Migrating from 2x4tb (mirrored) to 4x4tb RAID 5-like setup?

[derblub]

FreeBSD 9.3-RELEASE-p31

So currently I've a 2x4tb mirrored volume setup in freenas (zfs). While first I wanted to be more on the safe side, I've realized that I don't have that important data and storage space is not free and I'm running short.

I've ordered another two 4tb drives and they should arrive shortly. I thought about switching my current 100% mirrored setup to something like a RAID 5.

Usable space would be 12tb then, instead of just 8.
Now my questions:

• First, is it even possible to migrate my current setup to the wanted new 4-drive setup, without loosing any data? (I only have those 4 drives (while 2 have my data), nothing else where I can do a temporary backup)
• On a real RAID, I could copy my data to a single new disk > format the other 3 to a RAID 5 > copy data back to 3-drive setup > format 4th drive and extend RAID 5, right?. How would I do something similiar in freenas zfs?

Thanks in advance for any pointers and help!

 

[anodos]

FreeBSD 9.3-RELEASE-p31

So currently I've a 2x4tb mirrored volume setup in freenas (zfs). While first I wanted to be more on the safe side, I've realized that I don't have that important data and storage space is not free and I'm running short.

I've ordered another two 4tb drives and they should arrive shortly. I thought about switching my current 100% mirrored setup to something like a RAID 5.

Usable space would be 12tb then, instead of just 8.
Now my questions:

• First, is it even possible to migrate my current setup to the wanted new 4-drive setup, without loosing any data? (I only have those 4 drives (while 2 have my data), nothing else where I can do a temporary backup)
• On a real RAID, I could copy my data to a single new disk > format the other 3 to a RAID 5 > copy data back to 3-drive setup > format 4th drive and extend RAID 5, right?. How would I do something similiar in freenas zfs?

Thanks in advance for any pointers and help!

Most people do not consider RAIDZ1 / RAID5 to be a good choice for drives over 1TB. You can see dell's recommendation to this effect here: http://en.community.dell.com/techce...erations-and-best-practices-released#comments or the various "RAID5 is dead" posts.

Perhaps look into a 6-disk RAIDZ2 pool. There is no way to convert a mirrored vdev to a RAIDZ vdev. BTW, you should figure out a way to back up your data. RAID is about redundancy, storage capacity, and performance. It does not eliminate the need to back data up.

 

[Mr_N]

BTW, you should figure out a way to back up your data. RAID is about redundancy, storage capacity, and performance. It does not eliminate the need to back data up.

I don't see how you can include storage capacity in that statement :P

 

[Arwen]

I don't see how you can include storage capacity in that statement :p

Simple. RAID-0, (or Stripe or Concatenate), was about extending smaller disks into larger LUNs.

 

[Mr_N]

But you'd only do that for improved IOPS/BW not purely for storage space otherwise it'd be safer for your data just using single drives...

 

[Robert Trevellyan]

What if you want contiguous storage?

 

[anodos]

I don't see how you can include storage capacity in that statement :p

I'd say storage capacity is the primary reason people choose parity raid over raid10 / mirrors. In general, when you're planning storage arrays you have to consider capacity, speed, and redundancy. BTW, I have a hard time seeing how you can have a volume size of over 10TB without some sort of raid. :p

 

[Scharbag]

Most people do not consider RAIDZ1 / RAID5 to be a good choice for drives over 1TB. You can see dell's recommendation to this effect here: http://en.community.dell.com/techce...erations-and-best-practices-released#comments or the various "RAID5 is dead" posts.

Perhaps look into a 6-disk RAIDZ2 pool. There is no way to convert a mirrored vdev to a RAIDZ vdev. BTW, you should figure out a way to back up your data. RAID is about redundancy, storage capacity, and performance. It does not eliminate the need to back data up.

