How does IRC's federation model compare to ActivityPub? July 3, 2021 on Drew DeVault's blog

submited by

Style Pass

2021-07-03 21:00:07

Today’s federated revolution is led by ActivityPub, leading to the rise of services like Mastodon, PeerTube, PixelFed, and more. These new technologies have a particular approach to federation, which is coloring perceptions on what it actually means for a system to be federated at all. Today’s post will explain how Internet Relay Chat (IRC), a technology first introduced in the late 1980’s, does federation differently, and why.

As IRC has aged, many users today have only ever used a few networks, such as Liberachat (or Freenode, up until several weeks ago), which use a particular IRC model which does not, at first glance, appear to utilize federation. After all, everyone types “irc.libera.chat” into their client and they all end up on the same network and in the same namespace. However, this domain name is backed by a round-robin resolver which will connect you to any of several dozen servers, which are connected to each other1 and exchange messages on behalf of the users who reside on each. This is why we call them IRC networks — each is composed of a network of servers that work together.

But why can’t I send messages to users on OFTC from my Libera Chat session? Well, IRC networks are federated, but they are typically a closed federation, such that each network forms a discrete graph of servers, not interconnected with any of the others. In ActivityPub terms, imagine a version of Mastodon where, instead of automatically federating with new instances, server operators whitelisted each one, forming a closed graph of connected instances. Organize these servers under a single named entity (“Mastonet” or something), and the result is an “ActivityPub network” which operates in the same sense as a typical “IRC network”.