Index

There are two ends to each connection: the client, that is the host that initiates the connection, and the server, that is the host that responds to the connection. Clients and servers are connected by sockets.

On the server side instead of connecting to a remote host, a program waits for other hosts to connect to it. A server socket binds to a particular port on the local machine. Once it has successfully bound to a port, it listens for incoming connection attempts. When it detects a connection attempt, it accepts the connection. This creates a socket between the client and the server over which the client and the server communicate.

Multiple clients can connect to the same port on the server at the same time. Incoming data is distinguished by the port to which it is addressed and the client host and port from which it came. The server can tell for which service (like http or ftp) the data is intended by inspecting the port. It can tell which open socket on that service the data is intended by looking at the client address and port stored with the data.

No more than one server socket can listen to a particular port at one time. Therefore, since a server may need to handle many connections at once, server programs tend to be heavily multi-threaded. Generally the server socket listening on the port will only accept the connections. It then passes off the actual processing of connections to a separate thread.

Incoming connections are stored in a queue until the server can accept them. (On most systems the default queue length is between 5 and 50. Once the queue fills up further incoming connections are refused until space in the queue opens up.)

Index