Quality of service

Also known as: QoS

On this page 7

Quality of service, usually abbreviated QoS, is the set of mechanisms a network uses to classify traffic and treat each class differently, so that delay-sensitive or critical flows receive the bandwidth, latency, jitter and loss levels they need. In telecommunications, and in the wider field of network performance, the term also describes the measurable quality that a connection actually delivers against those targets.

QoS matters because network capacity is shared. When links become congested, a network without QoS treats a voice call, a video stream and a background software update the same way, and the most sensitive service degrades first. QoS decides who waits and who goes first, and it gives operators a way to promise specific levels of performance.

How quality of service manages traffic

QoS works through a chain of functions applied at each network node. Packets are first classified and marked according to their application or subscriber, then queued and scheduled according to their class, and finally policed or shaped so that no class exceeds its allowance.

In IP networks, the marking usually sits in the DSCP field of the packet header, defined by the Differentiated Services (DiffServ) architecture. Routers read the marking and place each packet in a matching queue. Schedulers such as strict priority, weighted fair queuing or deficit round robin then decide which queue transmits next. An older model, Integrated Services (IntServ) with RSVP, reserves resources per flow but scales poorly in large networks.

The table below lists the core QoS functions and their effect on traffic.

FunctionWhat it doesTypical tool
ClassificationIdentifies the traffic type or subscriberPacket inspection, access lists
MarkingLabels packets with a priority valueDSCP, 802.1p, MPLS EXP bits
Queuing and schedulingOrders transmission by classPriority queue, WFQ
PolicingDrops or remarks excess trafficToken bucket policer
ShapingBuffers bursts to smooth the rateTraffic shaper
Admission controlRefuses new flows when capacity is shortCall admission control
Core QoS functions

The metrics that define QoS

Quality of service is measured with a small set of network-level parameters. Each application tolerates a different range, and a service can fail on one metric while looking healthy on the others.

Latency is the one-way or round-trip delay, often checked with a ping. Jitter is the variation of that delay, which disturbs real-time audio and video. Packet loss is the share of packets that never arrive, and throughput is the useful data rate the connection sustains. Availability, the share of time a service works, is often added in contracts.

ApplicationOne-way latencyJitterPacket loss
Voice call (VoLTE, VoIP)Below about 150 msBelow about 30 msBelow 1%
Interactive video callBelow about 150 msBelow about 30 msBelow 1%
Online gamingBelow about 50 msLow and stableBelow 1%
Buffered video streamingSeconds toleratedAbsorbed by bufferUp to a few percent
Bulk downloads and backupsNot criticalNot criticalRecovered by TCP
Typical QoS targets by application

These values follow widely used guidance such as ITU-T G.114 for voice delay, but they are targets rather than hard limits. Codecs, jitter buffers and error concealment can hide moderate problems, while the same network may feel poor for a demanding cloud game.

QoS in mobile networks

Mobile networks implement QoS through bearers in LTE and QoS flows in 5G. Each flow carries a QoS identifier, called QCI in LTE and 5QI in 5G, that tells the base station and core network which priority, delay budget and loss rate to apply.

A key distinction is between Guaranteed Bit Rate (GBR) flows, which reserve capacity, and non-GBR flows, which share what remains. Voice over LTE uses a GBR bearer, which is why calls stay clear in a busy cell where web browsing slows down. In 5G, network slicing extends the idea by giving whole groups of services their own logical network with separate QoS rules.

5QIResource typePacket delay budgetExample service
1GBR100 msConversational voice
2GBR150 msConversational video
5Non-GBR100 msIMS signaling
9Non-GBR300 msDefault internet traffic
82Delay-critical GBR10 msDiscrete automation
Selected standardized 5QI values

Radio conditions still set the ceiling. A user at the cell edge with low SINR receives fewer resources per transmission, so even a high-priority flow can suffer when signal quality is poor.

QoS is often confused with neighboring ideas in network performance. The differences matter because each concept answers a different question and is measured in a different way.

ConceptQuestion it answersHow it is measured
Quality of serviceDoes the network meet its technical targetsLatency, jitter, loss, throughput
Quality of experienceIs the user satisfied with the serviceMean opinion score, surveys
Network capacityHow much traffic can the network carryAggregate cell or link throughput
Mobile data throttlingIs a user’s speed deliberately cappedPolicy rules, measured rate limits
Speed test resultWhat rate is achieved at one momentSingle download and upload test
QoS and related concepts

Quality of experience (QoE) is the user’s perception, while QoS is the technical delivery behind it. Good QoS usually produces good QoE, but not always: a video app with a large buffer feels fine on a jittery link, and a perfect link cannot rescue a poorly designed application.

Limitations and common misconceptions

The most common misconception is that QoS creates bandwidth. It does not; it only redistributes scarce capacity during congestion. On a lightly loaded link, QoS has almost no visible effect, and on a link that is permanently overloaded, it can only decide which services fail first.

  • QoS ends at the network boundary: markings are often reset when traffic crosses the public internet, so end-to-end guarantees exist mainly inside a single operator or managed network.
  • Encryption limits classification: when traffic is encrypted, networks must rely on markings, addresses or flow patterns rather than content.
  • Net neutrality rules restrict prioritization: in many jurisdictions, operators may manage traffic for technical reasons but may not favor commercial services arbitrarily.
  • Misconfiguration causes harm: too many traffic classes or an oversized priority queue can starve other traffic and make the network slower overall.

Effective QoS therefore combines sound capacity planning with a small number of clearly defined classes, consistent marking at the network edge and regular measurement against the targets.

A plain-language summary

Quality of service is the network’s way of deciding which data goes first when the road gets busy. It works like an emergency lane on a highway: ambulances, here phone calls and video meetings, get through quickly, while trucks carrying large downloads can wait a little without anyone noticing.

QoS cannot make the highway wider. It only organizes the traffic fairly so that the things that suffer most from delay keep working, even when many people use the network at the same time.

An example from practice

A regional hospital runs voice calls, nurse call alerts, medical imaging transfers and guest Wi-Fi over one shared network with a 1 Gbit/s uplink. During the morning shift, large image uploads saturate the link, and staff report choppy calls and delayed alerts.

The network team defines four classes. Voice receives the Expedited Forwarding marking and a strict priority queue capped at 15% of the link, alerts get the next priority, imaging uses an assured class with a guaranteed share, and guest Wi-Fi is shaped to 100 Mbit/s as best effort. The mobile operator also provides a private 5G slice with a GBR profile for the handheld devices used by nurses.

After the change, measured voice jitter falls from about 45 ms to below 10 ms, packet loss on calls drops near zero, and imaging transfers take only slightly longer at peak times. The outcome shows the core purpose of QoS: critical services stay reliable without buying extra capacity.