two hosts write to same storage

[Aaron_82]

Is NFS my only option? I have two hosts, one has the VG mounted and the other host mounts the VG via NFS. This is a ServiceGuard environment. The point is if one box goes down the other box can still write to the storage.

[Robert-Jan Goossens]

Hi,

You could also think about a SAN (storage area networks), the problem with nfs shared disks is like you mention. If one of the hosts fails the share is gone.

Regards,
Robert-Jan

[D Block 2]

are you using MC/ServiceGuard's "Highly Available NFS Package"?

In this case the two nodes forming a cluster both provide NFS servers like behavior. if one goes down, then the other will provide NFS services - Highly available.

On the other hand, you do have an option of using MC/SG Volume Groups, right. is the cmlvmd daemon running on both servers ?

[Aaron_82]

This is in a SAN environment(brocades, EMC DMX, etc...). Not a problem using HA NFS but I was just wondering if there was another way. The application will be up on both servers and both hosts need to read/write to the same storage. The one that has the NFS mount is going to be significantly slower so I was just wondering if there was another way to have two hosts write to the same disks. It doesn't have to be a HP solution it can be third party and I can throw more gigabit nics if need be.

[Sunil Sharma_1]

Hi,

It's not possible to activate same volume group on two hosts. So only option is NFS.

In some case like Oracle RAC both nodes can write on same Volume but for File system i don't think it is possible.


Sunil

[Jeff Schussele]

Hi Aaron,

Although one can do this, it will inevitably lead to tremendous problems. The problem is that host A will have *no* idea that host B has not written it's buffer cache data to disk and will go ahead & read the disk. Eventually one host will update the inode table unexpectedly to the other which will then declare the filesystem corrupt.
You just cannot do this any other way than NFS.

My 2 cents,
Jeff

[Ian Dennison_1]

Mount it on a third system? Maybe use SAMBA and host it on a Windows Machine?

Alternatively if one node goes down, the cutover action can include umounting the NFS share and remounting the Disk directly.

Share and Enjoy! Ian

[D Block 2]

are there any performance issues using NFS ? just wondering

[Dave Olker]

Hi Aaron,

Today, with HP supported products, NFS or CIFS is the only way to do this.

In the next major release of HP-UX, 11i v3, we are planning on introducing a Clustered File System, which will allow all nodes in the cluster to simultaneously access a shared pool of storage. You could then use all of these servers in the cluster as one big NFS server and export the shared filesystems to remote clients. However, the nodes inside the cluster would all access the filesystem locally.

Other vendors may offer a clusterd filesystem solution today, but if they did you would have to work with them for support issues. The 11i v3 solution will be an HP provided, and supported, solution.

Hope this helps,

Dave

[Hein van den Heuvel]

Writing to the same storage is trivial: Connect a fibre from the switch to an hba on each box, ioscan + insf and open the raw devices.

Making it useful for an application is hard. You need to higher level locking.

Since you mention 'nfs' i guess you are interested in file based access, not raw blocks.

Writing to a shared, static, pre-existing files from two nodes is relatively easy also.
Making is usefull for an application is harder. Again, external locking is needed.
Oracel essentially does this with it's OCFS.
This is a very light protocol to be sure the same space is not handed out twice and then the standard Oracle locking protects the blocks.

Writing lots of file from multiple nodes, but only accesses from one node at a time is relatively easy and requires little or no application code change. But is requires a good 'cluster file system'. The OCFS is one (free) possibility for limited use. Other commercial products exist (legato, ibm,...)
HP's VMS and Tru64 OSes have had cluster wide file systems for years. This code is (slowly) moving into hpux. Early performance results are very good (far out!), but availability is still also 'far out'.

You may want to clarify 'write to storage' for better help. Just plain file access? Shared file access? ..

Hein.

[Aaron_82]

I have the same application on 2 systems. The EMC mounted storage is on one of the servers and the other server uses NFS to read/write it's files to the same storage.
The way we want it to work is that both systems share the storage and if one of the systems goes down then the storage is still available to the other.
With NFS, we have to at least stop the application to reconnect to the storage.

[Dave Olker]

Hi Aaron,

Please explain why you need to shut down the application and restart it after the package moves from one server to the other. Is this a file locking problem, or some other issue?

Thanks,

Dave

[Hoefnix]

Hi,

If you have to use NFS, why not use a storage device that does it for you. NetApp has NAS storage that can be configured with HA hardware and gives you an NFS server. Still you have performance problems with you NFS connection like you mentioned.

We were having some problems in the past with uploading large files to an NFS NAS device.
It is covered by a thread started by Elmar P. Kolkman.
http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=400028
Elmar say's it's not solved but when we upgraded to HP-UX 11i it was solved, but maybe there are tips in the thread that helps in your case.

HTH,
Peter Geluk

[Aaron_82]

NAS would be a good solution if we had to use NFS. Just wondering if there is another way to share storage between hosts. Does anybody know if there is EMC software that could do it?

[Dave Olker]

Hi Aaron,

If by "share storage between hosts" you're implying a clustered file system, where all hosts directly attach to a shared storage pool and simultaneously access the storage while negotiating metadata to avoid corruption - HP doesn't offer anything like that yet. We plan to introduce that in 11i v3.

I'm still curious why you have to stop/start your application following an HA/NFS package migration. I know this used to be a common practice for applications that used file locking, but HP released an updated version of the HA/NFS product that resolves that problem by migrating file locks between NFS servers during a failover. That should allow applications using file locks to gracefully recover their locks after the package switch and avoid the need for restarting.

Is there some other reason why the application needs to be stopped and started? If so, perhaps there is a way to configure around that and still use NFS effectively until a Clustered FS solution is available.

Regards,

Dave