OK, the "two routers on one network issue."

This is where we have to be really precise in what we're talking about. Many of the devices that are sold to consumers as "routers" are, in fact, "switches" or "bridges." And I know I'm going to bore people to tears by explaining the difference, but it is important.

"Routers" are what we propeller-heads call a "layer 3" device. A real router makes packet handling decisions based on the destination IP address in your packet. A packet comes from your PC into the router: the router looks at the destination IP address in your packet, then looked in the "routing table" that tells it "OK, for this IP address (or corresponding network mask), that packet gets sent to this "next" particular IP address out a particular interface."

The important thing here is that the router doesn't care about Ethernet addresses until it gets to the point of actually transmitting that packet to the "next hop" IP address out some (usually) Ethernet interface. At that point, the router will look in what is called the "address resolution table" and resolve the "next IP hop" address into the Ethernet address for that next hop, put the Ethernet address onto your IP packet and squirt it out the appropriate Ethernet interface.

Clear as mud? OK, now "switches and bridges."

A switch/bridge (they both work the same, a switch is just faster, that's all) don't usually look at IP addresses - a switch is called a "layer 2" device because it looks only at Ethernet addresses. A switch or bridge "learns" the layout of your network by using something called a "spanning tree protocol" to figure out how to map your network without having any "loops" in the map. In fact, the "map" of the network, as created by a switch or a bridge, looks like what we turbine turbans call a "binary tree" -- imagine taking a real tree, yanking it out of the ground, and calling the base of the tree "the root node" -- in a common tree, none of the branches grow upwards and then connect to each other, right? Same thing here.

When a packet comes into a switch, the switch looks at the destination Ethernet address. The switch knows nuthin' about IP addresses, only Ethernet addresses. The switch looks in the spanning tree table to see out what interface the switch should send the packet to, and sends it out that interface. Either the destination Ethernet address is on that interface, or another switch is on that interface, who will pick up the packet and repeat the process. 

In the switch's view of your network, your switch is the "root" of the tree diagram. To do this, the switch has a setting or DIP switch on it that declares it to be "the root node." You can have only one root node in a switched/bridged network, and you must have one. When manufactures ship switches/bridges, they have every unit configured as the "root node", because they're assuming that this is the first (and possibly only) switch you're going to have in your network. So when you install more than one switch/bridge in a switched network, you have to disable the "root" setting  on all switch/bridges after the first one.

Most of the devices being sold into the consumer market are, in fact, switches, regardless of what the manufacture is actually calling them, but there are exceptions. A switch/bridge is easier for the customer to configure -- no IP routing protocols or static routing tables are necessary. The "spanning tree protocol" just magically "learns" your Ethernet network map and in small networks, "everything just works." A switch/bridge doesn't care whether you're running IP, XNS, Novell, AppleTalk or whatever higher level protocol, because the bridge looks only at the Ethernet map of the network to figure out where things are going. Switches/bridges have simpler software, are easier to manage in small networks and are just less of a pain to program all the way 'round.... until the customer wants a complicated network or to inject policy into networking, that is.