Case Study: RAID Tolerance to Failure

[wpirobotbuilder]

Case Study: RAID Tolerance to Failure

 

In this post I talked a lot about the pros and cons of running a RAID array, explained the different RAID levels, and some more specific aspects of RAID. In this snippet, I'd like to take a closer look at the tolerance of a RAID array to drive failures.

 

To do this, let's consider the storage configuration given here, by our own looney:

 

Looney has a RAID volume with 14 2TB drives, and he wants to use it for storage of media and his computer backups. Let's consider the effects of various RAID levels on this volume, and what the pros and cons are of each one.

 

RAID 0:

 

         Without a doubt, this would be the fastest configuration. 14 drives in RAID 0 would give higher random performance and insane sequential performance. But what happens if a drive fails? Well, since a RAID 0 array has no tolerance to failure, he'd lose all his data.

 

        Best Case: Not Applicable

        Worst Case: Any one drive failure

        Tolerance: None

        Space: 28TB

 

If he was using the RAID array as a recording disk, he might want a RAID 0 volume. But media playback performance wouldn't be impacted much with a RAID 0, and his home network would be bottlenecked at ~125 MB/s with gigabit Ethernet. Plus, that puts his entire movie collection at risk, as well as losing all his backups. Clearly, this isn't the choice for him.

 

RAID 1:

 

         Without a doubt, this would be the safest configuration. 14 drives in RAID 1 would have okay performance in both writes and reads. It can also sustain the loss of up to 13 drives, so he wouldn't have much to worry about.

 

        Best Case: Not Applicable

        Worst Case: All drives fail

        Tolerance: 13 drives.

        Space: 2TB

 

The problem with this configuration is that he doesn't end up with much space, so he can't have much of a media collection or many backups of his computer. This also isn't the configuration for him.

 

 

RAID 5:

 

         This is the most efficient configuration. 14 drives in RAID 5 would have good read performance, but would take a hit in write performance. It can also sustain the loss of any one drive, so he has less to worry about as far as data reliability goes.

 

        Best Case: Not Applicable.

        Worst Case: Any two drives fail.

        Tolerance: 1 drive.

        Space: 26TB

 

With this configuration, looney gets his space, and some redundancy. This wouldn't be a bad choice for him, but let's see what else there is.

 

RAID 10:

 

         This configuration wouldn't impact write performance at all, and would give us some redundancy. 14 drives in RAID 10 is 7 RAID 1 arrays striped together, so we get pretty beast performance. However, in this case, redundancy has a best and a worst case. For instance, if every drive we lost came from a different RAID 1 stripe, we could lose up to 7 drives, because each stripe would still have another drive to keep it going. On the other hand, if we lose two drives from the same RAID 1 stripe, the entire RAID volume fails.

 

        Best Case: 7 drives fail, one from each stripe.

        Worst Case: Two drives from the same stripe fail.

        Tolerance: 1-7 drives.

        Space: 14TB

 

With this configuration, looney gets good performance, but only gets half his maximum space. While it's true that he could tolerate a large number of drive failures, he might want to have a higher tolerance for the worst case scenario.

 

RAID 6:

 

         Basically a safer version of RAID 5, using two drives for parity calculations. 14 drives in RAID 6 would have good read performance, but would take a very large hit in write performance. It can also sustain the loss of any two drives, so he has even less to worry about as far as data reliability goes.

 

        Best Case: Not Applicable.

        Worst Case: Any three drives fail.

        Tolerance: 2 drives.

        Space: 24TB

 

With this configuration, looney gets lots of space, and even more redundancy. This would be a great choice, since much of what he does is read from his array, and backups can be run overnight, making the write performance less of an issue.

 

 

RAID 50:

 

         This configuration would gives us increased performance over RAID 5, and would still gives us some redundancy. 14 drives in RAID 50 is 2 RAID 5 arrays striped together, so we get better read and write performance. Once again, redundancy has a best and a worst case. For each RAID 5 stripe we can lose a maximum of 1 drive. If we lost two drives, one from each stripe, we would still have our array. However, if we lost two drives from the same stripe, we would lose the whole RAID volume.

 

        Best Case: Two drives fail, one from each stripe.

        Worst Case: Two drives from the same stripe fail.

        Tolerance: 1-2 drives.

        Space: 24TB

 

With this configuration, looney gets the same minimum tolerance as RAID 5, and more performance at the cost of another hard drive. This isn't a bad configuration, but it would require some advanced hardware to do, and it depends on his desired tolerance.

 

RAID 60:

 

         This configuration would gives us increased performance over RAID 6, and would still gives us the redundancy of RAID 6. 14 drives in RAID 60 is 2 RAID 6 arrays striped together, so we get better read and write performance.

