Common LUN for shared application between nodes of cluster

[Srikanth Arunachalam]

Hi,

We have a requirement that we need to place our application on shared LUN on EVA6K that is available to all nodes of the cluster (2-node serviceguard cluster). As far as the Oracle db is concerned we install them on induvidual LUNS zoned specific to each node. However, the application executables need to be installed on common LUN of EVA6K storage. Please let me know the possiblity of having application in shared LUN. How to configure the same? I understand the disadvantage of having application in shared LUN/disk between 2-nodes. But, please let me know if this is feasible in the first place and steps required to be configured as package.

[Chan 007]

Hi Srikanth,

Why can't you try a NFS/CIFS solution for this application LUN ( I mean export from one server) and mount on the next one.

We are using similar requirements solve by this NFS, but I don't know you code/app requirements may be very specific

Chan

[Warren_9]

hi,

as chan suggestion, NFS is one of the approach and no extra cost if you already have the toolkit installed.

another solution is install the RAC option which will let you to mount activate the VG in share mode (concurrent access)... of course it will need extra cost.

GOOD LUCK!

[Steven E. Protter]

Shalom Srikanth,

Yes you can do this. Depending on the application, you can't have both nodes using the shared storage at the same time.

Most installations install the binaries local and put the data on a shared LUN.

Example: Oracle RAC can handle shared simultaneous data mounting. Oracle server will not tolerate this setup.

SEP

[Srikanth Arunachalam]

Thanks everybody for your response. I will explore on service guard for NFS setup, as we have not purchased Oracle RAC for HP-UX. Actually the oracle binaries are in induvidual LUNs only. The Oracle database data are in shared LUN. For the convinience of the system administrator's style of performing upgradation..they have requested application binaries to be placed on common shared LUN of EVA6K.

[A. Clay Stephenson]

Actually this confguration is perfectly reasonable and NFS is not required. It is perfectly legal and supported to mount the filesystem containing the executables as long as ALL systems mount the filesystem read-only. When you need to update/patch the executables, you would have to umount the filesystem from all nodes and then mount it read-write on one system, do the update/patch, and then umount and mount read-only again.

[A. Clay Stephenson]

and a very neat way to mount it read-write on all nodes is the Cluster File System (CFS). That requires HP-UX 11.23 with MC/SG eRAC but it is a method to share database applications and data across all your nodes.

[PeterWolfe]

>and a very neat way to mount it read-write
>on all nodes is the Cluster File System (CFS).
>That requires HP-UX 11.23 with MC/SG eRAC

CFS is indeed a neat way to do this and
you can get CFS independently of Oracle.

The relevant SG product bundles are:

Bundle 5: HP Serviceguard Cluster File System (SGCFS) - T2775BA
Bundle 6: HP Serviceguard Cluster File System for Oracle (SGCFSO) - T2776BA
Bundle 7: HP Serviceguard Cluster File System for RAC (SGCFSRAC) - T2777BA

Bundle 5 is CFS without Oracle.

Here's a good summary of the Serviceguard
Storage Management Suite offerings (has
the decoder ring for what is included
in each bundle):

http://docs.hp.com/en/7413/SGStorageMgmtSuite_WP.pdf

[Srikanth Arunachalam]

Hi,

Thanks for your response. Unfortunately we do not have CFS product purchased for our systems. I would like to put down the configuration details below, please let me know if application binaries can be setup in seperate shared LUNS across nodes using NFS as solution.
(1) We have Hp-UX 11.23, pair of nodes, with MC service guard installed in them.
(2) Package contains Oracle, application accessing oracle and other performance monitoring tools.
(3) Oracle installed on zoned induvidual LUNS accessible seperately on each system.
(4) Oracle data setup in shared LUNS accessible across nodes of cluster.
(5) VxVM volume manager and VxFS type file system is used for Oracle data.

If NFS is a solution, please advice how do we setit up in package scripts (single package contains both Oracle in it and the application binaries accessing the Oracle).

Thanks in advance for your help.

[Chan 007 · Accepted solution]

Hi,

So far you have not assigned any points for any of your questions. All forum members are helping you in solving your problem, as a reward if you help them by assigning points there will be more to help.

"I have assigned points to 0 of 6 responses to my questions. "

You can script this as below in the primary package, will be pact of you package start/stop.

edit /etc/exports and add your File system that will be exported, to which all system, waht access,
e.g
/home -access=server2

This script will run after oracle/application starts on the primary server. Then the NFS will start. same way you should stop it first and the shut oracle.


FUNCTION=$1

case $FUNCTION in
start_msg) echo "Start FileSystem Export" ;;

stop_msg) echo "Stop FileSystem Export" ;;

'start') /usr/sbin/exportfs -vua
/usr/sbin/exportfs -va
for host in server2 server3
do
remsh $host mount 2>/dev/null
done

exit 0 ;;

'stop') for host in server2 server 3
do
remsh $host umount 2>/dev/null
done
/usr/sbin/exportfs -vua
exit 0 ;;

*) echo "usage: $0 {start|stop}"
exit 1 ;;
esac

exit 0

[Srikanth Arunachalam]

Thanks Chan, Its a good details for me.