[Leica] RAID controllers?

[hartzell at alerce.com (George Hartzell)]

Spencer Cheng writes:
 > A bit late but here is my experience.
 > [...]
 > I agree with those who says the OS should not be part of the RAID  
 > array if for no other reasons other than that RAID5 recovery can take  
 > a long, long, long time. If you have a deadline, the last thing you  
 > want to do is to sit there and wait, and wait, and wait, and pray that  
 > the OS can be recovered so you can actually boot your PC....

I agree with Spencer, RAID is not a replacement for separate backups,
preferably taken on a regular basis and stored offsite.  Add
catastrophic hardware failures to the list of coffee, soda, fire, and
theft.

But I guess that I now disagree with Spencer _and_ Gary.  Hopefully
this doesn't make me disagreeable....

If you have a redundant disk setup, RAID 1 or RAID 5 or RAID 6 or...,
and you lose a disk, that volume should continue work, albeit possibly
more slowly.  That's what RAID's do.  There's no mystery about it, no
praying, and no waiting for it to recover.  When you replace the
failed drive the system will spend a *lot* of it's time "resilvering"
the mirror aka rebuilding the array, during which time performance
will _suck_ but the volume will still be available.

It's not as if it's gone away and you're hoping that the recovery
process will somehow magically bring it back.  All of your data is
still there and still available.  No praying involved.  If you don't
believe that your RAID can survive the loss of a single disk, you
probably haven't played with it enough and I'm not sure what's it's
giving you.

If you're thinking about setting up a RAID, you really owe it to
yourself to experiment with it before you have all kinds of data on
it.  Read the manual.  Set it up.  Read the manual.  Power down and
disconnect a drive.  Reboot and see what happens.  Read the manual.
Power down and reconnect the drive.  What do you need to do to
reintegrate the drive?  Read the manual.  Disconnect two drives.
Etc....  Once you're comfortable with it as a tool, then you can put
your valuable data onto it.

Various RAID strategies offer varying degrees of redundancy.  A
two-disk mirror can survive the failure of one disk without losing any
data.  A three-way mirror can survive the loss of of two disks.  A
four-way mirror could survive the loss of three.  A simple RAID 3 or
RAID 5 can survive the loss of one disk, lose two and the whole
thing's toast.  There are various flavors of RAID 5 that offer more
redundancy, called things like RAID 6 and RAIDZ2 and..., and which can
survive the loss of multiple disks.

Just as when you have a disk die in a single disk setup and you can
sometimes peel some/most of the data off of it, you can sometimes peel
some/most of the data off of a completely FUBAR'ed RAID.  But that's
not the same thing that happens when they lose a single disk.

You should think about the risk of two disks failing in your
situation.  You might think that disk failure is a long shot and that
two failures is well nigh impossible.  On the other hand your first
disk might have failed because you keep the machine in the closet so
you don't have to listen to the fans and you've overheated it.  That
could make your second disk more likely to fail too.  Or maybe you
bought both of the disks at the same time, they came out of the same
box, and the FedEx guy dropped it on the way to your front door.  Or
they were both made on the Monday morning following Mardi Gras.  If
you want to see some real disk failure numbers w/out marketing crap,
check out this study the Google crew did.

  http://labs.google.com/papers/disk_failures.html

There are lots of solid reasons to keep some data on a separate hunk
of storage from other data (and the OS is really just another hunk of
data).  Performance.  Manageability (you don't want the fact that your
kid filled up his hunk of the disk with mp3's and videos to keep you
from being able to work with your images).  Move-ability (you'd like
to be able to take the hunk with you somewhere else w/out lobotimizing
the machine.  Sometimes you handle this by using a separate disk
(either a real one or a virtual one constructed from a RAID).  Other
times you partition a real/virtual disk into hunks and use them
accordingly.  One or more of these reasons might encourage you to put
your OS on a separate disk, but they don't otherwise mean that you
shouldn't keep everything together.

If you have a mac that can hold multiple disks, and you're willing to
put two disks into it, I'd say that you really should set them up in a
software RAID 1 and that you should put the entire kit-and-caboodle on
that RAID.  My money's where my mouth is, it's how I run my machine.

If you can put more disks into it, then it's a more complicated
decision.  Meet me in Oakland, CA and buy me a beer and I'll babble
about it until the beer's gone....

Sheesh.  When did I get so long-winded????

g.