Pubsub over Kademlia

[decanus]

One thing to look into is pubsub over Kademlia, this is something we will probably want to implement in the future or use, if there is already a solution out there.

We MAY be doing ourselves a disservice by using something like Kademlia. Discovery protocols such as Pastry already have pubsub implementations such as SCRIBE whereas the research on this topic using Kademlia seems to be quite shallow.

A protocol to look at is PSS which allegedly is exactly that pubsub over Kademlia!

We mainly want to leverage incentivized routing and accounting for this.

Acceptance Criteria

• We want to investigate if there are protocols already and whether or not we can use any of them?
• If there aren't any protocols, it may make sense to design one and write a paper on the subject.

[decanus]

Here are the PSS docs, and here is another protocol overview of PSS.

[decanus]

From my current understanding of PSS, it isn't really pubsub. Messages are sent to regions or peers directly, a topic is added to a message in order for a user to dictate whether or not they want to listen to a specific topic. This means to me means that it is not pubsub in the traditional sense where multiple parties subscribe to the same topic that is stored in some list.

[decanus]

Just a note from a conversation with Nick Johnson based on a post he wrote a while ago:

The problem with pubsub over a Kademlia that isn't a forward Kademlia is that messages aren't routed through intermediate nodes along the discovery path meaning you can't form a broadcast tree using the network

[decanus]

This may be relevant: https://cs.baylor.edu/~donahoo/papers/MCD15.pdf

[decanus]

see: https://forum.vac.dev/t/notes-on-distributed-pub-sub/44