 

        Aaaand, on to redundancy. For each RAID 6 stripe we can lose a maximum of 2 drives. If we lost four drives, two from each stripe, we would still have our array. However, if we lost three or more drives from the same stripe, we would lose the whole RAID volume.

 

        Best Case: Four drives fail, two from each stripe.

        Worst Case: Three drives from the same stripe fail.

        Tolerance: 2-4 drives.

        Space: 20TB

 

With this configuration, looney gets the same minimum tolerance as RAID 6, and more performance at the cost of two more drives. Like RAID 50, this requires advance hardware to do, but still gives us the redundancy we'd like.

-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

 

With all this in mind, how do we choose a RAID volume? We know that one requirement is that it should be redundant, since he's storing lots of data that would be hard to replace, plus backups which would be bad to lose if another of his computers went down. We also know that having high performance isn't necessary, and that he needs lots of storage space.

 

Given these requirements, we can pick RAID 5, 6, 50, and 60 as our viable candidates. So which one should he go with?

 

SPOILER ALERT: He chose RAID 6.

 

Based on his choice, he might not have wanted to go with a hardware solution (expensive, and a single point of failure), or didn't need the performance of a hardware solution to provide RAID 50 or 60. He also might have wanted the redundancy of RAID 6 over that of RAID 5.

-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

 

Final Thoughts:

 

Looking at striped volumes, I think it's useful to recognize that striping nested RAID volumes does not increase worst-case redundancy, only performance, and at the cost of more drives. In a striped RAID, your array is only as redundant as the lowest RAID level. In RAID 50, that's RAID 5. In RAID 60, that would be RAID 6. In RAID 10, that's RAID 1.

 

When chance is on your side, a striped redundant array can provide additional benefit. But for those who want to be absolutely sure, it doesn't help much.

 

Nesting RAID volumes can increase redundancy, unfortunately there are no hardware RAID cards that support such RAID arrays, like a RAID 1 array of nested RAID 6's (RAID 61?). That would provide 4 drives worth of redundancy, but in looney's case, he would only get 10TB of space (7 drives per RAID 6, 5 drives of space available per RAID 6, only get 1 RAID 6 worth of storage). RAID 65 (RAID 5 of nested RAID 6 volumes) would be intense, but he'd only be able to use 12 drives (4 drives per RAID 6, 3 RAID 6's for the RAID 5), and would get two RAID 6's worth of space, for a total of 8 TB. Also, good-bye to write performance.

[looney]

Based on his choice, he might not have wanted to go with a hardware solution (expensive, and a single point of failure), or didn't need the performance of a hardware solution to provide RAID 50 or 60. He also might have wanted the redundancy of RAID 6 over that of RAID 5.

 

Spot on

[tycheleto]

Case Study: RAID Tolerance to Failure

RAID 60:

This configuration would gives us increased performance over RAID 6, and would still gives us the redundancy of RAID 6. 14 drives in RAID 60 is 2 RAID 6 arrays striped together, so we get better read and write performance.

Aaaand, on to redundancy. For each RAID 5 stripe we can lose a maximum of 1 drive. If we lost two drives, one from each stripe, we would still have our array. However, if we lost two drives from the same stripe, we would lost the whole RAID volume.

Best Case: Four drives fail, two from each stripe.

Worst Case: Three drives from the same stripe fail.

Tolerance: 2-4 drives.

Space: 20TB

With this configuration, looney gets the same minimum tolerance as RAID 6, and more performance at the cost of two more drives. Like RAID 50, this requires advance hardware to do, but still gives us the redundancy we'd like.

You repeated the part on RAID 50 Redundancy for RAID 60, which doesn't match the correct values as stated in your best case and worst case scenarios.

[wpirobotbuilder]

You repeated the part on RAID 50 Redundancy for RAID 60, which doesn't match the correct values as stated in your best case and worst case scenarios.

Fixed, that was a result of judicious copy-paste. Thanks

[tycheleto]

Fixed, that was a result of judicious copy-paste. Thanks

 

Looked liked it... Always glad to help, don't want other people getting misunderstandings.

 

Edit: I realized it's still inaccurate. For RAID 60, you can lose two drives from each stripe right? You can lose up to 4 drives(2 from each stripe) and still have the array. However, if you lose three drives from the same stripe, you lose the entire volume

Edited October 12, 2013 by tycheleto

[amintha1992]

that was a lot of information... in a nice package.  thanks ...

[wpirobotbuilder]

Looked liked it... Always glad to help, don't want other people getting misunderstandings.

 

Edit: I realized it's still inaccurate. For RAID 60, you can lose two drives from each stripe right? You can lose up to 4 drives(2 from each stripe) and still have the array. However, if you lose three drives from the same stripe, you lose the entire volume

It was fixed, but after the fact. You might have picked it up before I had submitted changes.

 

Regardless, it's fixed now.