We have decided that RAID1 would be the best fit for our usage scenario for the following reasons:

1). Our overall storage requirements are relatively modest (< 500GB) 2). RAID1 offers the most simplicity in terms of controller overhead and ease of recovery from a failed drive

The only issue that worries us is simultaneous drive failure. In a standard two drive RAID1 setup this would be terminal.

So the question is, are there any hardware raid controllers that allow the mirroring of 3 drives (preferably with a 4th acting as a hot-spare)? The probability of three drives failing simultaneously is sufficiently tiny not to bother us.

As far as I can tell this might be possible in software raid but doesn't seem to be an option if you want to use a hardware controller?

Any advice (or alternative approaches) welcome!

  • Do you plan to use SAS or SATA drives? – user17642 Sep 18 '09 at 16:10
  • 1
    I'd point out that while odds are better the two drives won't fail there is the possibility that your one controller can fail... – Bart Silverstrim Sep 18 '09 at 16:13
  • Or a power supply. Or a fan. Or the server motherboard. Or anything else. It's very unusual for 2 disks in RAID 1 to fail together... unless of course one of them fails and you also fail to replace it. – Massimo Sep 18 '09 at 17:30

RAID 1 doesn't allow for three-disks mirrors; I know of no hardware RAID controller that can do that.

With some controllers, you can have a hot spare, which is a standby disk that's immediately used to rebuild the volume, should a drive failure happen.

  • +1 - and this is also why you have backups :) – TheCleaner Sep 18 '09 at 16:17
  • I was afraid this would be the answer. The thing that confuses me is that from a technical standpoint I can't see why if the controller can write to two drives simultaneously, it can't write to three. I guess RAID6 is the next best choice? – gpuk Sep 18 '09 at 16:18
  • Backups are all very well and good (and of course we have them) but it's the business interruption that is the problem. In the event that 2 disks die at the same time the server is off line until we can copy over the latest backup (unless you have clustered servers but we don't have the budget for this). – gpuk Sep 18 '09 at 16:21
  • performance, I'd say. And it's not a common spec. The most common you'll find are 0, 1, 5, and 10. – Bart Silverstrim Sep 18 '09 at 16:22
  • 1
    You can't ensure constant uptime without the budget to make it happen. You also haven't addressed controller or component failure. You'd need to look at partitioning your services or duplicating, physically, a server. – Bart Silverstrim Sep 18 '09 at 16:23

If you're worried about two disks dying at the same time then consider RAID6, possibly even with an extra hot-standby disk too, as this will support two disks failing without data loss. Be aware however that RAID6 is just about the slowest when it comes to disk writes.

Oh and there's lots of controllers out there than can support RAID6 by the way, let us know your server make/model and perhaps we can make some adapter suggestions.

  • RAID6 has performance issues when writing though. Each block write may result in activity on all four drives (the actual write to the drive the block you have change is on, reading the drive the next block is on (for parity calcs), and writing to the two other drives to store the parity results. While the three writes will be similar to the three writes using 3-drive RAID1 the extra read operation is a potential extra bottleneck. – David Spillett Sep 18 '09 at 16:39
  • 1
    I know, that's why I mentioned it... – Chopper3 Sep 18 '09 at 17:14

I don't know your specific scenario, but the odds of simultaneous disk failure is already really low, especially if you use two different brand drives with different manufacturing dates.

The way I've seen most deal with this is to have three drives with one designated as a hot spare, and the controller will automatically handle duplicating data immediately to the spare when a failure is detected.

For handling multiple drive failures I think you'd have to look at using something like RAID 10.

In either case, if you're worried about your critical data, you need a GOOD backup in place to recover data if there's a crash or error on some component that takes down the server. A hundred drives duplicating data won't help if the controller or motherboard or memory goes wonky (or your server is struck by lightning or pretends to be an anchor when a pipe bursts).

RAID isn't a panacea. I have had a RAID 5 array warn of a disk failure. During the rebuild when replacing the dead third drive, a critical error on a second disk, UNDETECTED, prevented a rebuild from completing. The drive controller would run repair after repair on the second disk and insist it was okay only to fail again.

There is a writeup about why today RAID 5 is worthless and RAID 10 was the way to go now. Basically with drives being so large there are a certain number of tolerated failed readable spots on the disk and as they get bigger the odds of encountering one of those spots increases, thus scenarios like mine get more common. A drive has an undetected error at a moment that is critical to recovering from an error. dead in the water. In my case we had to recover from bare metal from a backup. Lost just a couple days of files fortunately.


I don't know of any hardware controllers that can do this, but if you're using Linux, you can accomplish this with LVM mirrors. The lvcreate command supports an -m argument that lets you specify the number of copies of data to make. If you had each disk be a separate physical volume in LVM and used lvcreate -m2 you could effectively have a RAID 1 with three copies of the data.

  • Or just create a software RAID1 with 3, 4 or any number of drives. – MikeyB Sep 18 '09 at 16:40

Thanks to everyone for your insights and answers. I had (wrongly maybe?) assumed that drive failure was a lot more common than component failure (e.g. raid controller or mobo).

I think we'll opt for two disk RAID1 with a hot spare and simply mandate that the drives cannot come from the same manufacturer. If in the (apparently unlikely) event both drives go bang at the same time, we'll just have to minimise the downtime as best we can while we retrieve a backup.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy