upgrading storage raid10 or raid z2 or ...

[rwps]

I've had a HP proliant N36L setup with freenas 7 for a fair while. It has a 2tb disk in it & I keep another 2tb disk and a 1tb offsite (backed up monthly). I'm coming to the point I need to increase storage beyond the 2tb.

I've come across another couple of 2tb disks & 8gb of ram, and am tossing up how I should best set the NAS up.

I know have 4 x 2tb disks, and 1 x 1tb disk.

I'm leaning towards all 4 x 2tb disks in the NAS setup as either RAID10 (2 mirrors) or RAIDZ2 – both giving me 4tb usable. As I read it RAID Z2 would be the way to go given with 4 disks it shares the risk of failure across all the disks, where as a couple of 2tb mirrors could be cactus if the 2 disks in a mirror failed.

Obviuosly I'm looking at RAIDZ2 I'm also thinking to move to FreeNAS8

So overall I'm thinking
4x2tb in RaidZ2
move from freenas7 to freeNas8
keep 1tb disk offsite for important data (back up intermittently)
All of this to serve as the main data storage for the household (serving video, music & files)

Does this sound a good way to go?

 

[Kimba]

Sounds great -- far more robust than my setup.

 

[louisk]

I don't see how they would have a differing amount of risk. Both models essentially have 2 parity disks and 2 data disks.

I think (having run striped mirrors for many years) you are unlikely to see both spindles in a mirror give out at the same time. If you relaxed the timing a bit, I would say with in a month is probably not unlikely if they were bought and deployed at the same time. This can be easily resolved by keeping spares physically present however. I'm running 8 spindles and I keep 4 spares. I've had zero failures from FreeNAS).

How are you doing your "offsite"? replication? do you ensure a successful copy of new data before you expunge your old backup?

Sounds like a reasonable plan.

 

[Joshua Parker Ruehlig]

there is slightly more risk with 2 mirrors stripped together.
If you lost 2 disks on the same mirror you would lose all your data, but with raidz2 you would have to lost 3 disks to lose any/all your data.

I would still go with stripped mirrors though if I only had 4 disks as there is a large performance benefit; mirrors take little cpu compared to raidz2, and the stripped vdevs would increase IO.

 

[rwps]

How are you doing your "offsite"? replication? do you ensure a successful copy of new data before you expunge your old backup?

At the moment my offsite backup is achieved using rsync to an external HDD connecetd by USB (freenas 7)

I've just splurged and bought a 3TB Seagate usb3 external HDD. So my plan now is to do a full backup to this, and do a key files backup to the 1TB (photos, personal stuff that can't be replaced). Where I always leave one disk offsite and bring one home each month (or whatever frequency I go for) to perform the backup.

What's the best backup method I should be using in Freenas8?
I'm a noob with ZFS, but it seems snapshots would be a good thing to incorporate (space being available) on the 3TB, and then not sure whether I'd do a straight rsync style backup for the relevant files to the 1TB or could I do snapshots of the relevant selected folders (again space dependant)?

 

[rwps]

but with raidz2 you would have to lost 3 disks to lose any/all your data.
....
I would still go with stripped mirrors though if I only had 4 disks as there is a large performance benefit;

Thanks for confirming, this is what leans me towards raidZ2, my data throughput needs aren't huge, but I care greatly about not losing data. So will err on that side.

 

[louisk]

If you're going to send snapshots from one to the other, I would make sure they are both running the same version of FreeNAS.

 

[rwps]

there's only one box here, one that I hook external HDDs to for backup purposes.
I'm planning on changing that box from Freenas7 to Freenas8 and implementing the new setup.
Do snapshots make sense in this scenario?

 

[louisk]

Snapshots only exist within the context of a zpool. If you only have 1 host, I think you're probably stuck doing something like rsync and hardlinks.

 

[rwps]

can each of the external HDDs be their own zpool. Would that work?

 

[cyberjock]

It is a valid configuration, but you will have no redundancy. Failure of the drive will result in a loss of data in the zpool.

