TCP slow-start
Begin with a small congestion window; double per RTT until loss.
Origin
Van Jacobson, "Congestion Avoidance and Control," SIGCOMM 1988. Designed in response to the 1986 Internet congestion collapse where throughput dropped 1000×.
Where it shows up in production
- Every modern TCP stack Slow-start with initial cwnd of 10 segments (RFC 6928, 2013) is universal.
- HTTP keep-alive matters A fresh connection pays the slow-start tax; long-lived ones are already warmed.
On Semicolony
Sources & further reading
Found this useful?