1

I've currently got a 3 Hard Disk Linux Software RAID5 volume (mdadm). At least one disk is probably dying - long spin up times, and I've heard the click of death coming for a little while.

I've purchased a brand new disk, but while the old disk isn't fully dead, I'm not sure which makes more sense;

Option 1) Upgrade the 3-disk RAID5 array to a 4-disk RAID6 array, effectively using the new disk as additional parity, and meaning the array can take a 2-drive failure. This means I'd get a bit of extra life out of the old dying disk until it "really dies"?

Option 2) Just rip out the semi-dead disk and keep a RAID5 array.

Option 2.1) RAID5E? Move one of the disks (the new, or the dying) to be a online spare?

Option 3) ... that I didn't think of? :-)

I want to optimize for the option that gives me less chance of the array dying/gives me more parity - so Option 1 seems like the correct decision, but I worry that having the semi-dead disk still in the array might have the side effect of a more catastrophic failure later prehaps?

IO Profile/use case: The use case is very low I/O online file share storage, with a mixture of lots of little files, and lots of big files (ie, >1Gb each), it's also basically write-once, read often.

Caveat: I'm aware that RAID != backup, and all the array contents are safely backed up in multiple formats across multiple sites. The chance of real data loss is close to impossible with the amount of places it's backed up (the RAID effectively serves as the online faster access to that data). However, rebuilding the array from backups would be a timely pain in the ass. That's what I'm trying to avoid.

xconspirisist
  • 157
  • 7

2 Answers2

4

If you choose to keep the dying drive and convert your array to RAID6 immediately, you stand an increased risk of the drive croaking while you're rebuilding the new array; in the worst case requiring a time-consuming restore from backup.

I'd recommend replacing the drive and repairing your array, and then decide whether to splurge on an additional drive to go RAID6.

A viable alternative to a 4-drive RAID 6 could be going RAID10, depending on your I/O needs: If you do a lot of concurrent I/O as opposed to sequential reads/writes, you'll probably notice the extra write penalty in RAID6 compared to your current RAID5 setup, and a striped mirror setup could definitely help there. The drawback is that you can't easily go from RAID5 to RAID10.

Mikael H
  • 4,868
  • 2
  • 8
  • 15
  • Hey. The use case is very low I/O online file share storage, with a mixture of lots of little files, and lots of big files (ie, >1Gb each), it's also basically write-once, read often. I'd optimize for more resiliency - that's more important, than performance. So it sounds like RAID6 would not be a bad choice? Good point about the array being at a much higher chance of failure while rebuilding [potentially to RAID6]. – xconspirisist Feb 11 '19 at 19:42
  • In that case RAID6 should not be a bad choice. – Mikael H Feb 11 '19 at 19:46
2

Don't try to eck out any residual life from the dying disk. Put it out of its misery and replace it with a new disk, e.g. option 2.

doneal24
  • 561
  • 4
  • 10