As we mentioned in Chapter 1, An Introduction to Networks and Protocols, network programming is usually done using a client-server paradigm. In this paradigm, a server listens for new connections at a published address. The client, knowing the server's address, is the one to establish the connection initially. Once the connection is established, the client and the server can both send and receive data. This can continue until either the client or the server terminates the connection.
A traditional client-server model usually implies different behaviors for the client and server. The way web browsing works, for example, is that the server resides at a known address, waiting for connections. A client (web browser) establishes a connection and sends a request that includes which web page or resource it wants to download. The server then checks that it knows what to do with this request and responds appropriately (by sending the web page).
An alternative paradigm is the peer-to-peer model. For example, this model is used by the BitTorrent protocol. In the peer-to-peer model, each peer has essentially the same responsibilities. While a web server is optimized to send requested data from the server to the client, a peer-to-peer protocol is balanced in that data is exchanged somewhat evenly between peers. However, even in the peer-to-peer model, the underlying sockets that are using TCP or UDP aren't created equal. That is, for each peer-to-peer connection, one peer was listening and the other connecting. BitTorrent works by having a central server (called a tracker) that stores a list of peer IP addresses. Each of the peers on that list has agreed to behave like a server and listen for new connections. When a new peer wants to join the swarm, it requests a list of peers from the central server, and then tries to establish a connection to peers on that list while simultaneously listening for new connections from other peers. In summary, a peer-to-peer protocol doesn't so much replace the client-server model; it is just expected that each peer be a client and a server both.
Another common protocol that pushes the boundary of the client-server paradigm is FTP. The FTP server listens for connections until the FTP client connects. After the initial connection, the FTP client issues commands to the server. If the FTP client requests a file from the server, the server will attempt to establish a new connection to the FTP client to transfer the file over. So, for this reason, the FTP client first establishes a connection as a TCP client, but later accepts connections like a TCP server.
Network programs can usually be described as one of four types—a TCP server, a TCP client, a UDP server, or a UDP client. Some protocols call for a program to implement two, or even all four types, but it is useful for us to consider each of the four types separately.