sharing the volume group and mount point

[sheevm]

Hi,

Is it possible to share the same SAN area actively between two servers?

For example:

server A: VG01, LVOL1 /DATA
server B: VG01, LVOL1 /DATA

Both uses the same SAN area.

Can I mount it on both server? Can two servers write to the same filesystems?

Please help.

Thanks.
Raji

[A. Clay Stephenson]

Yes but with an extremely big if. If you are running HP-UX 11.23 and if you are running CFS (Clustered File System) then n nodes can mount a shared, fully-cooked filesystem and have full read-write access from every node. To each node, it simply appears as a conventional vxfs filesystem.
CFS also requires a dedicated network for its traffic.

If the above conditions are not met then your options with a fully-cooked filesystem is mounted read-write on one node and NFS mounted on all others. If you must have read-write access w/o NFS then your only option is raw i/o and the application is solely responsible for contention resolution. You can mount the filesystem read-only on ALL nodes.

The fundamental problem is buffer-cache consistency across all nodes. Unless CFS is used, Node A won't have a clue what Node B's buffer cache looks like and absolute chaos is guaranteed.

[sheevm]

Clay,

What about CIFS product? I understand it comes with HP-UX 11i v2 O.S and it is HP supported.

Does anyone has experience with this tool? Any comments, tips and documentation link would be helpful

Thanks.
Rajim

[A. Clay Stephenson]

CIFS is Samba (essentially NFS for PC's); it is not even remotely similar to CFS.

[sheevm]

Clay,

Can we use CIFS to share the filesystem between three HP nodes? Management is thinking this as an alternative to NFS.

If we do this what are the advantages and disadvantages?

Thanks
Rajim

[A. Clay Stephenson]

Yes, but is is dumb. If you are sharing files between UNIX boxes then NFS is the better choice because it is a native UNIX mechanism. Of course, if you want to emulate an extremely advanced, never-known-to-fail, friendly, intuitively-obvious, easy-to-use, troublefree operating system popular in many parts of the world (especially on desktops) then by all means use CIFS/Samba to share files between UNIX boxes.

CIFS really only makes sense when there is a need for Windows boxes and UNIX boxes to share the same files. Even then, NFS and CIFS are not mutually exclusive.

[Dave Olker]

I'm not Clay, but I have some opinions about sharing filesystems among HP-UX nodes using CIFS vs. NFS. :)

My preference would always be to use NFS when sharing filesystems among UNIX systems that do not have direct physical access to the filesystem. If all three nodes have a direct connection to the storage then a clustered filesystem is a viable alternative. That would require running VxCFS with ServiceGuard.

If the choices are between sharing the filesystem with NFS or CIFS (in this case it would be the server running CIFS server and the clients using the CIFS client) I'll take NFS every time. I've posted my reasons in previous posts on this subject.

Regards,

Dave

[sheevm]

I guess someone in the team raised a security issue with NFS, this is the reason for considering alternatives.

Thanks for your input.

Rajim

[sheevm]

Hi

One more question. Can we configure CFS (CLUSTER FILESYSTEM) to share the filesystem between nodes without it being the service guard cluster?

Do we have to use CFS in conjuction with the Serviceguard?

Because in our envirionment the two application servers need to share the filesystems is not a serviceguard cluster.

I hope my question is clear.

Thanks.

[A. Clay Stephenson]

No, MC/SG is a requirement for CFS -- and, as I mentioned earlier, all nodes must be at HP-UX 11.23 or later.

[Samir Boussoussa]

Hi, I'm concerned by the same question, specialy installing Oracle Application on multinode wih RAC. So I need to share filesystemes.

NFS, is it good for this ??
Or should I use another mechanisme like export/import volume groups ...

Thanks in advance.

[A. Clay Stephenson]

It's the same answer for OracleRAC. Either use CFS or raw/io.