[geeks] RAID0+1 and ideas, etc...

[Nick B]

Fusionio says "Don't mirror our cards, it's dumb", and in most cases they
are right.  Fairly frequently trying to mirror fusionio cards will result
in a significant performance drop, and they are pretty darn reliable.  Is
your use case so critical that the faint chance of data loss is a killer,
while the likely loss of performance is acceptable?  (then why do you have
fusionio cards?)

Man that was worthless.  I suspect the customer has spent a lot of money on
that third fusionio card, which will buy them next to nothing, but their
approach will work fine for what it is.
Nick


On Thu, Nov 7, 2013 at 11:29 AM, Sandwich Maker <adh at an.bradford.ma.us>wrote:

> " From: Jonathan Katz <jon at jonworld.com>
> "
> " All,
> "
> " Say you're running a Linux server. It has six Fusion IO cards, each
> " presents two disk devices giving you a total of 12 disk devices. You
> " want to RAID 1+0 this setup so that if any physical card or disk
> " device dies, you're still operational.
> "
> " My customer set this up using mdraid to stripe the two disk devices
> " within the individual cards. So we go from 12 disk devices to six md
> " stripes. Then the md stripes are mirrored card-by-card (so there are 3
> " md mirror/RAID1 devices.) Then LVM is used to stripe the three
> " remaining cards into one large volume that we use.
> "
> " I'm not sure this is the best way to do it, and I want to hear from
> " others who deal with this what they think (which is why I'm throwing
> " this out there.) The requirement is for one decent sized, reliable
> " storage pool of data. I think we have an extra layer of abstraction
> " here, but I'm not sure how to redo this, or if we even should.
> "
> " What say the mind trust?
>
> what disksuite does in that situation is
> a. mirror the drive pairs on each card
> b. stripe the 6 mirrors
> thus allowing you to survive any combination of 1-disk-per-card
> failures.  it does this even if you try to configure 2 stripes and
> mirror them, and it makes sense - with mirrored stripes, one failure
> on each side would take you down.
>
> i would mirror first, then lvm the mirrors.  it would make sense to
> mirror pairs across cards, to provide card-failure hardness.
> ________________________________________________________________________
> Andrew Hay                                  the genius nature
> internet rambler                            is to see what all have seen
> adh at an.bradford.ma.us                       and think what none thought
> _______________________________________________
> GEEKS:  http://www.sunhelp.org/mailman/listinfo/geeks