Reminder -- RAID 5 is not your friend

[Derek Atkins]

Richard Pieri <richard.pieri-Re5JQEeQqe8AvxtiuMwx3w at public.gmane.org> writes:

> On Mar 11, 2010, at 1:20 PM, Derek Martin wrote:
>> 
>> The odds are much, much lower than that, though you need to figure in
>> MTBF and the time period over which you're concerned.  If your time
>> period is long enough, you have a 100% chance of losing your data.
>> Only not really... because you're a good sysadmin, and you do regular
>> back-ups, keep your important data in replicated revision control
>> systems, etc.
>
> True.  I should have written something like: I don't consider a 1 in 2
> chance of data lost *in the event of a second failure* to be
> "reliable" beyond the basic no single point of failure.

Actually it's not a 1 in 2 chance..  If you have a 4-drive system and
lose 1 drive then you have a 1 in 3 chance of a second drive failure
causing data loss.

If you want to change those odds you can make each RAID-1 a 3-drive
mirrors, so you require 6 drives for RAID-10, and it requires the right
3 drives to fail to cause data loss.

But the way to prevent loss here is to come up with a process to change
out the disks before they fail.

-derek
-- 
       Derek Atkins, SB '93 MIT EE, SM '95 MIT Media Laboratory
       Member, MIT Student Information Processing Board  (SIPB)
       URL: http://web.mit.edu/warlord/    PP-ASEL-IA     N1NWH
       warlord-DPNOqEs/LNQ at public.gmane.org                        PGP key available