QoS

Quality of Service (QoS) is a set of techniques that give different classes of traffic different treatment on a network. Without QoS, every packet competes for the same queue on each router or switch and is forwarded in the order it arrived. That works while links are uncongested but breaks down under load: interactive traffic such as a voice call shares a queue with a large file transfer and inherits the same delays.

QoS exists because some applications are far more sensitive to delay and loss than others. A VoIP call needs end-to-end latency under roughly 150 ms and very low jitter (variation in packet arrival times); a few percent of packet loss is audible. A bulk download tolerates hundreds of milliseconds of latency and recovers from loss with TCP retransmissions. QoS lets the network favour the first kind of traffic when there is contention, instead of treating both equally.

The usual building blocks are classification, marking, queueing and shaping. A packet is classified by header fields (port, protocol, source) or by deep inspection, marked with a DSCP value in the IP header, sorted into per-class queues at each hop, and possibly rate-limited so it cannot starve other classes. End-to-end QoS requires every hop along the path to honour the markings, which is realistic inside one operator's network but not across the public Internet, where DSCP is often ignored or rewritten at the boundary.

A common real-world example: on a small office connection, the router is configured so VoIP traffic from the phone system goes into a strict-priority queue, while a backup job uploading to cloud storage is placed in a best-effort queue and shaped to leave headroom. The call stays clear even when the backup is saturating the upstream. Inside data centres and carrier backbones, QoS is also used to protect routing-protocol traffic (BGP, OSPF) and management-plane access from being affected by user traffic spikes.