Open-Loop vs Closed-Loop Congestion Control

Congestion-control policies are classified by when they act. Open-loop policies are decided before traffic is sent and never use feedback from the network; closed-loop policies rely on feedback to detect congestion that has already occurred and then correct it.

Basis	Open-loop congestion control	Closed-loop congestion control
Principle	Prevent congestion by design; no feedback is used	Detect congestion, feed information back to the sender/routers, then react
When applied	Before/during policy setup, ahead of transmission	Continuously during transmission, once per RTT
Feedback	None - the policy is blind to current load	Implicit (ACK rate, RTT, packet loss) or explicit (router marking, ICMP)
Goal	Avoid deadlock, starvation and queue overflow a priori	Keep the network in the comfortable operating region after congestion appears
Typical mechanisms	Retransmission policy, acknowledgment policy, error-control policy, admission/flow control at connection setup, packet-discard policy (drop-old vs drop-new), FQ/WFQ scheduling, differentiated service classes, traffic shaping (leaky bucket/token bucket)	Slow start, congestion avoidance (AIMD), fast retransmit and fast recovery (TCP), RED/WRED active queue management, ECN marking, adaptive routing
Response time	Slow - cannot adapt while traffic runs	Fast - adapts every RTT
Complexity	Simple, predictable	More complex; needs signalling between routers, receiver and sender
Where implemented	Mainly in routers and in the sender before sending	In the sender, driven by receiver/router feedback
Example	Shaping traffic at the edge so the core never overloads	TCP hearing three duplicate ACKs and cutting cwnd by half

Key differences in brief

1. Feedback: open-loop is fire-and-forget; closed-loop is a classic control loop (measure -> decide -> act).
2. Timing: open-loop acts at policy/setup time; closed-loop acts during the session and re-evaluates each RTT.
3. Objective: open-loop prevents bad states (guarantees are made so queues cannot overflow); closed-loop repairs them once loss or delay rises.
4. Adaptability: closed-loop can exploit changing conditions (grow the window while ACKs arrive quickly), whereas open-loop must stay conservative to remain safe.
5. Cost of failure: open-loop failures are hard to detect (no feedback), while closed-loop can oscillate around the optimum as it reacts to delayed feedback.

Rule of thumb: if a mechanism must hear from the network before it changes behaviour, it is closed-loop; if its behaviour is fixed in advance, it is open-loop.