Estimate network congestion level from link utilization and traffic load.
Network congestion occurs when offered traffic load approaches or exceeds link capacity. As utilization rises above ~70%, queuing delays increase nonlinearly. When load exceeds capacity, excess packets are dropped (tail-drop) or marked (ECN), causing retransmissions that further increase effective load.
Link utilization
utilization% = offered_load / link_capacity * 100
Queuing theory (M/M/1 model) shows that as utilization approaches 100%, average queue length and delay grow toward infinity — small increases in load cause disproportionately large increases in delay, which is why keeping utilization below 70-80% is a common engineering target.
TCP congestion control algorithms (Reno, CUBIC, BBR) reduce sending rate when packet loss or increased delay signals congestion, then gradually probe for available capacity — this feedback loop is what prevents network collapse but also causes throughput fluctuations.