The term Peer-to-Peer has expanded and shifted since the early days when it merely referred to a type of decentralized network topology where every computer on the network was completely autonomous and individually configured. Today the term is more often used to describe direct computer-to-computer communication to share files and data because they also have an autonomous, decentralized behavior.
A Logical Topology Used in File Sharing Services
You will most frequently hear the term peer-to-peer referring to a piece of software used for decentralized file sharing. Each computer running a copy of the software is called a peer because it can autonomously retrieve or share out files to all other computers running that same software and running the same protocol (not all file sharing software can communicate with each other).
Examples:
- I2P
- Torrent clients
- Gnutella
- Direct Connect
- Windows Update
As a Network Topology
In a peer-to-peer network, network IP addresses are often configured manually and any settings for file shares, printers and other services are unique to, and accessable from each sysetem. All access, logons, services and policies are provided to the users of that computer only. There is no centralized management in the network whatsoever. With no central database for logins and access, each user will have a separate login for each machine. This is fine for small networks, but does not allow them to grow larger (scale) very well.
In peer to peer networks, every single machine must have a separate account for each human who wants to use the machine. Worse, each machine can potentially have a different human administrator. This makes maintaining corporate-wide policies impossible, as each administrator can do whatever they want with their machine. To make things even more nightmare-ish, an administrator will often have to run round to each computer individually and log in locally in order to make changes. This keeps the admin running all day long, typically in circles and never rarely anything accomplished.
However, if only a few machines exist, and each user can be trusted, this model can often be more productive.