[BBLISA] disk corruption recovery ideas?

[Douglas Alan]

Theo Van Dinter <felicity at kluge.net> wrote:

> On Tue, Oct 04, 2005 at 09:37:51PM -0400, Douglas Alan wrote:

> > I've had two drives die at the same time on a RAID.  Good thing the
> > RAID was only used as our backup server.  I'd never trust RAID again to
> > be any kind of security against disk failure.

> So have I, but I still swear by RAID (HW RAID if I can get it) on servers
> with any importance.  A quote from LISA a few years ago comes to mind:

> "Low probability events do happen, which is why people still play the
> lottery."		- Elizabeth Zwicky at LISA '99

As I see it, that low probability is multiplied by the fact that if your
RAID fails you lose a lot more data than you would have if you had just
had two disk drives fail.  They also don't protect you against
accidental "rm -rf /"'s and the like.  Consequently, I'd just rather
have a good backup system in place, than rely on a RAID for reliability.
(The backup system, though, might very well be rsyncing daily snapshots
to a RAID.)

This is assuming that you don't need 24x7x365 uptime, in which case it
might certainly make sense to have a RAID that allows you to hot-swap in
and out replacement drives.  If your server is that important, though, I
should think that you'd want to mirror disks (RAID 1), rather than RAID
5.

I think it likely that RAID often hinders server performance, by making
all your disk drives work in tandem, rather than allowing different
processes to peform independent i/o on different disk drives.

In addition to RAIDs for backup servers, we've also used RAIDs for
manipulating large amounts of data that wouldn't easily fit on a single
disk drive.  In these cases, though the data could be recreated, albeit
a bit painfully, if need be, in case of failure.

|>oug