FS share on two nodes

[Ryan Kogelheide]

This sounds like a strange question, but my unix team is telling me it's not possible to mount (as fs) a single vg on two nodes at the same time.

Is this true? Why not? What am I not understanding about how this works?

(2xL1000-5X @ HPUX B.11.00. U)
(2xVA7100)

[harry d brown jr]

Well they are about 1/2 correct. You can have serverA mount a VG (probably on a san) in RW mode, and serverB mount it in RO (read only) mode. But any changes that serverA makes will not be reflected on the "view" that serverB is looking at, unless you (from server B) umount and mount it again. I've done it using a va7400.

live free or die
harry

[Andreas Voss]

Hi,

one possibility is to activate as readonly on the second host (vgchange -a r ..)
Then you can mount on both nodes but only one should/could be able to write to the fs.

Regards

[harry d brown jr]

I'm also thinking that maybe Veritas has something that will allow both servers to see the VG in RW (read/write) mode. I know that Oracle and Informix have new products that allow a single database to be updated from many servers.

live free or die
harry

[Duncan Edmonstone]

It *is* possible to access a volume group on two nodes at the same time. What *is not* possible is for two nodes to mount a file system in a shared volume group at the same time. This is because there is no concept of distributed file locking and sharing within standard file systems (like vxfs and hfs). If you can guarantee that the data is not changing, then you can mount the volume group and file systems in read-only mode without any issues...

If you have some kind of distributed lock management system, like that used by Oracle Parallel Server, then raw logical volumes within a volume group can be shared and accessed.

I believe Veritas sell a file system called 'Veritas Cluster File System' which allows access from multiple nodes at the same time... but of course as a Veritas product this will probably be extremely expensive!

Perhaps if you let us know what you are trying to achieve, we can give you some pointers for solving your problem...

HTH

Duncan

[harry d brown jr]

Boy, is Duncan correct about Veritas pricing. A license for Veritas VxVM on a sun web server (t1) is more expensive than the server itself.

The other day I found a reference to CxFS that allowed multiple servers to update a single VG, but I'm not sure as to it's portibility.

live free or die
harry

[Duncan Edmonstone]

This is the Veritas product which enables file sharing :

http://eval.veritas.com/downloads/pro/fso_hp-ux_datasheet.pdf

Be prepared to take out a second & third mortgage!

HTH

Duncan

[Roger Baptiste]

Ryan,

As Duncan mentioned, we can share only RawVolumes of a VG - which happens in a distributed OPS cluster environment where oracle raw volumes are shared . We are not yet there for Filesytems. Hence the bigbucks for the veritas product.

-raj

[Ryan Kogelheide]

Ahh... so there's no DLM for the Unix FS like there is for Oracle. OK. I wonder why not? I'm sure it's not trivial, but FS locking seems like it should be the exact same problem.

Anyhow, thanks for your responses. I think I understand a lot better now.

The problem will have to be solved at the Oracle level. It makes archivelog management much simpler if the logs can go on a shared volume because a recover (which must happen on a single node) needs the logs from both sides. I'll use separate vgs and mount both on one side when a recover is needed.

[Duncan Edmonstone · Accepted solution]

Ryan,

If your intention is to use the 'shared file system' for archived redo logs, and read those into a recovery database, then you can either:

Just use NFS - have the archived redo logs written to a local disk which is then NFS mounted by your recovery server.

-OR-

If you are using 8i, there are features that allow you to ship archived redo across Net8 and straight into a recovery database.

HTH

Duncan