Hey, I worked in a related research field a few years ago and published some of the results. Here are some ideas I find important:
- Generally I found that overlay DHT based routing is not very scalable for large graphs. It performs well for random looking graphs, but becomes very inefficient for large planar-like graphs. If pinecone is designed for small local networks, it might work well, but it will probably not scale well to large networks with the current design.
My intention in dealing with planar networks is mostly the case of deploying a local mesh network. Most mesh networks behave like planar graphs. Not sure if this is an actual use case for matrix though.
- Generally I found that overlay DHT based routing is not very scalable for large graphs. It performs well for random looking graphs, but becomes very inefficient for large planar-like graphs. If pinecone is designed for small local networks, it might work well, but it will probably not scale well to large networks with the current design.
- With a small local hack to the Chord network, it is possible to achieve eventual connectivity, see here: https://www.freedomlayer.org/articles/chord_connected_routin...
Other pieces can be found here: https://www.freedomlayer.org/research/ Of special importance are the experiments performed to large networks, which can be found on this page: https://www.freedomlayer.org/research/landmarks-lookahead/
If anyone from the matrix team is reading this, I will be happy to help if any help if needed.