Socket Foundations, Byte Order, and Encoding

Sockets

A socket exposes network-protocol services through a programming interface, allowing processes to communicate without implementing the entire protocol stack themselves. Programming languages wrap the underlying mechanisms in accessible APIs.

• TCP maintains connection state, establishes a connection through a handshake, and supports reliable delivery. Connection management has costs.
• UDP sends datagrams without TCP's connection handshake or teardown. It does not guarantee that every datagram arrives; suitable addressing and options also allow broadcast use.

A TCP handshake establishes a connection, not every send on that connection. Reliability also depends on sequence numbers, acknowledgments, and retransmission. Performance depends on workload and environment; TCP is not universally slower. Ordinary UDP unicast does not automatically broadcast to multiple receivers.

Endianness

Endianness determines the byte order of a multibyte value in memory or on a wire. Little endian places the least significant byte at the lowest address; big endian places the most significant byte there. Consider 0x12345678 and 0x11223344:

CPU designs do not all use the same order. Network byte order is big endian, as specified for Internet numeric fields in RFC 1700. Programs convert applicable host-order values to network order for sending and back for receiving, independently of CPU and operating system.

htons() and htonl() convert host to network order; ntohs() and ntohl() convert network to host order.

This convention applies to multibyte numeric fields defined by a protocol, not to reversing arbitrary payload text. Python's struct.pack("!I", value) encodes an unsigned 32-bit integer in network order; struct.unpack decodes it. ! explicitly selects network order rather than the host's memory layout.

Text and Bytes

Python str contains text, while sockets transmit bytes. text.encode("utf-8") produces bytes; data.decode("utf-8") recovers text from a complete encoding. Character count need not equal byte count: “网络” has two characters but six UTF-8 bytes.

A UTF-8 character can span several bytes. A TCP receive need not end at a character or message boundary. Accumulate data according to framing rules or use an incremental decoder rather than assuming every received fragment is complete text.