[GRLUG] RAID levels

[Adam Tauno Williams]

On Mon, 2011-10-03 at 02:37 -0400, Mike Williams wrote:
> Do you think it's possible to construct a table of number of drives
> available versus recommended RAID level? Not that it would be an
> absolute bible, but a place to start.

There are many such tables, and sites, that allow you to calculate
storage at various RAID levels.
<http://www.ibeast.com/content/tools/RaidCalc/RaidCalc.asp>

> I'm thinking
> 2 drives -> RAID 1 (obviously)
> 3 drives -> RAID 1 + spare

Or a three way mirror.

> 4 drives -> RAID 5 + spare? RAID 6? RAID 10?

Or a pair of mirrors, or a three way mirror with a hot spare.

> 5 drives -> RAID 6

Or two mirrors with a hot-spare.

> 6+ drives -> RAID 6 + spare

Or two mirrors of three drives.

I've also seen mirrored RAID-5 sets [each of three drives].  But that
was before the advent of RAID-6.  Mirrored RAID volumes [even mirrored
sets of mirrors] present interesting performance.   One upside [not
related to this usage] is that a volume can be safely degraded and the
application continues running - still with redundant storage - while you
can one by one replace the drives [with faster or greater capacity
drives].  These days that would accomplished more often with at the SAN
or NAS, but it you have lots of DASD...

> Fairly often, like in the 5-drive box that started this, the question
> is whether a drive that will not add any storage space is better as a
> hot spare or as a live, already functioning member of the array. In
> the case of 3 drives, if you're really worried about the array going
> down, would it be better to run a 3-way mirror rather than a 2-way one
> with a hot spare? 

That is a consideration, but I'd typically go with a hot spare, unless I
needed trainloads if I/O.

> The only thing you gain with the hot spare option is a vastly
> different amount of wear on the spare since it isn't spinning until
> it's needed.

Yep.  Buy three *different* drives, configure two as a mirrored pair,
and the third as a hot spare.  At least order the drives from different
suppliers; increase your odds of getting different manufacture dates as
much as possible.

>  What's more likely: the 2nd half of a mirror failing before you get
> the data copied or 2 of 3 drives failing before you buy and install a
> replacement? I know statistics as well as the next guy, but these
> calculations get arcane fast!

I don't need statistics; as soon as one drive fails a second failure is
imminent until the volume is consistent again.  The best thing you can
do is make sure recovery begins as-soon-as-possible.  Maybe the
"statistics" say that two concurrent failures is a 1-in-a-million; my
experience supports O'Toole's corollary.

> On 10/02/2011 09:29 PM, Richard Nienhuis wrote: 
> > To bring this back to its actual application...  Its a personal
> > music server made from scrap parts.

In that case is RAID-6 even an option?  Software RAID (MD) doesn't
support RAID-6.

> >   It doesn't require 5 9's uptime of hands free operation.  The most
> > likely failure mode of this thing is that one day a drive will fail
> > on reboot/startup.  Hot spare isn't going to save it from a second
> > drive failure at startup. 

Failure of drives as start-up really isn't a RAID issue.  Although that
is a very likely time for drives to fail [on the other hand - the
tap-tap method gets at least 3 out of 4 stall-at-boot drives to spin
up].

Is this a solution that is going to be powered down frequently or left
running?

> > For its application integrity is a priority over everything else.
> > Downtime is a relatively inconsequential cost. 

A failed volume means 0% application integrity.



-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.