Any other DU sysadmins? RAID advice request/rant

[Recursion]

OK, today was the fifth time in ten years that a RAID controller has simply died, caput, destroying basically all the stripes on it.

Allow me to make an heretical suggestion: RAID is a waste of money. Take whatever money you were going to spend on a RAID controller and amp up your backup system.

Anybody here tried that? Am I overlooking something?

[pokerfan]

1. Someone explain RAID for the average user for me

If your application is so mission critical that uptime is a primary factor then RAID 1 might make sense and I've worked on systems (military) that require this. But it's no replacement for an off-site backup as obviously both drives must be local. Yeah, it's a pain in the butt recovering from a drive failure but honestly, how often does that happen? So once every few years I might spend some time restoring a system from backup. Big deal.

As far as performance goes, RAID 0 can add some performance but at the expense of reliability. A nested RAID can offset that reliability hit but that requires a minimum of four drives.

I honestly don't get it. :shrug:

[Recursion]

2. Me too!

And basically all of my work has been military. Seriously, it's a scam. RAID 0 adds some performance at the expense of redundancy (it's "RAID", sure, but RAID 0 is actually negatively redundant).

Take the RAID controller money and buy a better tape/optical backup system.

[pokerfan]

3. Yep

The BIOS on this machine nags me on cold-boot with a 6 second prompt to configure RAID. Pain in the ass and I have dug through the all the settings and apparently there's no way to get rid of it short of re-flashing. Fortunately, since I'm running Linus, I don't have to boot that often. :) Takes 6 seconds off of what should be about a 30 second boot. :banghead:

I agree. Spend the money that you would have used on a redundant drive and buy an external drive that you can use for off-site backups. With larger and faster drives/interfaces (SATA 3 is 2X SATA 2) I just don't see any need for it these days.

[HopeHoops]

4. I've had excellent luck with RAID-5, but it takes some configuring.

Specifically, you really don't have a reliable volume unless each of the physical disks is on a separate controller. The idea behind RAID-5 is good - the volume can suffer a physical drive loss and keep running without data loss. Recovery generally just involves putting in a replacement drive and telling it to "get the hell to work rebuilding the damn thing", the complexity of which varies from system to system. But if all of the disks aren't on individual controllers, the loss of a controller is equivalent to a multiple disk failure and the entire volume is hosed. That isn't to say that it can't be recovered, but it sure as hell isn't going to keep running while you wait for a new controller to show up.

For example, a 30 disk unit would likely have six controllers. A good configuration would be to throw five RAID volumes of five disks each spread over five of the controllers. Even if a controller dies, all five volumes will still have four disks each on functional controllers and will continue operating as if nothing had happened, other than a slight but noticeable decrease in performance as it has to reconstruct the parity stripe or use it to reconstruct the data stripe for the missing disks during I/O operations. On the sixth controller, make one of the disks a hot-swap disk and use the other four as two two-disk volumes mirrored. They will have no protection in the event of controller failure, but the mirroring will provide data redundancy.

There are a lot of ways to configure RAID-5, but again, unless each physical disk is on a different controller you've got a major point of complete failure waiting to happen. If you've only got one controller, you might as well go with mirroring. You won't get the same potential disk capacity as RAID-5 but performance will be better and recovery from a hardware failure is generally less complicated. In the scenario I described above, recovery on the RAID-5 volumes was painless - and I dealt with a shitload of failed disks as well as a few controllers. There was one major all-nighter, but that was caused by a technician royally fucking up a drive replacement - I was pissed. For PC systems with internal drives, I just stick with mirroring.

On another note, USB external drives are relatively inexpensive now and your best safety comes from regular backups. My development box has two 1.5TB internal drives (mirrored) and six 1TB externals, 3 on one hub for general purpose storage and 3 on another hub for backups. The backup software is configured for nightly backups and I regularly make manual copies of really important directories. I use DVDs for long-term archives. I'm not as consistent with backups of the laptops (6 of which are in regular use) and really just rely on manual copying of directories to an external drive. I like having at least two copies of everything important anyway.

Last point: if you do go with RAID-5, I don't recommend striping multiple volumes over the same physical disks since serious thrashing is likely to occur unless the volumes are accessed almost exclusively at different times of the day. The performance hit generally isn't worth whatever potential benefit there may be.