Mathis Engelbart

Interactive real-time media transport requires specific congestion control algorithms that pose different requirements and challenges from the ones known from TCP. I present what I learned from implementing and testing bandwidth estimation for Pion in Go.

The internet offers an unreliable, best-effort packet transport, where delays,
bandwidths, and loss rates can vary at any time. How is it possible to send
reliable and high-quality real-time media over a network with zero guarantees?

Real-time applications usually use UDP as the underlying transport to avoid
head-of-line blocking and late retransmissions. UDP does not provide any
congestion control, and the congestion control algorithms known from TCP do not
satisfy the requirements of interactive real-time media flows. Most TCP
congestion control algorithms work well for reliable transfers, e.g., files or
web pages, but impose too much packet loss and delays that are too long for
real-time media transmission.

In real-time communication, peers must send data at a valuable bit rate while
keeping transit delays and packet loss low. At the same time, it has to be fair
to other flows on the same link, be it real-time traffic or bulk transfers.

Learn about my journey in building a bandwidth estimator in Go. I will share
fundamental concepts that make it possible to implement congestion control for
real-time media and the software needed to test it.

Mathis Engelbart

Talks