Kademlia has an interesting property most folks ignore, but until circa 2000 nodes, the performance is not great. Not broken but flakey. What Engineers can tend to do is try to “fix” that, and that is a slippery slope.
Like an ant colony that until it’s about 1500 ants is also flakey, kademlia needs
- A decent size network (say 2000 nodes)
- Nodes to be communicating across the network (i.e. our DBC store means nodes need to check the parent is Spent. That performs the cross-network lookups to keep routing tables fresh)
This is why doing a bit of Safe is not great. That cross-network chatter is important. The same goes for network size.
The network size is why I am keen on quic and not TCP. quic is much more resource friendly and running several hundred nodes per computer is easy.