7

I've been wanting to do a raid 1 setup in my home with a pair of sata drives. Someone told me that if the controller fails, you can't just get a new controller because you'll have to reformat the drives.

Or is that true only in some implementations? I was originally just looking at an onboard raid controller, or an entry level nas drvice like the intel SS4200-E, but If the hardware (controller) ever fails, will I be out of luck accessing the data if I can't get the exact same hardware to replace it?

Jeremy
  • 903
  • 2
  • 9
  • 19

8 Answers8

4

Maybe, maybe not. That's *exactly* why I use software raid. I could care less about performance, in exchange for guaranteed compatibility. I know it's a religious issue, but that's a no-brainer in my world.

pboin
  • 1,086
  • 8
  • 11
  • honest question: how often are you ripping disks out of one server and plugging them in to another? I've never had a problem procuring a replacement RAID card in the event of a failure from HP, Dell, or IBM, so the only real value (if you're doing this strictly for compatibility) is so that you can yank disks from one server and stick them in another, which I've never seen anyone do ever. – MDMarra Oct 22 '13 at 01:44
  • 1
    I work in a large laboratory with hundreds and hundreds of Linux boxes of all sizes. We have failed RAID cards often enough to make this a consideration. – pboin Oct 28 '13 at 17:56
3

Typically yes the config is per controller type. So if you have a failed controller and can't replace it with the equivelant you'll probably be SOL.

As long as it's the same kind though, I've never seen one which wasn't able to import the config from the drives. This is usually stored on each drive.

Brian Desmond
  • 870
  • 4
  • 7
2

If you're using RAID-1, then the controller is usually not relevant. In fact, you SHOULD be able to take a drive from a RAID-1 pair, and plug it into a normal controller and it should work. I've done it in the past.

The only thing you will need to be careful of is which controller you replace it with. Some controllers (the ones in our Dell blades, for example) will only make a RAID-1 array by destroying both disks. Some controllers will let you choose a Source, or Primary disk will not be wiped when creating a RAID-1 array.

Mark Henderson
  • 68,316
  • 31
  • 175
  • 255
1

Or use the OS level RAID1 feature instead of a card - then the controller for the disks wont matter (and as we're talking RAID1, mostly any OS will be able to boot from it for the reasons Farseeker put through as well).

Oskar Duveborn
  • 10,740
  • 3
  • 32
  • 48
1

If you know the parameters the RAID container used, and it was something simple like RAID 0, 1, 3, or 5, then you can perhaps get lucky if you know someone who can code.

I had a CMD raid controller fail, and after a bit of (easy) sniffing around on the drives, I found that each drive had the RAID set configuration stored in the first few blocks. Skipping that, and writing a simple program to read from each drive in the correct order (including RAID 5's rotated parity) I managed to recover all the data.

Michael Graff
  • 6,588
  • 1
  • 23
  • 36
0

It would depend on how the specific controller implemented the array. If it stores all the config in the controller and does no format abstraction, it may be able to be bypassed. You could test for that by setting up the initial array and then pulling one of the drives and connecting it to a regular disk controller.

If the array can be moved between systems ("use disk config or use controller config" type prompt) or can accept non-identical drives in the array, there is likely some abstraction happening. You'd likely need a controller which uses the same format (which in practical terms means an identical card) if the existing one failed. Or you could just decide that on controller failure you'd fall back to the most recent backup.

Jeremy M
  • 819
  • 4
  • 10
0

Based on my home experience - do not spend your time with anything below RAID 5 :) Troubles of migration or crash recovery are much higher than performance gained.

alexm
  • 458
  • 3
  • 11
-4

Software RAID is horrible for a number of reasons and people who think that they are somehow more reliable than hardware, simply do not know how enough about hardware controllers. First, software controllers (even HBA - hardware that uses the computer's CPU and RAM for its function) have a huge failure rate from unexpected shutdown, server lock-up and power loss. So many times we have seen a hung server with a software RAID that could only be rebooted through a hard shutdown and about 50% of the time it will come with the RAID status 'degraded' and will be dog-slow for the next X hours while its rebuilding itself. On a RAID1 mirror, that's an inconvenience, but running RAID 10 or 5 (where a multi-disk degradation means loss of the entire volume) is asking for trouble.

Hardware RAIDs are FAR more efficient, have better performance and higher fault tolerance. Also, although the one controller is a single point of failure, controllers rarely fail. There are no moving parts and as long as they are kept cool and have clean power, they almost always outlast their usefulness. We have tossed many controllers in the trash that originally cost $2-5k. There was nothing wrong with them, they were just obsolete.

Recovery from a hardware RAID controller is not difficult. Unfortunately, RAID has taken a technological step backwards in the last decade. I am not sure why, but new SATA/SAS controllers are missing a lot of functions that old school IDE/SCSI controllers used to have. One particularly annoying fact is that you can no longer take a single drive (with data on it) and make it into a RAID1 volume - unless the drive was created as a RAID member, it must be wiped. Likewise, you can no longer take a RAID 0 array and make it fault tolerant - you got to back it up, wipe everything, build the array and then restore.

The notion that if one hardware controller fails, you may need to wipe the drives is inaccurate, but I think I know what they may have been referring to. If your controller fails, each one of the drives has the RAID information on them. You can replace the controller with an identical model and it will see the old raid configuration and take off where the failed controller left off. The problem happens when you cannot get your hands on an identical controller. For whatever reason, different controllers do RAIDs differently. However, almost always the new controller will recognize the RAID and mount it in some sort of limited mode. The performance may be bad, or it may not let you rebuild, but the data will be accessible. This will allow you to back up the data to another place, then wipe your disks, build a new array and put your data back. RAID 1 is particularly awesome because RAID members can be put into a computer without even a RAID controller! You can actually take either of the drives that were part of a RAID 1, connect them to a computer via any controller and the data will be there. You may have trouble booting from this drive because it will be missing drivers and boot paths for your new controller, but that can be addressed manually. Bottom line, if your hardware RAID controller fails (highly unlikely) - you can take either of the drives, connect it to any machine and your data will be there.

Tyson
  • 9
  • A lot of your rant is somewhat controller specific - especially the one about transforming single disks into RAID 1. That can be done on both HP and Dell controllers without issue (and I'm sure on plenty of other controllers as well). Also, nowhere in this rambling is there an answer to the actual question. While I do agree with you that true hardware RAID controllers from Dell, HP, IBM, Cisco, etc are almost universally better options than software RAID, I have to downvote this because it's just plain not an answer to the question. – MDMarra Oct 22 '13 at 01:43
  • 1
    Even if it can be done without issue, you still have to know what you're doing. LSI IR & MR controllers, for instance, will convert a single drive into a R1 array retaining data, but you had better make damn sure you left room at the end of the drive for the 1MB size rounding and the array metadata that's stored at the end of the disk or you'll find you have truncated your data partition. – MikeyB Oct 22 '13 at 01:51