Understanding socket basics

A socket, in C parlance, is a data structure in kernel space, corresponding to one end-point of a UDP or TCP session (I am using session very loosely when talking about UDP). It's normally associated with one single port number on the local side and seldom more than one "well-known" number on either side of the session.

A "raw socket" is an end-point on, more or less, the physical transport. They're seldom used in applications programming, but sometimes used for various diagnostic things (traceroute, ping, possibly others) and may required elevated privileges to open.

Sockets are, in their nature, a binary octet-transport. It is not uncommon to treat sockets (TCP sockets, at least) as being text-based streams.

I have not yet seen a programming model that doesn't involve something like sockets, if you dig deep enough, but there have certainly been other models of doing networking. The "/net/" pseudo-filesystem, where opening "/net/127.0.0.0.1/tcp/80" (or "tcp/www") would give you a file handle where writes end up on a web server on localhost is but one.

Socket is a software mechanism provided by the operating system. Like its name implies, you can think of it like an "electrical outlet" or some electrical connector, even though socket is not a physical device, but a software mechanism. In real world when you have two electrical connectors, you can connect them with a wire. In the same way in network programming you can create one socket on one computer and another socket on another computer and then connect those sockets. And when you write data to one of them, you receive it on the other one. There are also a few different kinds of sockets. For example if you are programming a server software, you want to have a listening socket which never sends or receives actual data but only listens for and accepts incoming connections and creates a new socket for each new connection.

Short answers:

1. Socket is an abstraction of an IP connection endpoint - so if you think of it as an API structure, you are not very far off. Please do read http://en.wikipedia.org/wiki/Internet_socket
2. Internet layer i.e. IP Protocol. In practice you usually use explicitly sockets that bind to a certain transport layer parameters (datagram/UDP or stream/TCP)
3. Sockets send data, in network byte order - whether it is text or binary, depends on the upper layer protocol.
4. Theoretically, probably yes - but in practice all IP traffic is done using 'sockets'