Congestion & packet queues

Its job in the system

Packets arrive in bursts, while each outgoing link can transmit only at its available rate. A queue absorbs some temporary mismatch between arrivals and service. If offered traffic persistently exceeds what the bottleneck can handle, a finite buffer eventually fills or an active management mechanism intervenes. A connected path can therefore be slow without being broken.

How it works

A larger buffer can hold more waiting packets but does not increase the outgoing link’s capacity. It can instead lengthen delay when the buffer remains occupied. Dropping packets or, where supported, marking congestion gives responsive senders information that the path needs a lower sending rate. Active queue management aims to signal before persistent queues cause excessive delay.

Congestion control at an endpoint uses feedback to adapt transmission. The detailed behaviour depends on the transport and algorithm. It is different from receiver flow control, which limits data to what the receiving endpoint can accept. A sender may be constrained by either mechanism, and the smallest sustainable bottleneck along the path can dominate an end-to-end transfer.

What it connects to

Queues connect access-network capacity, routers, transport recovery, and user-perceived latency. Uploads can compete for the outgoing connection, and a link failure can concentrate traffic onto a surviving route. Having an alternate route therefore establishes neither spare capacity nor unchanged delay. Routing reachability and performance require separate questions.

Different places, different designs

Queue discipline, traffic scheduling, explicit congestion signalling, and congestion-control algorithms vary between networks and devices. Different flows do not automatically receive identical shares. The simple scenarios on this site show selected relationships rather than reproducing a particular operator’s scheduler or a complete TCP or QUIC implementation.