 

[rwps]

but given that these external drives are in themselves the offsite redundancy, that's ok isn't it? i.e. that's the same risk that would always exist with a drive failure in one of these usb external HDD devices, you can lose that copy of the data, but will still have what other copies you have made.
The "original" data is living in a raidZ2 box.

 

[Joshua Parker Ruehlig]

can each of the external HDDs be their own zpool. Would that work?

Of course, though you get no benefits of a larger more manageable zpool, performance increase from striping, or redundancy.

 

[rwps]

Of course, though you get no benefits of a larger more manageable zpool, performance increase from striping, or redundancy.

Thanks, but did I miss something here. I thought RaidZ2 gave me 2 redundant disks, and 2 offsite disks give me two more redundancy avenues (which I've been told need to be their own zpools). Or is it that the external HDD's used for sync & then offsite backup don't need to be zpools?

Avoiding data loss is king for me, and given limitations of home budgets/practicalities I think from my reading and the assistance in previous posts that this hardware:

-hp 4 bay server (proliant n36L)
- 4 x 2TB disks
- 1 x 3TB external usb3 disk
- 1 x 1TB external usb2 disk

Would now go into:
- 1 x server with raidZ2, using 4 x 2TB, for 4TB storage with 2 redundant disks
- 2 x offsite disks with backed up data (1x 3 TB, 1 x 1TB)

 

[paleoN]

Thanks, but did I miss something here. I thought RaidZ2 gave me 2 redundant disks, and 2 offsite disks give me two more redundancy avenues (which I've been told need to be their own zpools). Or is it that the external HDD's used for sync & then offsite backup don't need to be zpools?

Yes, exactly. The offsite disks would be 1 redundancy overall. Though 1 disk could fail and you would still have the other part of the backup.

I would lean toward formatting the offsites UFS, but maybe each in their own zpool works better with the current FreeNAS. You will need to export and import them all the time.

The proper thing to do is have a second NAS setup, but you know that already.

Would now go into:
- 1 x server with raidZ2, using 4 x 2TB, for 4TB storage with 2 redundant disks
- 2 x offsite disks with backed up data (1x 3 TB, 1 x 1TB)

I'm with louisk with using striped mirrors. If you look at the [post=28410]graph in this post[/post] you can see mirrors are very safe. That being said you still lose 2 drives to redundancy either way.

 

[cyberjock]

I'm not sure why you'd choose a RAID10 versus RAIDZ2. With a RAID10 a failure of the same 2 disks on the 2 mirrors will result in a loss of all data. With RAIDZ2 any 2 disks can fail and you will lose nothing. Statistically, your chances of a loss of data are higher with a RAID10.

 

[louisk]

Performance really. There is no checksumming going on, only writing of data.

In terms of loss stats, once you have a mirror (or set of mirrors) up and running, typically, you don't lose both disks together, and as soon as they are staggered in terms of age, its extremely unlikely that they will double fault. I've got about 60T at work that is all RAID10. We lose spindles every month or so (we have ~150 total that run 24x7). Even with all this loss, we've never had a double fault. Its run this way for ~4yrs now.

 

[paleoN]

Louis said it better than I could have. Also, you only need 2 additional disks to expand the pool vs 4 disks for raidz2. Though I think his box only has room for 4 so that doesn't really apply in this case.

 

[louisk]

The 2850 is fairly limited, but the MD1000 that is plugged into it will hold 15 spindles.

 

[cyberjock]

Performance really. There is no checksumming going on, only writing of data.

In terms of loss stats, once you have a mirror (or set of mirrors) up and running, typically, you don't lose both disks together, and as soon as they are staggered in terms of age, its extremely unlikely that they will double fault. I've got about 60T at work that is all RAID10. We lose spindles every month or so (we have ~150 total that run 24x7). Even with all this loss, we've never had a double fault. Its run this way for ~4yrs now.

If you are running gigabit, the "overhead" of checksuming will not matter unless you are using a low power Atom CPU. Besides, the first time you lose your data will be the last time you take that slight chance.

Strangely, I've found that running drives 24x7 is better for them than on/off cycles. My guess is that the thermal stress accounts for alot of 'wear and tear'. I saw a study somewhere that provided some evidence that supported this.