Fallback Bottleneck

[neonphog]

If a connection fails to establish over webrtc and switches to fallback messaging, it can quickly overwhelm the SBD rate-limit making it almost impossible to establish new connections that might otherwise be able to correctly connect over webrtc.

Options:

• Consider if this fallback mechanism is even a requirement. Maybe webrtc connectivity is good enough without it? Maybe it is okay if Holochain cannot be run from within a UDP-eating convention center firewall?
• Consider treating connectivity messages as higher priority than fallback messages, something like:
  • divide the available rate-limit in half,
  • allocating only half for fallback messaging, and divide that by connection count that has fallen back
  • always send ice messages, relying on the sdb-client to hit its internally tracked RL, but still count those bytes for fallback messaging RL