2.2.1 The Congestion Collapse

When a network is congested, packets are dropped. The "sender" (TCP) then retransmits the packets, which adds more traffic to the already congested network. If not stopped, this leads to Congestion Collapse, where the actual useful data delivered (Goodput) drops toward zero even as the link stays 100% busy.

2.2.2 Congestion Control vs. Flow Control

• Flow Control: Prevents a fast sender from overwhelming a slow receiver (End-to-end).
• Congestion Control: Prevents a group of senders from overwhelming the network infrastructure (Intermediate nodes).

2.2.3 Reactive Congestion Control

Reactive strategies respond after congestion is detected.

1. Choke Packets: A router sends a special packet back to the source saying "Slow down!"
2. Implicit Signaling: The sender notices packet loss (timeout) or increased delay and automatically reduces its transmission window.
3. Explicit Signaling (ECN): Routers mark bits in the packet header to indicate "I am getting full" without actually dropping the packet.

3.2.4 Proactive Congestion Control (Avoidance)

Proactive strategies try to prevent congestion before it starts.

1. Admission Control: Rejecting new connections if the network is near capacity.
2. Resource Reservation: Hard-coding bandwidth for specific high-priority flows (e.g., RSVP).
3. Traffic Polishing: Dropping or marking packets that exceed a pre-agreed rate.

2.2.5 Case Study: The 1986 ARPANET Collapse

In October 1986, the throughput of the ARPANET link between UC Berkeley and Lawrence Berkeley Laboratory dropped from 32 Kbps to 40 bps.

• The Cause: Lack of congestion control in the original TCP implementation.
• The Fix: Van Jacobson developed the Slow Start and Congestion Avoidance algorithms that still power the modern internet.