[Dovecot] RAID1+md concat+XFS as mailstorage

[Stan Hoeppner]

On 6/30/2012 6:17 AM, Костырев Александр Алексеевич wrote:
> So, you say that one should use this configuration in production with
> hope that such failure would never happen?

No, I'm saying you are trolling.  A concat of RAID1 pairs has
reliability identical to RAID10.  I don't see you ripping a mirror pair
from a RAID10 array and saying RAID10 sucks.  Your argument has several
flaws.

In a production environment, a dead drive will be replaced and rebuilt
before the partner fails.  In a production environment, the mirror pairs
will be duplexed across two SAS/SATA controllers.

Duplexing the mirrors makes a concat/RAID1, and a properly configured
RAID10, inherently more reliable than RAID5 or RAID6, which simply can't
be protected against controller failure.

By stating the concat/RAID1 configuration is unreliable simply shows
your ignorance of storage system design and operation.

-- 
Stan


> -----Original Message-----
> From: dovecot-bounces at dovecot.org [mailto:dovecot-bounces at dovecot.org] On Behalf Of Stan Hoeppner
> Sent: Saturday, June 30, 2012 4:24 PM
> To: dovecot at dovecot.org
> Subject: Re: [Dovecot] RAID1+md concat+XFS as mailstorage
> 
> On 6/28/2012 7:15 AM, Ed W wrote:
>> On 28/06/2012 13:01, Костырев Александр Алексеевич wrote:
> 
>>> somewhere in maillist I've seen RAID1+md concat+XFS being promoted as
>>> mailstorage.
>>> Does anybody in here actually use this setup?
>>>
>>> I've decided to give it a try,
>>> but ended up with not being able to recover any data off survived
>>> pairs from linear array when _the_first of raid1 pairs got down.
> 
> The failure of the RAID1 pair was due to an intentional breakage test.
> Your testing methodology was severely flawed.  The result is the correct
> expected behavior of your test methodology.  Proper testing will yield a
> different result.
> 
> One should not be surprised that something breaks when he intentionally
> attempts to break it.
> 
>> This is the configuration endorsed by Stan Hoeppner.
> 
> Yes.  It works very well for metadata heavy workloads, i.e. maildir.
>