Tilted Forum Project Discussion Community - View Single Post

Kadin · 09-07-2010, 10:51 PM

I recently had a RAID-5 setup go south on me. It was the typical failure mode for these things. One drive failed, out of four in the array, I swapped in a new one, and while it was rebuilding one of the other 3 in the array failed.

One failed drive is tolerable, because it can calculate the missing chunks using parity calculations that it sprinkles around on the other drives. But with two drives down, you're S.O.L.

When that second drive failed, I knew I was hosed. Luckily I had backups and wasn't depending on the RAID system, so nothing of value lost, but restoring everything was an unpleasant way to spend a weekend.

Just a cautionary tale. A lot of people set up RAID arrays with several identical (same manufacturer, model, purchased at the same time) drives. If you do that, there is a fairly good risk that more than one of them might decide to conk out at the same time, or close enough to it so that you might be rebuilding during the second failure. In RAID-5, that means badness. In RAID-10 (stripes across mirrors) or 0+1, you have a 1 in 3 (maybe 1 in 2?) shot of losing your data with two drives out, depending on which two go.

Definitely not a replacement for a good backup strategy. Reduces you risk of downtime, for sure, but not a backup system.

And if you do decide to go with RAID, particularly RAID-5, do yourself a favor and buy a mix of drives (same capacity and RPM, obviously), different manufacturers or at least models. It's not like buying tires for your car, where you want them to wear at the same rate; you very much *don't* want that. You want them to all fail at different times, and the only way I know how to do that (aside from retiring them prior to failure after a certain number of hours in service, which is something to strongly consider!) is to go for diversity.

09-07-2010, 10:51 PM	#12 (permalink)
Kadin Upright Location: Virginia	I recently had a RAID-5 setup go south on me. It was the typical failure mode for these things. One drive failed, out of four in the array, I swapped in a new one, and while it was rebuilding one of the other 3 in the array failed. One failed drive is tolerable, because it can calculate the missing chunks using parity calculations that it sprinkles around on the other drives. But with two drives down, you're S.O.L. When that second drive failed, I knew I was hosed. Luckily I had backups and wasn't depending on the RAID system, so nothing of value lost, but restoring everything was an unpleasant way to spend a weekend. Just a cautionary tale. A lot of people set up RAID arrays with several identical (same manufacturer, model, purchased at the same time) drives. If you do that, there is a fairly good risk that more than one of them might decide to conk out at the same time, or close enough to it so that you might be rebuilding during the second failure. In RAID-5, that means badness. In RAID-10 (stripes across mirrors) or 0+1, you have a 1 in 3 (maybe 1 in 2?) shot of losing your data with two drives out, depending on which two go. Definitely not a replacement for a good backup strategy. Reduces you risk of downtime, for sure, but not a backup system. And if you do decide to go with RAID, particularly RAID-5, do yourself a favor and buy a mix of drives (same capacity and RPM, obviously), different manufacturers or at least models. It's not like buying tires for your car, where you want them to wear at the same rate; you very much don't want that. You want them to all fail at different times, and the only way I know how to do that (aside from retiring them prior to failure after a certain number of hours in service, which is something to strongly consider!) is to go for diversity. Last edited by Kadin; 09-07-2010 at 10:56 PM..