Backup, backup, backup!!! I am still amazed at the number of people that take the time to figure out how to setup a NAS with RAID and then choose to not backup. I guess sometimes you do not need to backup. But I would be loathe to "replace" many TBs of data over my cable modem :)

So yeah, only way to change types of RAID would be to have an intermediate storage solution. RaidZ2 is, IMHO, the most reasonable way to ensure your data's availability and its performance is more than adequate for home use. I would avoid RaidZ1 as resilvering can take a while with large drives. If that second drive fails during a resilver, you better have a backup:)

 

[Mr_N]

If you absolutely had to have a large block of storage in a single chunk then sure, otherwise there is no reason from a capacity argument to use raid over single drives :P

 

[Scharbag]

If you absolutely had to have a large block of storage in a single chunk then sure, otherwise there is no reason from a capacity argument to use raid over single drives :p

There is a misunderstanding here. Capacity, when talking about data, is not just raw capacity but it is always implied to be "reliable capacity."

When you purchase or build a storage solution, there are the three items, as mentioned above, that define the system: Speed (IOPS and/or throughput), Capacity and Reliability (fault tolerance - not a backup replacement!!!). By definition, RAID is a Redundant Array of Independent Disks (yes, RAID0 should just be called AID but I digress). The initial RAID releases primarily dealt with the redundancy aspect but RAID was very quickly adopted to increase the reliable storage capacity as well.

So, answer me this: how would you build a 100TB filer that offers the equivalent data availability(reliability) of a RaidZ2 system without using some form of RAID?

Cheers,

 

[Mr_N]

So, answer me this: how would you build a 100TB filer that offers the equivalent data availability(reliability) of a RaidZ2 system without using some form of RAID?

Cheers,

Were getting off track or your misunderstanding my viewpoint. if you want the reliability of Z2 then you use Z2 but your not setting it up saying im gonna have so much capacity, your setting it up saying look how much space i have thats got multiple vdevs using z2 so it can withstand multiple drive failures and have great performance.

If you say I want 100TB of space and i don't care about redundancy or performace then you say 100TB divided by the sized drive which is the best $/TB then build a suitable system to house them all, done.
And this person only loses or has to restore from backup a single drive's worth of data should a drive fail, not the entire 100TB because he went and used Raid0.

One situation your spending alot more $, using alot of capacity up for redundancy/parity and the other your saying i want the most storage space i can get period.

You simply cannot make a use raid argument if your aim is raw capacity. (I don't consider raid0 as raid as you said above its Aid)

 

[Scharbag]

Were getting off track or your misunderstanding my viewpoint. if you want the reliability of Z2 then you use Z2 but your not setting it up saying im gonna have so much capacity, your setting it up saying look how much space i have thats got multiple vdevs using z2 so it can withstand multiple drive failures and have great performance.

If you say I want 100TB of space and i don't care about redundancy or performace then you say 100TB divided by the sized drive which is the best $/TB then build a suitable system to house them all, done.
And this person only loses or has to restore from backup a single drive's worth of data should a drive fail, not the entire 100TB because he went and used Raid0.

One situation your spending alot more $, using alot of capacity up for redundancy/parity and the other your saying i want the most storage space i can get period.

You simply cannot make a use raid argument if your aim is raw capacity. (I don't consider raid0 as raid as you said above its Aid)

RAID0 is still "RAID" as defined by storage vendors etc. Just like in ZFS, a striped volume (no redundancy) is still an array of disks. You were not clear on your above statement that you did not consider stripes or Raid0 as being RAID. 100TB made with 4TB drives would be 25 individual drives to manage if they were not grouped in an array. It would suck to figure out which volume you put your pr0n on with so many individual disks... :)

The argument still stands though. You are just choosing to not spend any money on reliability to maximize speed (stripes can be fast) and capacity. As a result, you have to live with the probability of the stripe array surviving (from here) = 65% for 1 year, 8% for 2 years, 1% for 3 years and 0% for more than 3 years.

So, the decision is still: how fast, how big, how reliable.

Cheers,

 

[Mr_N]

the decision will always be how reliable/fast and then way down the track how big can i afford to make it, if your taking about raid, if you want to run individual disks its easy, 25 is nothing :)

You will never win an argument for raid if you base it on raw capacity.

I'm sure we could do this all day but I think we've drawn our lines in the sand.