What is QoS?

QoS (Quality of Service) is the general name for treating some network traffic differently from the rest: limiting a link to a set rate, giving voice and video priority, reserving bandwidth, or sharing a connection fairly between users.

All of it comes down to queues. A network only needs QoS where traffic arrives faster than it can leave, typically at the slow link between an office and the internet, because that's where packets wait. QoS decides how long they wait and in what order. The main tools are classification (sorting packets into classes, often by the DSCP field that DiffServ defines in the IP header), shaping (sending no faster than a set rate, so a queue forms where you control it), policing (dropping traffic above a rate, which is harsh on TCP), scheduling (deciding which class or flow goes next), and active queue management (AQM), which keeps each queue short.

On a small office or home router, the most effective QoS is usually the simplest: shape the internet link just below its real speed with CAKE, which fixes bufferbloat and shares the line fairly without any traffic rules. Elaborate priority schemes help only if traffic is marked consistently, and markings are often missing or rewritten across the internet. Inside a network, QoS matters most for voice, video and storage traffic on links that are sometimes full. The Traffic shaping with CAKE on Debian lesson measures latency under load before and after.