Cost-effective and Energy-efficient Resource Allocation in Virtualized Telecommunication Networks

Abstract

Virtualization technologies have transformed telecommunication networks through abstraction of physical hardware as virtual resources. A central challenge in this paradigm is the Virtual Network Embedding (VNE) problem, which involves the optimal allocation of resources from substrate networks to service requests in the form of Virtual Networks (VNs). This thesis investigates two specific VNE applications across heterogeneous network environments: (i) network slicing in edge networks and (ii) VNE in cloud Data Centers (DCs) with composable/disaggregated infrastructure, with an emphasis on profit maximization and energy efficiency improvement.

The rise of latency-sensitive and bandwidth-intensive applications (e.g., autonomous driving, augmented reality) has driven computation closer to data sources via edge computing. Multi-access Edge Computing (MEC) extends this paradigm by equipping edge nodes, such as base stations and access points, with computing and storage resources, thereby reducing latency and alleviating network congestion. In this thesis, we address network slicing in a large-scale MEC-enabled Radio Access Network (RAN) comprising heterogeneous edge nodes with varying computing and storage resource capacities. These resources are dynamically allocated to slice requests and released when the service of a slice request is completed. Our objective is to optimize the resource allocation for each admitted arriving slice request, considering its demands for computing and storage resources, to maximize the long-run average Earning Before Interest and Taxes (EBIT) of the MEC slicing system. We formulate the optimization problem as a Restless Multi-Armed Bandit (RMAB)-based resource allocation problem with a nonlinear cost rate function. To solve this, we introduce a new policy called Prioritizing-the-Future-Approximated earning per request (PFA) where for each admitted slice request, we always prioritize the allocation of the resource combination that gives the highest achievable earning, considering the future effects of this allocation. PFA is designed to be scalable and applicable to large-scale networks. We numerically demonstrate the superior performance of PFA in maximizing long-run average EBIT through simulations, comparing it with two baseline policies under various combinations of parameter values. Moreover, our findings offer insights for network operators in resource allocation policy selection.

While edge computing shows potential to support real-time applications through low-latency processing, resource-intensive workloads still need to be handled by large-scale cloud data centers. Motivated by this, we also address a specific VNE problem in the context of Composable/Disaggregated Data Center (DDC) networks, characterized by the decoupling and reassembly of different resources into resource pools. Existing research on the VNE problem within DC networks primarily focuses on the Server-based DC (SDC) architecture. In the VNE problem within SDCs, a virtual node is typically mapped to a single server to fulfill its requirements for various resources. However, in the case of DDCs, a virtual node needs to be mapped to different resource nodes for different resources. We aim to design an optimization method to achieve the most efficient VNE within DDCs. To this end, we provide an embedding scheme that acts on each of the arriving VN requests to embed the VN with minimized power consumption. Through this scheme, we demonstrate that we also achieve a high long-term acceptance ratio. We provide Mixed-Integer Linear Programming (MILP) and scalable greedy algorithms to implement this scheme. We validate the efficiency of our greedy algorithms by comparing their performance against the MILP for small problems and demonstrate their superiority over baseline algorithms through comprehensive evaluations using both synthetic simulations and Google cluster trace-driven test.

The power consumption associated with VN requests already being serviced in a cloud DC can be further reduced through workload consolidation. In this thesis, we investigate the workload consolidation problem for VNE in DDCs, consolidating workloads scattered across underutilized resource nodes onto fewer nodes. There are two objectives, to minimize power consumption and to minimize migration cost. We formulate the problem as a MILP and generate a Pareto front that provides a set of optimal solutions reflecting different trade-offs between the two objectives.

Date of Award	12 Jan 2026
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Chi Hung Sammy CHAN (Supervisor), Jing Fu (External Co-Supervisor) & Moshe ZUKERMAN (Co-supervisor)