Does anyone see any flaws in my reasoning?

 

[Jessep]

A better pool design would be 3x6 RaidZ2, more IOPs, reasonable redundancy.

Remember RaidZ2 isn't a backup so plan for that as well.

 

[Heracles]

Hey Ash,

I would also go with the 3 vdevs of 8 drives each using RaidZ-2. As for redundancy, you will probably have pre-failure warning signs, so you can replace the drive even before it fails. Also after first failure, 16 drives out of 23 are from different vDev, so you already have a high probability not to drop to no redundancy on a vdev.

Looking at that in another way, the probability to loose a vdev is : 7 / 23 * 6 / 22. That gives 42 / 506. That is 1 / 12 or 8.5%.

91.5% of probability to survive the loss of 3 drives is still very good but in all cases, it will be up to you to monitor your server and replace a drive fast enough for never dropping to as many as 3 failed drives at once.

And as Jessep already mentioned, snapshots and RaidZ-x are no backup and you still need to plan for actual backups. What will you in case a fire destroy your server or a misconfiguration corrupts your pool ?

Better to plan for it before it happens instead of after... So many people do their backups only after they loss everything. Don't be one of them

 

[Stux]

I use 3x 8-way Raidz2 in my own 24 bay server.

I think it’s a good compromise between capacity, iops, and allows you to start with one or two vdevs, add a third and then grow another as you replace drives etc.

If you needed more iops (which you probably don’t) then you would consider 4x 6-way

 

[pschatz100]

I use 3x 8-way Raidz2 in my own 24 bay server.

I think it’s a good compromise between capacity, iops, and allows you to start with one or two vdevs, add a third and then grow another as you replace drives etc.

If you needed more iops (which you probably don’t) then you would consider 4x 6-way

With that big an investment in drives, I would also keep a cold spare that is already burnt in - for quick replacement.

 

[Stux]

Well, I keep a backup. Thus I have time to purchase a replacement drive when I need too.