End-to-end latency

Description

End-to-end latency is a fundamental quality-of-service (QoS) metric in 3GPP networks, defined as the time that takes to transfer a given piece of information unidirectionally from a source to a destination, measured at the communication interface, from the moment it is transmitted by the source to the moment it is successfully received at the destination. This definition, standardized in 3GPP specifications such as TS 22.261 and TS 22.104, provides a precise and measurable way to characterize the delay experienced by data packets as they traverse the network from an application at one endpoint to an application at another endpoint. It is a key performance indicator (KPI) for many services, especially those with stringent real-time requirements like industrial automation, autonomous driving, and immersive media.

The concept of end-to-end latency encompasses all the delays introduced by the network and the endpoints. These include processing delays at the source and destination, transmission delays over the radio and backhaul, queuing delays in network nodes, and propagation delays. In 3GPP, the latency is typically measured at a reference point, such as between the UE and the application server, or between two UEs. The measurement is unidirectional, meaning it captures the delay in one direction of a data flow, which is important for applications that are sensitive to asymmetric delays. The definition explicitly states that the measurement is taken at the communication interface, which means it includes the time from when the packet is handed to the communication stack at the source until it is delivered to the application at the destination.

3GPP has defined various latency requirements for different service categories. For example, ultra-reliable low-latency communication (URLLC) services in 5G target end-to-end latencies as low as 1 ms for some industrial use cases, while enhanced mobile broadband (eMBB) services may tolerate tens of milliseconds. The specifications TS 22.261 (Service requirements for the 5G system) and TS 22.104 (Service requirements for cyber-physical control applications in vertical domains) provide detailed latency budgets and guidelines. The network architecture, including the radio access network (RAN) and core network (CN), is designed to minimize latency through techniques such as edge computing, optimized scheduling, and reduced processing times.

Key components that influence end-to-end latency include the radio interface (e.g., numerology, slot length, HARQ round-trip time), the transport network (fiber, microwave), the core network (UPF placement, session management), and the application servers. In 5G, the introduction of the service-based architecture and network slicing allows operators to tailor latency performance for specific slices. The 5G QoS model uses 5QI (5G QoS Identifier) values that map to latency requirements, and the network can enforce these through scheduling and resource allocation. Measurement of end-to-end latency is typically done using timestamps at the source and destination, and may involve synchronization protocols like PTP (Precision Time Protocol) for accurate one-way delay measurement.

End-to-end latency is a critical enabler for many advanced use cases. In industrial automation, low latency is essential for closed-loop control; in vehicular communication, it is vital for safety; in cloud gaming and AR/VR, it directly affects user experience. The 3GPP specifications continuously evolve to support lower latency, with each release introducing enhancements. The definition has remained largely consistent, with minor wording adjustments to clarify unidirectionality and measurement points.

Purpose & Motivation

End-to-end latency was defined to provide a standardized, measurable metric for the delay experienced by data as it travels from a source to a destination across a communication network. Before its formal definition, latency was often described in vague terms or measured differently by various stakeholders, leading to inconsistent performance claims and difficulty in comparing network capabilities. The 3GPP definition addresses this by specifying that the measurement is unidirectional and taken at the communication interface, ensuring that all parties use the same reference points. This is crucial for services that require deterministic delay, such as industrial control, where a delay of even a few milliseconds can cause production failures.

The motivation for a precise definition also stems from the need to support new vertical industries in 5G. The ITU IMT-2020 requirements include latency targets for URLLC, and 3GPP needed a clear definition to guide network design and verification. The definition helps operators and vendors to design networks that meet specific latency budgets, and it enables the creation of service level agreements (SLAs) with measurable KPIs. It also facilitates the development of latency-critical applications by providing a common understanding of what the network can deliver.

Historically, the term was introduced in Release 15 as part of the initial 5G specifications, though earlier releases had latency requirements for LTE. The definition has evolved slightly: in Rel-15 it did not explicitly state 'unidirectional', but from Rel-16 onwards it was clarified to be unidirectional. In Rel-20, the wording was further refined to remove 'unidirectional' but kept the essence. These changes reflect ongoing efforts to harmonize terminology across specifications and to ensure clarity for measurement and testing.