MUST_USE_LOCAL_DRIVE (MULD)and media sharing and maximum number of part full media in pool.

[jim_dalton]

This combo seems not to work: Ive got a media server with MULD set, Ive also got set 'enable unrestricted media sharing for all media servers' and no server groups , as per the doc, and Ive got a volume pool with max partially full media=3.

And Ive got eight policies that use the pool , all running to a server that is also a media server and all using the vol pool described above. Any_available is the policy storage directive.What happens is: it kicks off three polices to the local media server, each using one medium.Thats correct. Then it kicks off another three, ignores the MULD, uses another media server ( the master as it happens) and uses a further 3 tapes. So we're up to six on a pool with a limit of 3. Then it decides the pool limit is reached (as seen in activity monitor), and waits until a drive and its media come free, and carries on like that. 

I dont want to use another media server and I dont want to use more than three drives/media, but Netb refuses to follow the directives.

So maybe Im missing something here, but if media are shared across all media servers then that limit ought not be breached, and if Im saying it MUST use local drive, then Im insisting it do so, therefore theres a breach of its own rules twice over.

 Looks like a bug or maybe two associated with tape sharing and active media.

Solaris 10 6.5.2A, Solaris9 client and media server 6.5.2A,Oracle agent,LTO4.

If you know of a solution or workaround it would be appreciated: one thing it has to use a local drive - these backups are running from clustered Ora so whichever side of the cluster the Ora instance is on, it has to use the drives it has there (and nowhere else), or queue.  

Thanks,Tim

[jim_dalton]

Ive half sussed out whats going on here...I'll rephrase the question: Ive got a number of oracle instances running on a Sun active-active cluster. Each of the nodes in the cluster is a media server and each of the Ora instances resides on one node or the other. Each Ora instance has a service address which can be on one node or the other (provides us with failover). So when I run my backup, netbackup wont accept that , say, oracle-dbA, when on serverone, should use serverone as its media server.

It appears to think along the lines of : whats my name? ( oracle-dbA ), is there a media server called oracle-dbA? (no), therefore i need a media server. 

Rather than: whats my name ?(oracle-dbA), whats my true hostname? (serverone), is there a media server called serverone? (yes), or similar ( ie when the backup fires off on the client, look at its parameters - theres a MUST_USE_LOCAL_DRIVE=true in there, look for a local media server.

Im sure theres a way around this: its too useful a function to not have! 

Thanks in advance,Tim

[samagee]

I have noticed this article about it as well. I have not tested it yet though.

http://seer.support.veritas.com/docs/278996.htm

[J_H_Is_gone]

what about your storage unit....  how many drives are in the storage unit?

try setting up a storage unit with just 3 drives. and have the policy use that one.

[jim_dalton]

Not sure I understand your point, JH: each media server has its own storage unit that happens to be the same on all media servers as this is SSO.

The idea is that I dont tell it which storage unit- I say 'Any Available' and netb works out that is has a storage unit attached locally regardless of whether its on node a or node b or the cluster. That way it always writes to locally attached drives and not over the network and thus is performant. This works fine when the node being backed up (client name) matches the media server name, but I have service addresses that can exist on either of the two nodes courtesy of suncluster, so netb needs to realise that the service address exists on a node which is a media server and hence should use the locally attached drives.

There does seem to be the notion of clustered media servers , but mine arent clustered, they are shared and its the service that is clustered ( ie SAP, Oracle). Ive not found any clarity in the documentation on this point.

Thanks,Tim