Cluster attached storage to linux .... anyone? HP MSA 1000 for example.

[Lennart Sorensen]

On Sun, Apr 01, 2007 at 05:07:27PM -0400, tleslie wrote:
> I am thinking of buying a HP MSA1000 storage array,
> with a 2gB fiber channel (standard)
>
> if I put the 2gB fiber card in any given linux box,
> and assuming thats all fine ....

Sounds awfully expensive.

> do the drivers just make the MSA1000 appear as a device in  /dev
> no different then /dev/sda or dev/hda .... ????

Fibre channel just has devices appear as scsi disks.  What device
provides the actual fibre channel storage is not relevant (just like
firewire disks all just appear as storage to firewire).

> and if so ...
> how do i specify (get at) a particular unit/partition
> on the MSA, given i could have umpteen of them.

Each unit will be assigned a scsi device, with a serial number that
should make some sense for the device.  The driver should also have some
way to show which unit number is which scsi LUN.

> i read one article way back about some firewire cluster setup
> and it did talk about the drivers making it as easy as a  /dev ref. to
> get at the storage, and in that case it was easy because it was a single
> drive system so i'd imagine it was just /dev/xxa1  /dev/xxa2  for the
> different partitions on the drive array setup.
> But for the MSA1000 (MSA 500 or 1500 for that matter),
> you can have a shit load of defined drives sets and partitions.

Well linux does have a limit to how many scsi devices it will support,
although I think there are kernel patches to change that.  It has been a
few years since I used fibre channel and a SAN.  In that case we had
dual controllers and redundant paths to the SAN and ended up with 4
copies of each device (since each card had a hostcontroller number, and
then each path had an ID, and each LUN had an id, giving 4 ways to reach
any given LUN).  We then used the md drivers to run them in failover
mode so that it would detect lost connections to a LUN and switch to an
alternate path to that device.  Of course it means 4 scsi devices per
LUN and then an md device for each real LUN.

> anyone got any insight?
> 
> I'd hope i could just see the partitions over the arrays as  /dev/xxxYn
> format, but why am i thinking it might not be that easy.
> And then what about if you are addressing multiple MSA devices ?

Well you can partition each volume normally as sdXY with up to 14
partitions (3 primary and 11 logical, or 4 primary) per scsi device.

> Maybe there is some translation layer config file that has to be set up
> or something.
> 
> doesn't even have to be a HP insight, a DELL/IBM/SUN experience might be
> just as valuable.

Well my experience was with a qlogic 2200 (as far as I remember)
connected to an IBM shark or something like that.

If I was in charge of things I would NEVER consider spending money on
fibrechannel.  SAS looks like a much better idea, and I expect it to
kill of the fibre channel market over the next few years.  Personally I
will just stick with SATA for most things though.

--
Len Sorensen
--
The Toronto Linux Users Group.      Meetings: http://gtalug.org/
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists