Dave Hall wrote:
> You are correct, you can't "un-RAID" a RAID-0.
>
> Also, I'd like to debunk the suggestion that a RAID-0 is no more risky than a
> single drive
>
> With a single drive, if the disc fails you lose all your data. With a RAID-0, if
> either drive fails, you lose your data. The probability of data loss doubles.
> Think of it like a game of Russian roulette where each drive is a bullet. The
> chance of getting shot doubles when when you put two bullets in the revolver,
> the probability of getting shot doubles.
>
hmmmm... I understand what you are saying, and I'm NOT a statistician,
but would it?? - in that would it really matter - (assuming each drive
has the same MTBF type stats) if either drive dies, then effectively,
as far as the computer/OS/etc are concerned it's the same as if you had
only 1 drive and that 1 drive died, as since they are raid 0 they are
'inseperable'.
Let's try to word this differently as that doesn't quite say what I am
pondering.
Let's say you had only 1 drive, and it's gonna fail at 77,777 hours
(pulling number out of air and in this example assuming no backups and
no recovery options etc) - well that's when you are SOL.
In 2nd example you had another drive (could be identical but doesn't
matter) and you have same machine but with 2 drives in raid 0 and the
same failure hours for the 1st drive as in example 1 above. Now, drive 2
is gonna fail at 78,585 hours, but it doesn't matter since drive 1
already failed before that, when it was gonna anyway in a single drive
setup, so you are still SOL at 77,777 hours, so there isn't any
difference in failure time
In 3rd example you had another drive (could be identical but doesn't
matter) and you have same machine but with 2 drives in raid 0 and the
same failure hours for the 1st drive as in example 1 above. Now, drive 2
is gonna fail at 76,555 hours, so again that's when you are SOL, this
time 1222 hours of use before you would've with only the 'original' drive 1.
Since one does not know when a drive is gonna quit (and probably would
want to have the raid 10/50 setup as you mentioned to protect against
that, but this is only talking about raid 0 anyway) you aren't really
taking any more of a chance unless one KNEW for sure that the added 2nd
drive to the raid 0 setup is gonna fail BEFORE the 1st drive, but since
one can't/doesn't know that (or if one did then backup/replace before
that) it is no more risk than just one drive.
Of course one is paying the cost of the 2nd drive in exchange for faster
performance.
If the probability of loss doubles with 2 drives (and then triples with
3, etc if there were such a thing) then wouldn't that imply that the
'2nd' drive has a detrimental effect on the 1st drive (or vice versa,
but in any case doesn't that suggest some kind of interoperability
relationship - but these are 2 seperate independant (as in
R-A-Independant-D) pieces of hardware - sure a PSU problem could take
out either or both, but again that problem could've happened to the
only/single drive in a system also).
<smartalec mode on>
And if it doubles, can I just use 1/2 a drive or say, take a drive with
3 platters and remove 2 platters to lessen the probability of
failure/loss - effectively doubling or tripling the probability of NOT
having a loss/failure. :-) {gets into garage with tools, asks Mark
for help on opening drive.....}. And with the gun/bullet russian
roulette example, since they ARE designed to do damage - eg fail (rather
than to try to NOT fail, as a harddrive would be trying to do), wouldn't
it be that they are working better, like one would want a harddrive to
do, when the bullets do go off. Wouldn't it be more like a harddrive
failure when/if they didn't go off?
<smartalec mode off>
> RAID-0 is purely for improving performance by spreading reads and writes over
> two spindles. RAID-10 or 50 stripes two arrays so it is resistant to a single
> drive failure in either or both RAID-1 or RAID-5 arrays while compensating for
> the slower performance of the RAID-1/5. The RAID is not for disaster prevention
> but minimizing unscheduled downtime. It's not a substitute for backups.
>
> As an acid test, if you can't justify spending money on a big UPS and/or a
> generator, then RAID is probably not necessary for your use case.
>
:-) I have both, but want to sell the generator.
Received on Mon Jun 30 02:38:13 2008
This archive was generated by hypermail 2.1.8 : Mon Jun 30 2008 - 02:38:16 CST