QoS Metrics

1. QoS metrics

Quality of Service (QoS) is described through measurements. QoS metrics are quantifiable parameters of network behavior during data transmission. Network administrators use them to diagnose problems, configure policies, and check whether the network meets application and user requirements. The same metric can have different significance for a call, a file transfer, and an interactive game.

Measurements help distinguish bandwidth problems from delay and irregular packet delivery. They connect observed video-call quality with a specific network parameter. A result indicates whether a problem concerns link capacity, transit time, timing variation, or packet delivery. Comparing results at several points can identify the path segment associated with degradation. Four commonly used QoS metrics are bandwidth, latency (delay), jitter, and packet loss.

2. Bandwidth and throughput

Bandwidth is the maximum capacity of a link. It is expressed in bits per second (bps), with prefixes such as kilobits (kbps), megabits (Mbps), and gigabits (Gbps). Higher bandwidth permits more bits per unit of time, but it does not by itself describe the result of a transmission.

Throughput is the measured rate of successful data transfer. Bandwidth describes the link limit, while throughput also depends on congestion, protocol overhead, delay, and packet loss. It is therefore an observation from a particular transmission rather than the interface capacity itself. Transport windows and available buffers also affect throughput.

Bandwidth in QoS policies

Bandwidth is a finite resource. When several applications use a congested link, QoS mechanisms can allocate its capacity according to policy. A minimum allocation requires a traffic profile and conditions for its use:

• Bandwidth guarantee: A reservation mechanism can assign minimum bandwidth to an application. A high-definition video-conferencing system might require $2$ Mbps. For traffic that conforms to its profile, the reservation maintains that share while other data is served.
• Bandwidth limiting: QoS can limit the maximum bandwidth available to lower-priority applications. A large file transfer or peer-to-peer application then uses its configured share, leaving capacity for other traffic. The limit reduces the effect of one stream on other applications.

3. Latency (delay)

Latency (or delay) measures the one-way time for a packet to travel from sender to receiver. RFC 7679 defines this one-way measurement. End-to-end delay includes the components encountered on successive path segments. The interpretation depends on measurement points and clock synchronization.

Latency components

• Transmission Delay: The time it takes for a router to push all the bits of a packet onto the network link. It depends on the packet's size and the link's bandwidth. A larger packet on a slower link will have a higher transmission delay.
• Propagation Delay: The time it takes for a signal to travel the physical length of the medium. This is constrained by the speed of light in the medium (e.g., optical fiber or copper wire). This delay increases over long distances, such as transcontinental or satellite links.
• Processing delay: The time a router takes to process a packet's header, check for errors, and select the next interface. The value depends on the device and the functions it performs. Header checks and route selection can take different times in different devices.
• Queueing Delay: The time a packet spends waiting in a buffer (queue) at a router before it can be transmitted. This delay is highly variable and is the primary cause of latency in a congested network. When more traffic arrives at a router than can be sent out immediately, queues form, and packets have to wait their turn. Scheduling and buffer size affect this component.

Latency in QoS policies

Interactive real-time applications react strongly to delay. QoS mechanisms mainly manage queueing delay through queue selection and scheduling. ITU-T G.114 gives 150 ms as a reference value for one-way voice delay in many connections. The value concerns the complete voice path under assessment, so it is not a standalone limit for every network. A complete assessment also includes terminal and codec delay.

4. Jitter, or delay variation

Jitter, or delay variation, describes differences in the transit time of selected packets. RFC 3393 defines IP Packet Delay Variation as the difference between one-way delays for a selected pair of packets. Variation can result from queues, route changes, and traffic treatment. In an IPDV measurement, the selected packet pair and measurement points affect the result.

Jitter in QoS policies

Delay variation makes it harder for real-time applications such as VoIP and video conferencing to play a continuous stream. When packets arrive at different intervals, the receiver may lack the next audio or video fragment at playback time. The result is interruptions in sound and frozen video.

An application can use a jitter buffer that delays the first packet and stores subsequent packets. The buffer absorbs some arrival-time differences, but increases total delay. It can hide short differences between arrivals, but it cannot recover packets that were permanently lost. The value 6.5 ms in the table is an illustrative target for the video example. The requirement depends on the codec, playback method, and receiver buffer.

5. Packet loss

Packet loss means that a transmitted packet does not arrive at its destination within the time specified by the measurement method. Queue overflow during congestion is a common cause. Link errors and changes in network state also contribute. RFC 7680 defines a one-way packet-loss measurement. The result should be related to a defined direction and packet population.

Application responses to packet loss

The impact of packet loss depends heavily on the transport protocol being used by the application:

• TCP applications, such as file transfer, email, and web browsing: TCP uses acknowledgments and sequence numbers. After detecting a missing segment, the sender retransmits the data. The application preserves stream integrity. Retransmissions reduce throughput and extend the transfer. This mechanism transfers the cost of loss to waiting time and retransmission.
• UDP applications, such as VoIP, online gaming, and live streaming: UDP provides no built-in retransmission. A lost packet is not recovered by UDP itself. An application can add its own recovery method. In a call, loss produces an interruption, while in a video stream it produces a missing image fragment. The effect depends on the codec and the method used to conceal missing data.

Packet loss in QoS policies

QoS can limit loss through traffic prioritization and queue selection. During congestion, a policy can drop lower-priority packets first and protect real-time traffic. The value 1% in the table is an illustrative target for the VoIP example. Requirements depend on the codec, recovery mechanism, and transmission conditions. A quality assessment also specifies the observation period and measurement points.

6. Illustrative application requirements

The table compares example application requirements across the four metrics. The values are planning targets; the choice depends on the codec, transport protocol, receiver buffer, and network conditions. The table values do not replace measurements for a specific service. In practice, measure both directions under a load representative of the planned use. A single sample does not describe service variation over a longer observation window. Repeated measurements support capacity planning and diagnosis over longer time periods.