>>From ADSM's point of view, we set up the clients to use IP addresses that
>>are used by the backup network (10.0.x.x).  The server replies to the
>>clients using whatever network the client started using.  We never give
>ADSM
>>a machine name to resolve, therefore it never uses the main network.
>
>Mitch;  I don't understand how the ADSM server can be made to reply using
>whatever network the client started using.  The routes set up on your ADSM
>server system will control the routing of packets sent back to the client,
>no?  ADSM specifies the target IP address for the packets, but the
>underlying OS/TcpIP will determine which interface card to send them out
>on, based on the routes defined.  Seems to me that if you only have one
>default route specified, then all outbound traffic from ADSM will go out
>that default route.

The server doesn't keep a record of the client's IP addresses- only the node
name.  (This is handy for desktops using DHCP, among other things.)  The
only way the ADSM server knows how to contact a client is to reply back to
it.  That's why the scheduler service contacts the server upon startup.

It's important to remember that different network cards use different IP
ranges.  In our case the backup network is 10.0.0.x and the production
network is 38.x.x.x.  The ADSM server has *two* IP addresses- one for each
card.  If a client contacts the server over the backup network, it's using
10.0.0.x addresses for the server and itself.  Default route settings
shouldn't play into this at all.

>>We have had occasion to run backups on both cards at the same time. It
runs
>>fine.  The only thing I'd recommend is using straight IP addresses and
avoid
>>name resolution.
>
>Can you expand on the recommendation to not use name resolution?  I don't
>see what the problem would be, and not using it can introduce other
>headaches in managing your systems.

Name resolution is the process of resolving a name (say, SERVER1), to an IP
address (say, 10.0.0.1).  When you have a machine with two network cards, it
therefore has two IP addresses. (say, 10.0.0.1 and 38.1.1.1)  With NT it's
not real clear which IP address you should get when you say "SERVER1".

A good example of this is to type "PING SERVER1".  What IP address will it
use? Well, it depends...

You can get around this by using a LMHOSTS file, but that's a lot of manual
work.  The idea there is to have two names for the server, each with a
unique IP address.  Another option is to remove the second address from WINS
and DNS, so that it always resolves to the first address.  The only time the
second address is used is if you explicitly use that IP address.

-Mitch
=========================================================================