Very Computer

Here's the layout:
1.5Mbps DSL with 16 static IP addresses
      +
1.5Mbps DSL with 8 static IP addresses (ok, 5 usable)

Both of these circuits terminate on isolated LANs, however, the Solaris
server in question has 2 NIC cards, one for each network. Ok, so far so
good.

Now we add DNS entries for the server, two entries with the same name,
different IP addresses. We also have 2 default routes in the server, one to
each of the DSL gateways.

Something is wrong. Sometimes can come in on one circuit, sometimes not.
After connection on a particular circuit, the connection "goes away". I'm at
my wits end on this. Sorry if this is a "basic" problem for all you routing
gurus. It's just not my thing.

Thanks in advance. Email replies would be best. Thanks!!!

Rob Bruce

See my comments imbedded below
-Tim

Sounds ok so far.  I assume both interfaces come up and each can ping at
least one other node on their respective subnets without a problem?

Quote:>Now we add DNS entries for the server, two entries with the same name,
>different IP addresses.

Ooh - not so good.  While this technically works, the DNS server is only
going to give one address to the client.  That'll be random. That means
that sometimes a client will get the IP address of the local network
interface - other times the client will get the address of the non-local
network interface.  If the client has a valid route to the non-local
interface then I guess this would be ok.  So... do the clients have
valid routes between networks?

Quote:> We also have 2 default routes in the server, one to
>each of the DSL gateways.

You *can* legally have two or more default routes in Solaris, however
you cannot predict how they'll work.  I seem to recall that in Solaris
2.6 they would round-robin.  I'm not sure what the behavior is in
Solaris 8.  I generally do not recommend having more than one default
route unless you have a really good reason for it.  Even Solaris
multipathing uses only one network interface and fails-over to the 2nd
interface if & when there's a problem with the first.

Technically a "connection" is based on the protocol type plus the IP
address & port of the two endpoints.  Intermediate routing nodes have
nothing to do with it (besides conveying the invidual segments) - it's
completely acceptable for different segments of the same connection to
take different routes (and even arrive out of order) and for the network
path in one direction to be different from the network path in the other
direction.  But... all this assumes that both interfaces can, in fact,
reach all the same networks... which may be assuming too much.

When you use multiple default routes, this isn't like most routing
protocols where a network outage will cause reconvergence to locate
another working path.  Solaris assumes that both 'default' routes are
equally valid at all times.

Is your Solaris box supposed to be a router between the two networks?
 If so is it properly configured as a router and do the other clients
know that it works as a router.  Otherwise pinging the non-local
interface can cause a client to be unable to locate the box (creating
the illusion that sometimes you can reach it and sometimes you cannot).

You might give us some configuration data.  Do an "ifconfig -a" and a
"netstat -nr".

About this "connection goes away" issue.  If the other clients are
reaching this box by name in DNS (which means they get random addresses
back), and if they don't know how to route to the non-local interface
(even if the Solaris box is setup as a router you can still have a
problem if the remote clients don't know that they can use the Solaris
box as a router) then you could get the symptom you describe.