Mechanisms that regulate the rate at which senders inject packets or data into a network so that aggregate offered load does not exceed the network's carrying capacity, thereby preventing excessive queuing, packet loss, latency, or collapse.

Principle

Use feedback (implicit or explicit) and capacity estimation to adapt send rates so that demand tracks available service; trade throughput, delay, and fairness via rate/window adjustments and active queue management.

Demonstration

A TCP sender reduces its congestion window when it observes packet loss or ECN marks, then probes for available capacity by slowly increasing the window; a router applies RED or AQM to mark/drop packets before queues grow unbounded.

Misapplication

Applying only static rate limits without feedback, treating any packet loss as persistent network failure, or applying endpoint-only controls when in-network AQM is required, leading to oscillation, unfairness, or underutilization.

Consequence

When correctly applied, traffic avoids collapse, resources are shared more fairly, end-to-end latency is bounded by queue management, and throughput approaches available capacity subject to fairness constraints.

Reversal

Absent congestion control, senders continue to increase load until queues overflow and many packets are dropped, causing throughput collapse and widely increased latency.

Boundary

Concerns packet networks and layers that control transmission rates (transport, adaptive applications, and network devices); does not subsume application-specific flow control that addresses semantic ordering without regard to network capacity.

Semantic Tension

Often confused with flow control (receiver-driven pacing) and traffic shaping (policing enforced at ingress); congestion control emphasizes system-wide feedback and capacity matching rather than per-connection correctness.

Synthesis

Congestion control is the family of feedback-driven algorithms and queue management techniques that adapt sender behavior to network capacity, balancing utilization, latency, and fairness to avoid collapse while preserving throughput.