Quantifying the Impact of 5G Network Slicing on Quality of Experience and Survivability

Abstract

Beyond 5G (B5G) networks provide support for a diverse array of services with heterogeneous, often very stringent, and varying requirements. This versatility places the modern mobile network at the cornerstone of the ongoing transformation towards digitalized economy and society.

Mobile network operators aim at delivering good performance of the services utilizing their network. Considering that modern services are getting more personalized and immersive, relying solely on objective metrics like Quality of Service (QoS) to evaluate network performance is no longer sufficient, and more user-aware metric are needed. For this purpose, Quality of Experience (QoE) can be used, representing an overall degree of users’ satisfaction with the provided service, and representing a key factor for success and user retention in a highly competitive market.

Furthermore, due to the support for critical services, it is essential that the networks provide good service quality even in adverse circumstances caused by massive outages. Those services include public safety and emergency communication, transportation, power grid and telemedicine. Under large scale outages caused by natural disasters, attacks or hazards, communication services can be crucial for first responders and rescue mission. Therefore, the operators should ensures uninterrupted connectivity and service availability even during critical times. This nonfunctional property of the network infrastructure is called survivability.

Historically, mobile networks were heavily based on the economy of scale design principle, to optimize costs and increase efficiency. However, due to extremely stringent requirements of the services, a design shift towards more differentiated traffic management is needed including advanced resource allocation mechanisms that go beyond best-effort approaches. QoS Flows and network slicing have been identified as key enablers for quarantining specific QoS and QoE requirements, in an efficient and flexible manner.

QoS Flow is the finest granularity for differentiated QoS forwarding treatment in B5G networks, which given the number of flows in nowadays network easily results in scalability and configuration complexity challenges. A more coarse granular and scalable approach to traffic differentiation can be achieved with slicing, where several flows can be mapped into one slice. A network slice is defined as a logically isolated network partition, existing on top of a shared physical infrastructure and providing specific network capabilities.

The thesis takes a dual perspective on QoS Flows and network slicing by focusing (i) on a user-centric network and service optimization by utilizing the Quality of Experience (QoE) for normal operation, and (ii) including the user-centric approach in the context of network survivability, to ensure uninterrupted connectivity and design optimized recovery strategies in case of massive outages. The contributions are multi-faceted, including both methodological advancements and research tools needed for evaluation of performance under normal and adverse circumstances, as well as performing the evaluations and quantifying the impact of different slicing configurations on Quality of Experience (QoE), system utilization and survivability.

Although there exist high-level standards related to the architecture of network slicing formulated by 3GPP and ETSI, the details of the actual implementation of network slicing are often hidden behind commercial solutions or business policies of the network operators. This hinders a comprehensive evaluation of the performance of network slicing. Therefore, the first major contribution of this thesis is in the methodology to implement and evaluate network slicing, in a reproducible and scalable manner. Such methodology and its key design principles should conform with key concepts of network slicing, including the ability to maintain a satisfactory QoE for heterogeneous services, guarantee adequate allocation and sharing of the resources. For this purpose the related contributions utilize prominent network simulator OMNeT++, and custom built module implementing the logic of Hierarchical Token Bucket (HTB). The HTB is chosen due to its hierarchical tree structure that mimics the way network slices can be managed under a common shared link (root node), as well as throughput guarantees evolving from its high rate conformance. Such an approach allows for a broad spectrum of evaluations of non-functional properties, including performance and survivability.

The second major contribution of this dissertation is the quantification of the impact of network slicing on QoE and system utilization, performed through comprehensive evaluations using the established simulation environment. To cover a wide range of application types, both bandwidth-hungry and delay-sensitive applications were considered, including video streaming, file download, Voice-over-IP (VoIP) and SSH. A key outcome of those studies is that network slicing and QoS Flows can be used to satisfy and guarantee good service QoE, while at the same time allowing for higher traffic load (more admitting clients), resulting in more efficient and sustainable usage of the available resources. In addition, a set of often neglected configuration parameters for slicing is evaluated, leading to the creation of guidelines for slice orchestration resulting in optimized QoE and system utilization. However, merely analyzing the impact of network slicing on QoE in fully operational networks is not sufficient and performance should also be evaluated in adverse circumstances, since the availability and quality of communication services can play a crucial role for the first responders in cases of natural disasters. Therefore, the survivability of different slicing configurations and their performance when unwanted events occur, should be evaluated. For this purpose, this thesis also proposes a framework which allows to quantify both spatial and temporal evolution of general network performance under massive outages. The framework allows for modeling dynamic and escalating disasters, as well as a variety of recovery actions (e.g. repair, re-route, relocate). As a last step, this framework is applied to assess the survivability of different slicing configurations, evaluating the performance for various critical and non-critical service types. The insights obtained demonstrate notable survivability differences across various slicing configurations, showing that mapping of the traffic types and criticality, as well as prioritization schemes can improve network performance under massive outages and reduce the time it takes to restore their connectivity.

To summarize, the present thesis contributes to the quantification of the impact of network slicing on QoE and survivability. With respect to the methodological aspects of the contributions, two milestones are achieved: (1) simulation environment for evaluations of network slicing, and (2) framework for quantification of the network survivability under various disaster types and recovery actions. Using this methodology important insights into the relationship between slicing and QoE, system utilization and survivability are obtained, focusing on the three key elements identified by Industry 5.0: human-centric, sustainable, and reliable systems.