topic Re: Shared filesystem for multinodes in Operating System - HP-UX

Shared filesystem for multinodes

Abid Iqbal — Mon, 27 Dec 2010 11:39:03 GMT

There are six servers running HP-UX 11.31 DCOE.
And four mount points (file system) are on one server.
How to share these four mount points between all six servers so that these are accessible to application on all six nodes for read write operations equally.
Oracle is not being used so SgeRAC is not the option.
Please share any reference document.

Re: Shared filesystem for multinodes

Shibin_2 — Mon, 27 Dec 2010 11:54:11 GMT

Use NFS share.

Add the entries in /etc/dfs/dfstab

then do shareall command.

Re: Shared filesystem for multinodes

Abid Iqbal — Mon, 27 Dec 2010 12:06:51 GMT

Thanks for the reply.
NFS is not the option. Some servicegaurd solutions......?

Re: Shared filesystem for multinodes

R.O. — Mon, 27 Dec 2010 12:36:56 GMT

You can create a nfs serviceguard package...

Re: Shared filesystem for multinodes

Emil Velez — Mon, 27 Dec 2010 23:34:37 GMT

HP cluster file System (CFS). Is a viable option for this. It is a purchased product. Yes it is used for Oracle RAC but it can be used without it too. It is easy to setup using Cluster VxVM and allows all nodes to read and write the local disks from any node.

There is no separate locking api so the last one that saves data will write so you need to make sure that the systems do not step on each other (oracle takes care of it when using RAC) but it is a good option.

Why not NFS. NFS is a great protocol and can deal with locking of files since one is the nfs server and all requests go through that node. IF that node fails the NFS package starts on another node. The cross mount capabilities of the NFS package allow the package to be mounted locally and on all of the nodes on the same mount point.

Re: Shared filesystem for multinodes

Abid Iqbal — Tue, 28 Dec 2010 12:01:55 GMT

Thanks Emil,
With out locking management, it would be difficult to use CFS as there is a great chance of data loss. The servers are running a critical customer application.
NFS is not because the operations require faster performance.
Any other suggestions?

Re: Shared filesystem for multinodes

Dennis Handly — Tue, 28 Dec 2010 14:42:06 GMT

>Without locking management, it would be difficult to use CFS as there is a great chance of data loss.

What locking API is the application currently using?

Re: Shared filesystem for multinodes

DeafFrog — Wed, 29 Dec 2010 04:33:45 GMT

Hi ,

As Emil already conveyed -Application should be intelligent enough to handel and understand that underlying filesystem is a cluster filesystem , and not a usual one ...why don't you discuss this first with the application support , just a thought.

Re: Shared filesystem for multinodes

Abid Iqbal — Wed, 29 Dec 2010 05:09:04 GMT

>What locking API is the application currently using?
No locking API is available in application currently.

Re: Shared filesystem for multinodes

Dennis Handly — Wed, 29 Dec 2010 05:17:42 GMT

>No locking API is available in application currently.

I'm confused. What do you mean by this? The FS doesn't support locking so the application doesn't bother?

If so, how does the application work for a local filesystem, single threaded/process?

Re: Shared filesystem for multinodes

Abid Iqbal — Wed, 29 Dec 2010 06:02:55 GMT

Means no locking API for CFS.
It is working on local file system on single server.

Re: Shared filesystem for multinodes

Dennis Handly — Wed, 29 Dec 2010 06:16:51 GMT

>Means no locking API for CFS.

I suppose this more accurately means that calls to lockf(2) and fcntl(2) are ignored?

Re: Shared filesystem for multinodes

Abid Iqbal — Wed, 29 Dec 2010 07:25:20 GMT

Yes Right.
Final conclusion, CFS is not the solution for this?
Is it confirmed?

Re: Shared filesystem for multinodes

Dennis Handly — Wed, 29 Dec 2010 12:33:33 GMT

>CFS is not the solution for this?

It doesn't seem like it. It seems all CFS does is to synchronize filesystem metadata but not user data.

Re: Shared filesystem for multinodes

Abid Iqbal — Mon, 03 Jan 2011 10:29:12 GMT

Thank you all for the support and help.