Hi! Thanks for joining @User. Maybe I'm misreading your question, but the central server is not acting like a relay. It's not a hub and spoke model, it's a full mesh. the clients are all set up with direct tunnels to each other. The central server just holds and modifies configs so that the clients know how to find each other. Even if the server goes down, all the clients still have functional tunnels to one another, they just don't get updated (so if clients are added/removed/modified, this will not be recognized by the network). That said, it's bad to have one point of failure, and long term we would like a good solution to this. We would like to add the option for something like Kademlia, but likely would need funding for a developer, unless someone from the community is willing to take it on, as it would be a pretty big overhaul. In the near term, our solution is just going to be adding some form of load balancing for the server, and allowing clients to become failover servers.