That's not a NBU problem, but a routing problem. Since the media server will contact the client before asking to start the backup, you may try to add individual routes on your media server, to the clients, and set them to use the other interface. Start with one client to see if it works...

But really hard to explain.  I will give it a try.

You don't configure networks in Netbackup. You only have one primary IP and gateway, the DNS entry for the primary network is also the server name (most often, cluster have exceptions). So far so good. When you add a secondary network it must have a different DNS name. This name must also be listed on Netbackup client's server list or you will get at status code 59 -"access to client was not allowed".

When ever Netbackup connect to a client, it uses the hosts intern routing table to see if it direct connect to the destinationsubnet, if that not the case the packet is send to the default gateway. This mean master to media server communication may start using the secondary network if it listed earlier in the internal routing table (use netstat -rn to see it).  Because of this both vm.conf and bp.conf must be updated with the secondary network interface as well.

Seen from a network separation side. This where the assumptions come into play. The two network are they flat (eg 255.255.255.0) or do they contain multiple Ip segments ?

If they are flat, the internal routing table will make sure traffic stay the right place, because traffic is never send through default gateway if a NIC is local attached to the destination subnet.

If there are multiple IP segments you will need to manipulate the routing table with routing statements to separate traffic as there can only be one default gateway.

Hope this helps.

So we gave this a try. 

In one phsycial media server, we had one NIC with one IP (10.60.xx.xx) and one name (MServerA) to back up the Windows servers, and another NIC with another IP (10.70.xx.xx) and a different name (MServerB) to back up the Unix servers. 

From any server on the 10.60.xx.xx network, we ran traceroute to MServerA; the output (correctly) showed only one IP, which was the IP for MServerA.  We repeated the same from the media server to the same clients, and it (correctly) showed the same route, no hops, to any of those clients. 

We repeated the above for the 10.70.xx.xx network's clients and got the correct results we were hoping for.  So the backups for each network were going through their respective interfaces, just as we had planned.  Backups were successful, restores worked without a hitch.

Then we hit a problem: When that night's backups kicked in and we had a couple hundred backup jobs active/queued, we started getting sporadic backup failures of the same file systems on each server.  Some Windows backups of "System State" or the LAST filesystem (G: drive, etc.) failed with either status 24 (socket write failed) or 42 (network read failed); Unix backups to disk would fail on various file systems with a status 213 (no storage units available for use).  Once the backup load on the media server came down, we could restart these backups successfully.

Oddly, our network monitoring system indicated a shortage of memory/swap space on the media server during this time, but showed 10GB of free memory, and the swap space was nowhere near full. 

After a week of trying to tweak NFS & TCP settings, we are reverting back to our previous setup to confirm the problem is related to the extra 10G interface.

A old rule of thump say you need 3.5 times the memory in swap space. Not that it will ever use it, but the system need to be able to allocate/reserve it for operation.

We removed the 2nd 10G interface and everything ran successfully again.

I've currently got a case open with Symantec to confirm whether this configuration is supported.  The NBU Planning and Performance Tuning Guide, p.123, shows how to configure "Using multiple interfaces", but we have been told by two different support folks that Symantec doesn't support the use of more than one NIC/IP/DNS name per media server.  The technical aspect of getting this to work is now on hold until we can resolve this supportability issue with Symantec.

Thanks for the advice, Nicolai!

An option would be to create aliases for the media server so that NetBackup will know that multiple host names point to the same media server.

From the master server:

C:\Program Files\Veritas\NetBackup\bin\admincmd\nbemmcmd –machinealias –addalias –alias media_server_alias1_name –machinename media_server_name –machinetype media

C:\Program Files\Veritas\NetBackup\bin\admincmd\nbemmcmd –machinealias –addalias –alias media_server_alias2_name –machinename media_server_name –machinetype media

Verifying aliases for the media server

Nbemmcmd –machinealias –getalias –machinename media_server_name  –machinetype media

Per the Symantec tech, so long as both interfaces are listed as valid servers for the clients, the configuration is supported.