Predictive Service

Description

The Predictive service is a Quality of Service (QoS) model defined in 3GPP specifications 21.905 and 22.060. It represents a middle ground between best-effort and guaranteed services, offering reliable performance while allowing a specified variance in the measured performance criteria. This service model is particularly relevant for applications that require predictable network behavior but can tolerate some fluctuation in parameters such as throughput, delay, or jitter. The concept was introduced in Release 4 and has been maintained through subsequent releases, reflecting its ongoing relevance in network design.

In terms of architecture, the Predictive service is typically implemented through a combination of traffic classification, policing, and scheduling mechanisms within the core network and radio access network. The network operator defines a target performance level and an acceptable variance range. For example, a service might guarantee an average throughput of 1 Mbps with a variance of ±10%. The network then uses QoS parameters such as Guaranteed Bit Rate (GBR) and Maximum Bit Rate (MBR) to enforce these targets. The predictive nature comes from the network's ability to statistically ensure that the performance stays within the specified bounds, rather than providing a hard guarantee.

Key components that enable Predictive service include the Policy and Charging Control (PCC) framework, which allows operators to define QoS policies per service data flow. The PCRF (Policy and Charging Rules Function) or its 5G equivalent, PCF (Policy Control Function), communicates these policies to the gateways (P-GW, SMF/UPF) and base stations (eNodeB, gNB). The base station scheduler then allocates radio resources to meet the predictive targets, using algorithms that balance between different users and services. The variance allowance provides flexibility to the scheduler, which can be important in wireless environments where channel conditions fluctuate.

The Predictive service plays a crucial role in supporting applications such as video streaming, VoIP, and interactive gaming, where a certain level of predictability is required for good user experience, but strict guarantees are not necessary. It allows operators to offer differentiated services without over-provisioning network resources. Compared to a guaranteed service, which might require dedicated resources and strict admission control, the predictive service can achieve higher statistical multiplexing gains, leading to more efficient network utilization.

In the context of 5G and beyond, the Predictive service concept aligns with the broader goal of supporting diverse use cases with varying QoS requirements. It can be mapped to 5G QoS Identifiers (5QIs) that are configured with specific packet delay budgets, error rates, and guaranteed flow bit rates. The service model continues to be referenced in 3GPP specifications, ensuring backward compatibility and providing a foundation for more advanced QoS mechanisms.

Purpose & Motivation

The Predictive service was created to address the need for a QoS model that offers more reliability than best-effort service but is less rigid than a guaranteed service. Before its introduction, network operators had limited options: they could either provide best-effort delivery, which offered no performance assurances, or hard guarantees, which were resource-intensive and often impractical in wireless networks due to inherent variability. The Predictive service fills this gap by allowing a specified variance in performance criteria, thus accommodating the stochastic nature of radio channels while still providing a predictable experience.

This service model solves the problem of efficiently supporting applications that require consistent performance but can tolerate some fluctuation. For instance, a video streaming service might need a certain average bitrate to maintain quality, but temporary drops can be buffered. By allowing a variance, the network can statistically multiplex resources among multiple users, improving overall capacity. This is particularly important in cellular networks where spectrum is limited and user mobility causes channel conditions to change rapidly.

Historically, the Predictive service was introduced in Release 4 as part of the early efforts to define QoS frameworks for UMTS. It has been carried forward through all subsequent releases, demonstrating its enduring value. The concept has influenced later QoS developments, such as the introduction of GBR and non-GBR bearers in LTE and the 5QI framework in 5G, which incorporate similar ideas of statistical guarantees and variance tolerance.