congestioncontrol

Congestion control is a family of mechanisms used in computer networks to regulate the traffic entering a network so that resources such as bandwidth and buffers are not overwhelmed. The goal is to keep networks utilized efficiently while keeping packet loss, delays, and jitter within acceptable bounds. Congestion control works by adjusting the rate at which a sender transmits data based on observed network conditions, typically using feedback from the network or inferred signals.

Most transport-layer congestion control schemes are end-to-end and operate on the sender's congestion window (cwnd). Early designs used slow start and congestion avoidance, applying an additive increase/multiplicative decrease (AIMD) rule: increase cwnd gradually when the path appears uncongested, and reduce it sharply when congestion is detected, e.g., via packet loss. Modern schemes refine these ideas and may include delay-based components or explicit congestion notification.

Congestion control can be implemented as end-to-end, relying on receiver feedback (ACKs), or with network-assisted mechanisms such as Explicit Congestion Notification (ECN) and Active Queue Management (AQM) in routers. AQM schemes like RED and more modern approaches aim to preempt congestion by signaling congestion before queues fill, reducing packet loss and controlling latency.

Common examples include TCP variants such as Tahoe, Reno, and NewReno, and more recent implementations like CUBIC and Google's BBR, which use different models to estimate available bandwidth and RTT.

Congestion control is also employed by other transport protocols such as QUIC and SCTP. Challenges include bufferbloat, unfairness between flows with different RTTs, interaction with wireless links, and ensuring performance across diverse network conditions. Ongoing work combines end-to-end strategies with network-assisted feedback to improve robustness and efficiency.