Resource Allocation for Service Chaining in Multi-layer Edge-Cloud Networks

  • Yu Bi

Student thesis: Doctoral ThesisDoctor of Philosophy (PhD)

Abstract

Architectural innovation is one of the leading development directions of the telecommunication network. It overcomes barriers left by the traditional network, such as inefficient resource usage, costly network growth, and ossification service levels. Network Function Virtualisation (NFV) plays a crucial role in promoting architecture innovation by disaggregating hardware and software. In NFV-enabled architecture, end-to-end (E2E) network services can be deployed flexibly as ordered Service Function Chains (SFCs). Another critical technology to stimulate architectural innovation is Multi-access Edge Computing (MEC), which brings computing resources to the edge to reduce latency, reduce load, and improve performance and user experience. To gain the advantage of architectural transformation, the resource allocation problem for SFCs has been extensively investigated. However, there is still a gap in serving a new class of advanced high capacity and ultra-low latency services. Motivated by the emergence of diversified and sometimes extreme service Quality-of-Service (QoS) requirements, the necessity for resource management on densely developed but geographically distributed MEC nodes, as well as the difficulty of dealing with complex and rapidly changing network and environment, this thesis focus on solving the resource allocation problem for QoS-aware SFCs in multi-layer edge-cloud networks.

This thesis creates a comprehensive road map for addressing the QoS-aware SFCs resource allocation problem under various scenarios. The developed Mixed Integer Linear Programming (MILP) model can find optimal solutions for small-scale networks. Meanwhile, the proposed heuristic and Deep Reinforcement Learning (DRL) approaches can deliver solutions in a reasonable time for large-scale networks. For centralised control, the designed meta-heuristic and multi-objective DRL algorithm can balance service performance and resource utilisation. For decentralised control, the adopted congestion game model and multi-agent DRL model assure privacy and scalability. For ultra-low latency services, adding an optical layer reduces the transmission and queueing latency on intermediate switches, and proposed algorithms achieve impressive service acceptance performance. In practical scenarios, on the one hand, real test-bed experiments bridge the gap between theoretical and practical algorithms' performance. On the other hand, provided online solutions cater to dynamic and unforeseeable service requests and network environments. The thesis contributions on the resource allocation for SFCs in multi-layer edge-cloud networks are of significant value to unleash the great potential of the evolved telecommunication network.
Date of Award1 Aug 2022
Original languageEnglish
Awarding Institution
  • University of Bristol
SupervisorReza Nejabati (Supervisor) & Dimitra Simeonidou (Supervisor)

Cite this

'