[F7] Most reliable RAID level

[Chris Tyler]

On Wed, 27 Jun 2007 15:06:31 +0200 Stuart Murray-Smith wrote: 
> 
> What is the preferred/most reliable level to implement software RAID
> on a Fedora 7 server?

Some thoughts to add to what has already been said:

- Raid 1 is expensive on a cost-per-gigabyte basis. Although it's
usually used in a two-bank configuration, there is no reason why you
have to stop at two copies of your data -- you could have three or four
copies if you wanted. If the disks are on separate controllers, you can
read and write them in parallel, yielding good performance (1x write, Nx
read).

- Raid 5 provides a good cost-per-gigabyte, because only one drive is
storing redundant information for each stripe. The write performance is
worse than raid-1 and worse than a single drive, but the read
performance is somewhere between a single drive and raid-1 (you can read
in parallel, but not every drive has every block).

- Raid 6 is like raid 5 with extra paranoia.

But in all of this, a few things to consider:

- You'll want to use software raid managed by the kernel in all but the
most extreme cases (where it may make sense to use hardware raid). The
CPU overhead is minimal on a modern system. Avoid motherboard raid,
since it is in most cases a type of software raid that's tied to the
motherboard BIOS. Using the kernel raid facilities gives you maximum
flexibility in the case of a hardware failure (and resulting crisis):
you won't need a source a particular motherboard model and software
version or a special disk controller to recover your data.

- If you're really paranoid, buy different brands or at least different
batches of disks. There have been cases where several disks of the same
brand fail within a very short time of each other (see for example
http://tinyurl.com/34rrlo ) which can lead to irrecoverable data loss.

- Hot spares enable a drive array to be rebuilt without your
intervention.

- Don't forget power supplies and controllers as possible points of
failure. Power supplies are particularly bad because a failure can take
out multiple drives plus the controller(s).

- Obviously raid is no substitute for backup, because it doesn't protect
against lightning strikes, fire, server theft, data corruption due to
bugs, or user stupidity.

--
Chris Tyler
http://dailypackage.fedorabook.com