How exactly do switches extend the network? Can you give a more concrete example? Sure - imagine that you had two computers 4000m apart that you wanted connected via a 10Mbps Ethernet. The maximum cable length for one of these is 2500m. So, you could have 2 cables, each one 2000m(*) and then have a switch/bridge in the middle connecting them together. The (*) here is because I believe that this distance includes the repeaters necessary. With bridges, let's say one fails and creates two disjoint networks. Do they get remerged later? Yes. The same protocol that creates the spanning tree is periodically run by the root. If you take a network that has become disjoint because a critical node died, it looks like two networks that were created independently. When you add a bridge to join them, the same protocol will be run and the combined network will have a single root. Let's say you have ---B1----B2---B3--- and B1 sends a broadcast. B2 copies it and B3 sees it. How does B3 know not to send it back to B1 in some way? The bridges know which wire had the incoming packet, so when they rebroadcast it, they exclude the wire on which the packet was received originally. This means that when B3 sees the packet from B2, it'll only send it on its right link. No stretch break! Sorry - I get the sense that we're falling behind. This was a packed lecture, and even without any dilly-dallying, we still didn't make it completely through. So, maybe that's a lesson to me that we should always have a stretch break. Does datagram switching have all of the performance benefits of virtual circuits? In theory, one could have virtual circuit identifiers that are smaller than the addresses contained in the datagrams. So, there's a small amount of wasted space in that regard. Also, with virtual circuits, one could more easily reserve a certain amount of resources (bandwidth, memory space at the switches/routers, etc) more easily than with standard datagram switching. Windows Network Neighborhood kills networks due to broadcast messages unless it is really intelligent. I can believe that. It probably wasn't meant for giant networks. Can an Ethernet switch send two messages simultaneously if the messages don't involve any of the same computers? If done right, the answer is yes. If all of the machines are connected to the switch itself, what you have is a star topology instead of a shared medium. So, only machines communicating with each other need to see the traffic. If you have N machines and they're having N/2 distinct communications, the switch should be able to allow all of them to take place simultaneously as long as it's well-designed. This is one of the scalability aspects of switching. Who decided datagram switching over circuit switching for the internet? The Defense Department (via ARPA) funded development of what we now call the Internet. I don't know how they picked the particular proposals to do packet switching over circuit switching, but that was the reason for this development. I think it's generally attributed to Professor Leonard Kleinrock of UCLA (check out his homepage), but the folks at the company BBN also claim credit for it. When is virtual circuit switching used? I believe the book explains how this is used with X.25 and ATM. Looking around on the Internet, it also seems to be the case that it's used with a new technique called MPLS (multiprotocol label switching). projector is XGAVPL PX31 I get conflicting information on the web about this - some seem to say the resolution is 1024*768, while others put it at 1280*something. In any case, I'll try to use thicker lines.