I attempt here to gather together and make a succinct chronical of the comms network(s) efforts discussed in the logs over time.
The first item I can recall proposed was gossipd.
This item is no longer under discussion for reasons that are not clear to me due to my inability to properly study and digest the comment thread in the above article.
As Freenode IRC's ultimate decline and collapse has become evidently imminent, asciilifeform has begun evacuation procedures. The first step he took was to stand up dulapnet, his own ircd running the Unreal ircd.
Subsequently, members of #asciilifeform, including signpost, gregorynyssa, asciilifeform, and myself attempted to form a new type of IRC network in which each participant runs his own ircd. Each ircd would be linked to each other ircd. In addition to being decentralized, such a network would have the advantage of being accessible via any existing IRC compliant client. We then discovered that IRC networks do not allow for cycles, and can only exist in a tree topology. This topology was not acceptable to asciilifeform or signpost due to centralizing requirement for there to be a root node1.
Discussion moved on to another type of network in which the peers would communicate via UDP, but clients would connect to peers using the IRC protocol. It would be trivial to architect such a network using shared secret keys. However, asciilifeform also proposes that this network should allow unknowns to broadcast to the network via a PoW mechanism. This would allow unknown and untrusted entities to gain entry into the network without manual intervention, which would allow anyone to drop in for a chat while at the same time minimizing the attack surface of the network, eliminating the need for anyone to check their email ever again.
At the present time, this type of network is not feasible due to the apparent intractability of sharing the required difficulty level for the PoW mechanism. I'm unaware of any further actions I or any other interested parties can take to make progress down this path.
Signpost is investigating a third network type based on the Kademlia DHT. Objections to this approach are here
It is interesting to note that Freenode ultimately did not succumb to any sort of technical attack such as a ddos, but due to sabotage by long time admins who walked away with user data stored in a centralized database.
- This network is notable that it is the first publicly available prototype of any potential comms network discussed in the logs [↩]
"At the present time, this type of network is not feasible due to the apparent intractability of sharing the required difficulty level for the PoW mechanism."
Why not let that be a server-owner operated knob?
"This topology was not acceptable to asciilifeform or signpost due to centralizing requirement for there to be a root node"
Technically there is no root node, it just has to be an acyclic network. So you could have servers connected in series A-B-C-D, where B and C are equally vital to the health of the network. If either B or C goes down, or if there is a connection lost by either of them, there will be a netsplit.
Simply by adding in a de-duping mechanism, cyclic networks can be made, which completely removes the whole netsplit problem from a well interconnected network.
"eliminating the need for anyone to check their email ever again." -- This seems like a big jump, what does this even have to do with email? Do you get emails from people wanting you to give them voice privileges in channels? I use email and chat very differently, so I don't think email should even enter into the discussion here?
Deduping would not be compliant with the IRC RFC.
Email here is used metaphorically here to represent a method of accepting messages from randos.
David, I believe asciilifeform's argument was that an unknown individual with a computer of unknown capabilities might, with good probability, end up nearly locking up their computer trying to send a message.
FTR here's my long-promised spec.