Ocfs2 users - Oct 2006 - lvm2 not cluster aware - okay, so how should I stripe my LUNs?

Hi, I'm testing ocfs2 in a 2-node ocfs2 cluster.  I have 12 SAN LUNs that
I'd like to stripe and mount up as an ocfs2 filesystem.  I've been told
that
because LVM2 is not cluster aware, that it can't be used.  I'm trying to
make mdadm work.  (using EMC powerpath for multipath).  I can create an md
raid0 device on one node using the /dev/emcpowerxx pseudo-devices, and on
the other node the md device is there, but it has scanned for the devices,
and it shows the /dev/sd* devices, not the pseudo-devices that really need
to be used for multi-path.  

 

Does anyone know how to make the other node only use the pseudo-devices for
multi-path.  And what is recommended for RAID in ocfs2?

 

Thanks.

 

-phil

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://oss.oracle.com/pipermail/ocfs2-users/attachments/20061004/c7cc7fa1/attachment.html

Humpherys Philip D Contr 532 CMMXS/CL wrote:
> Hi, I'm testing ocfs2 in a 2-node ocfs2 cluster.  I have 12 SAN LUNs
> that I'd like to stripe and mount up as an ocfs2 filesystem.  I've
been
> told that because LVM2 is not cluster aware, that it can't be used. 
I'm
> trying to make mdadm work.  (using EMC powerpath for multipath).  I can
> create an md raid0 device on one node using the /dev/emcpowerxx
> pseudo-devices, and on the other node the md device is there, but it has
> scanned for the devices, and it shows the /dev/sd* devices, not the
> pseudo-devices that really need to be used for multi-path.  
> 
>  
> 
> Does anyone know how to make the other node only use the pseudo-devices
> for multi-path.  And what is recommended for RAID in ocfs2?
I don't know about RAID0 striping over emcpower* devices, but you can
use EVMS clustered within Heartbeat 2.x to stripe LUNs using LVM2
metadata (that's what I do). But this will add additional point of
failures to your system.



Regards,
Fabio

On Wed, 2006-10-04 at 15:41 -0600, Humpherys Philip D Contr 532 CMMXS/CL
wrote:
> Hi, I'm testing ocfs2 in a 2-node ocfs2 cluster.  I have 12 SAN LUNs
> that I'd like to stripe and mount up as an ocfs2 filesystem.  I've
> been told that because LVM2 is not cluster aware, that it can't be
> used.  I'm trying to make mdadm work.  (using EMC powerpath for
> multipath).  I can create an md raid0 device on one node using
> the /dev/emcpowerxx pseudo-devices, and on the other node the md
> device is there, but it has scanned for the devices, and it shows
> the /dev/sd* devices, not the pseudo-devices that really need to be
> used for multi-path.  
I don't believe anything included in RHEL is supported across clusters
for software RAID.  This means LVM and MD are out of the picture.

Oracle only supports RAW devices, OCFS2 filesystems (one filesystem per
SAN LUN) or ASM.  They also support Red Hat GFS, but only if you use
external locking (GLM).  For a 4-node RAC cluster, you'd need 7 nodes (3
dedicated to locking).  This is because Oracle insists that none of the
RAC nodes participate in the GLM locking.  I believe Oracle may be
certifying DLM, the newer lock manager in GFS, but I'm not sure.

Either way, GFS is an added cost on top of RHEL, but it includes RHCS
and RHCS has cluster-aware LVM (I believe).

/Brian/

-- 
       Brian Long                             |       |
       IT Infrastructure                  . | | | . | | | .
       Data Center Systems                    '       '
       Cisco Enterprise Linux                 C I S C O

On 2006-10-04T15:41:47, Humpherys Philip D Contr 532 CMMXS/CL
<Phil.Humpherys@hill.af.mil> wrote:
> Hi, I'm testing ocfs2 in a 2-node ocfs2 cluster.  I have 12 SAN LUNs
that
> I'd like to stripe and mount up as an ocfs2 filesystem.  I've been
told that
> because LVM2 is not cluster aware, that it can't be used.  I'm
trying to
> make mdadm work.  (using EMC powerpath for multipath).  I can create an md
> raid0 device on one node using the /dev/emcpowerxx pseudo-devices, and on
> the other node the md device is there, but it has scanned for the devices,
> and it shows the /dev/sd* devices, not the pseudo-devices that really need
> to be used for multi-path.  
Software striping isn't such a great idea in the cluster. md's handling
of metadata isn't cluster safe either, just like LVM2/dm.

This is being developed, but not done yet.

The recommended RAID for clusters is implemented in the SAN.


Sincerely,
    Lars Marowsky-Br?e

-- 
High Availability & Clustering
SUSE Labs, Research and Development
SUSE LINUX Products GmbH - A Novell Business	 -- Charles Darwin
"Ignorance more frequently begets confidence than does knowledge"