So where do reliability and congestion control come in?
Network cost will be non trivial
Applications can adapt send rate and to delay variance
In between mean rate and peak, we need to tolerate loss and react to congestion
Protocols for non-time-critical applications can help in that range.