RAID redundancy - RAID 5 vs RAID 6

[batfastad]

Hi everyone
Just starting the process of building a new shared files NAS system. At the moment we have a simple RAID 1+0 setup... 2 stripes of 2 200GB drives, mirrored. Giving us total capacity of 400GB

It's time these drives were replaced (3 years of heavy service) and I wanted to incorporate a bit more redundancy.
We're going to make the purchase of a decent 3Ware 8 port RAID card which will give us access to RAID 5 and RAID 6

I've read that RAID 6 is slightly better, as it can cater for 2 drives to fail and still be ok, whereas in RAID 5 you can only have 1 drive fail.

For RAID 6 you need a minimum of 5 drives... so 5x500GB drives should equal 1.5TB of disk space (according to the formula on Wikipedia... http://en.wikipedia.org/wiki/RAID_6#RAID_6)
I will also configure a 6th drive as a hot-spare, so should anything happen then that 6th one will automatically start to be rebuilt into the array.

Here's my question:
The more drives you add, the higher the chance you have of a drive failing.
So if for example you had 50 drives in a RAID 6 array, can you still only get away with 2 drive failures? I don't like those odds!
Or is the number of failures somehow proportional to the number of drives you have?

Thanks, B

[Mblaster]

Hi everyone
Just starting the process of building a new shared files NAS system. At the moment we have a simple RAID 1+0 setup... 2 stripes of 2 200GB drives, mirrored. Giving us total capacity of 400GB

It's time these drives were replaced (3 years of heavy service) and I wanted to incorporate a bit more redundancy.
We're going to make the purchase of a decent 3Ware 8 port RAID card which will give us access to RAID 5 and RAID 6

I've read that RAID 6 is slightly better, as it can cater for 2 drives to fail and still be ok, whereas in RAID 5 you can only have 1 drive fail.

For RAID 6 you need a minimum of 5 drives... so 5x500GB drives should equal 1.5TB of disk space (according to the formula on Wikipedia... http://en.wikipedia.org/wiki/RAID_6#RAID_6)
I will also configure a 6th drive as a hot-spare, so should anything happen then that 6th one will automatically start to be rebuilt into the array.

Here's my question:
The more drives you add, the higher the chance you have of a drive failing.
So if for example you had 50 drives in a RAID 6 array, can you still only get away with 2 drive failures? I don't like those odds!
Or is the number of failures somehow proportional to the number of drives you have?

Thanks, B

In a RAID6 array you can always 'only' get away wtih having 2 drives failed at any one time, no matter how many drives are in the array. As you're planning to use a hot-spare it improves your redendency even more. The chances of 3 HDD's all failing with 12 hours (before the hot-spare has re-built), are extremely unlikely, espcially in an array of 6 disks

[Moby-Dick]

why not use 6 drives and go for Raid 0+1 ?

[batfastad]

@Mblaster
Right ok. So how do huge organisations deal with that? I guess they use some sort of huge clustering system that I don't even know about.
So I should be fine using RAID 6 with up to 8 drives in use right?

@Moby-Dick
Because with RAID 0+1 don't I get exactly the same total capacity?
(6 x 500GB) / 2 = 1.5TB
But without the redundancy of RAID 6... ie: 2 simultaneous drive failures and I'm still fine.
I think with a mirror of stripes (0+1) you can get away with 2 drive failures, but they have to be from the same stripe.
RAID 1+0 is better than 0+1 I think.

Thanks, B

[Moby-Dick]

Raid arrays are no substitute for regular backups at the end of the day it makes very little difference to your fault tolerancy as to which mechanism you use ( Raid 5 with a hot spare is as good as any ) If an array is going to die , its going to die

[Mblaster]

@Mblaster
Right ok. So how do huge organisations deal with that? I guess they use some sort of huge clustering system that I don't even know about.
So I should be fine using RAID 6 with up to 8 drives in use right?

@Moby-Dick
Because with RAID 0+1 don't I get exactly the same total capacity?
(6 x 500GB) / 2 = 1.5TB
But without the redundancy of RAID 6... ie: 2 simultaneous drive failures and I'm still fine.
I think with a mirror of stripes (0+1) you can get away with 2 drive failures, but they have to be from the same stripe.
RAID 1+0 is better than 0+1 I think.

Thanks, B

I don't know very much about the huge organisations storage setups, biggest thing I've dealt with is smallish RAID 5 arrays, but I'd imagine they use multiple RAID 5/6 arrays, or some other custom solution.

RAID 0+1 doesn't provide the same redendency as RAID 6, and scales much worse with adding disks, since you need to add another pair each time, only getting half he space you add, where as with a RAID 6 array you'll get all the space from any drives you add to the array. RAID 0+1 can sustain more disk failures theoretically, but only if you're lucky enough to not get the failures on the same mirror.

[Moby-Dick]

on the large HP EVA SAN's I run, we use a mixture of RAID 5 and RAID 0+1 disk groups ( 0+1 primarily for performance of log file drives )

[Splash]

As Moby said, RAID!=Backups.

[peterb]

@Mblaster
Right ok. So how do huge organisations deal with that? I guess they use some sort of huge clustering system that I don't even know about.
So I should be fine using RAID 6 with up to 8 drives in use right?

MobyDick has given one solution, but medium size enterprises would probably use RAID 5 with a hot spare and monitoring software to alert them when a disk fails, so the defective one can be hot swapped out. They may also use a mirror server so if one server dies, the other takles over seamlessly. However you will almost certainly find a large tapestore, using multiple tapes, configured to do daily backups, Some organisations might backup more frequently (with two servers, you could back up each one daily, but staggered by 12 hours).

However, the backup strategu really depends on a risk assessment. With a 6 disk array, whayt are the chances of two disks failing undetected? What is the impact of a total array failure? How much current data could you afford to lose? How long could you afford to wait to get the system up and running again?

the answers to these questions depend on the business you are running - and ultimately is a management decision. The lower the downtime you can afford, and the higher the resiliance, the more it will cost - both in hardware amd administration effort.

[batfastad]

Ok that all sounds fine.

But why would you go RAID 5 over RAID 6?

I'd be using 5 drives with a hot-spare which gives me 2TB for RAID 5 and 1.5TB with RAID 6.
1.5TB is fine capacity for us at the moment, and if we need to add more, then there's still another 2 free slots in our hot-swap cages.

[peterb]

RAID 5 maximises capacity for a given number of disks - at the expense of resiliance (only one disk failure) - which doesn't matter much if the raid is monitored. As I said earlier, it comes down to risk assessment and what is deemed an acceptable risk for your business model.

[batfastad]

Right ok.
Well it seems 5x500GB in a RAID 6 with a hot-spare is definitely the way to go for us.

500GB drives are pretty cheap at the moment.
And with 2 spare hot-swap bays in the case I'm building, then we can easily increase capacity from 1.5 to 2.5TB if necessary.

Thanks for all the info guys!