Use UDP for hydra network

[ch1bo]

We drafted an ADR on this already: https://hydra.family/head-protocol/adr/17/

It has been thought of, but we are not quite sure whether we want / need to continue down that road. Let's have a discussion! Anyone seeing pros / cons about this? Anyone preferring UDP over TCP connections in one or the other use case scenario?

[ghost]

One of the big advantage of UDP over TCP is that forming a head can be dynamic, eg. there's no need to set the peers' connection info up front. Each hydra-node listens on a dedicated UDP port when it starts and can communicate this information with peers when one wants to start a head. This communication could happen on-chain, by putting the connection info into the commit validator's datum, possibly encrypted using a dedicated symmetric key provided in the initial's datum and itself encrypted with each peer's cardano verification key. Then the only information needed to start a head would be the hydra and cardano verification key pairs, something which can be completely public and therefore published on the chain or off-chain.

Also UDP is much more malleable than TCP when it comes to relaying and NAT traversal, checkout Tailscale's doc: https://tailscale.com/blog/how-nat-traversal-works/

[ghost]

After my hydra-chess experiment, I am all the more convinced that we need to find a way to make ad-hoc Head networks formation much more dynamic and simple than it is today. A rough design for this is forming in my head with:

• hydra-node listens on a single UDP port and can use hole punching techniques to reach other nodes and form p2p connections
• when a node initiates a head, it also creates a symetric encryption key that's shared with other parties
• each party uses this key to encrypt connectivity information in the commit output datum
• each party can decrypt each other party's datum and use this information to "connect" to them
• (optional) symetric key can also be used to encrypt communication