LANs, Ethernet and Contention Protocols

1. Having now solved (or at least beaten to death) the problem of reliably getting packets exchanged between a pair of connected hosts, we want to look at ways to interconnect more than two hosts. This essentially brings us to the network layer.

   Warning: I added the "essentially" to the last sentence because most people add an extra layer called the "Medium access control" or MAC layer to address the issues associated with local area networks that I plan to discuss first as part of the network layer.

2. The approaches used to building networks of machines that span relatively small geographic areas (a mile or three) are quite different from those used in much larger networks (spanning hundreds or thousands of miles). Accordingly, networks are categorized as:

   Local Area Networks

   or LANs which span small geographic areas,

   Wide Area Networks

   or WANs which span huge areas, and sometimes

   Metropolitan Area Networks

   for things that fall in the middle.

3. Another feature that can be used to categorize networks is the approach taken to making it possible for a large number of distinct machines to exchange messages.

   • Directly extending the approach we have assumed in our discussion of the data link layer (i.e. connecting all pairs of machines that want to communicate to each other with a wire or fiber) is not feasible. Even for small networks, the number of wires required, n(n-1), gets too large too fast.

   • One alternate approach is to just use enough wires to ensure that there is a path through wires and other computers between every pair of computers that need to communicate. The computers along such path can then help pass the messages along by acting as switches.

   • The other alternative is to transmit each message in such a way that all other machines can receive it.

     • Broadcasting in the sense of radio is the obvious way to do this.

     • You can also "broadcast" through a wire with many computers connected to it as is done in cable TV systems.

     • In a broadcast system, all computers will receive all messages, but each message will contain an address that will tell one machine to pay attention to it and all others to ignore it.

   • The broadcast approach is most commonly used in LANs. It will be the first approach we consider.

4. The big problem with using the broadcast approach is finding a way that the machines in the network can take turns broadcasting.

5. We could conceivably assign a distinct frequency to each station (effectively using FDM), but:

   • This would preclude the use of many simple transmission techniques (like NRZ and Manchester encoding) that can be implemented cheaply in hardware.

   • Dividing up the spectrum would limit the bandwidth availble to any station. It would be nice if instead when one station had a big message to send it could use all the bandwidth for a short period (i.e. TDM may waste less bandwidth).

6. Taking turns is tricky, however, when you don't have any way to "talk" about who goes next without using the network itself.

7. Ethernet is a protocol based on broadcasting that takes a conceptually very simple approach to figuring out how to take turns.

   • When a machine on an Ethernet wants to send a message, it first listens to see if any other machine is already sending a message.

     • If another machine is sending, it waits a random amount of time and tries again.

     • If no other machine is sending, it starts sending its message.

   • While sending a message, a machine continues to listen. If it hears another machine's message while still sending its own, it stops sending, waits a random amount of time and then tries again.

8. The basic approach used by Ethernet is called contention.

9. While this approach it conectually simply, to appreciate Ethernet fully one has to appreciate some of the fine points of the exact ways in which Ethernet stations behave. To appreciate these it is useful to discuss some of the choices one has to make when designing a contention protocol. These included:

   • Slotted vs Unslotted transmission,

   • degree of persistence (or maybe it should be called insistence).

   • the use of carrier sensing

   • the use of collision detection, and

   • the collision resolution procedure.