[GRLUG] RAID levels

[Adam Tauno Williams]

On Fri, 2011-09-30 at 18:42 -0400, Philip J. Robar wrote:
> On Sep 30, 2011, at 4:59 PM, Mike Williams wrote:
> > Last night at The Warehouse we had a fairly involved discussion about the various RAID levels.
> Hopefully someone brought up the RAID 5 Write Hole problem:
> http://blogs.oracle.com/bonwick/entry/raid_z
> You don’t mention how large the drives being used are, but for terabyte and larger drives the chances of a RAID 5 array experiencing a fatal error, with an accompanying loss of all data, during a rebuild are uncomfortably high: 

To become a BAARF member visit <http://www.baarf.com/>

"B"attle "A"gainst "A"ny "R"AID "F".

The primary purpose of RAID 5 is to get System Administrators fired.

RAID-5 *is* an OK solution for providing a modicrum of redundancy for
data where (a) you are OK with a bit of downtime and (b) you have a
*current* off-line backup.  That is - for storing primarily static data.

Personally I always push hard for plain old boring RAID1 [mirroring].
It is simple, fast, and no fancy footwork involved.  When you swap a
drive in a mirrored set you know exactly what data you're risking and
what is going to happen to rebuild the volume.

If mirroring doesn't provide enough capacity RAID6 is better.

But, never in any case, install any RAID configuration without a
hot-spare.  If you need the data available you must have a hot-spare.
Odds are you simply will not know a drive failed [Wow, how I have seen
these really cool monitoring system fall flat on their faces in
fascinating ways],  you just won't physically be there to replace the
drive, or you just won't have a drive around of the type you need [if
you have a cold-spare - mount it right in the server case if possible or
store it in some box that can't be easily moved - if you don't even have
a cold-spare - the RAID configuration is pointless].

Due to O'Toole's corollary a single failure dramatically raises the
probability of a 'coincidental' second failure.


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