Optimal Resource Orchestration of Distributed 5G Applications over the Cloud Continuum

Abstract

The rapid advancement of fifth-generation (5G) wireless network technology has enabled a wide range of innovative applications with diverse requirements, including ultra-low latency, high reliability, and scalability, especially in the context of Internet-of-Things (IoT). To effectively support these applications, efficient resource management and orchestration are crucial. The Cloud Continuum (CC), which encompasses heterogeneous network infrastructures ranging from centralized cloud data centers to edge devices, offers a distributed computing and networking environment to meet the specific needs of 5G IoT applications deployment. However, orchestrating resources in this complex setting poses significant challenges. This thesis aims to address these challenges by proposing novel approaches for the management and orchestration (MANO) of computing and network resources and the CC layers, aligned with established frameworks like the European Telecommunications Standards Institute - Network Functions Virtualization (ETSI-NFV), in several aspects of the application's life-cycle, striving to strike a balance between the objectives of all the involved stakeholders. Focusing on enabling the efficient deployment of emerging 5G services, optimized resource allocation, and enhanced overall application performance and Quality of Service (QoS), in this dissertation, the following challenges are tackled: (i) the infrastructure selection for network service deployment in the cloud continuum by taking into account the differences between infrastructure capabilities and the plethora of various individual requirements, (ii) the establishment of the Cross-Service Communication (CSC) and the Network Service Marketplace (NSM) to enable Virtual Network Function (VNF) sharing to further improve the resource utilization and enhance the QoS, (iii) distributed the virtual network embedding, to support modern IoT applications, which composed as interconnected VNFs, deployed distributed in the cloud continuum ecosystem, in the form of a hybrid Service Function Chain (hSFC), under strict resource capacity constraints, and (iv) design efficient and low-complexity algorithms to provide solutions to the corresponding problems, meeting the need for real-time decision making and re-optimization of such a dynamic environment. In this context, initially, this work proposes an Edge Cloud (EC) Selection framework for network service deployment, focusing on network slices. It introduces a distributed service discovery framework for shared-enabled VNFs using a cache-based protocol, in order to effectively identify the demanded VNFs while eliminating the communication overhead. The framework, also, utilizes a Multi-Criteria Decision Making (MCDM) algorithm, specifically, the Analytical Hierarchy Process (AHP), to identify the most suitable edge infrastructure for Network Slice deployment in a low execution time, considering the functional, performance, and cost requirements of both the infrastructure provider and the user. In addition, focusing on the QoS guarantees for IoT-based application deployment, where delay minimization plays a crucial role, a two-stage approach to provide distributed virtual network embedding (DVNE) solutions with minimized round-trip delay is introduced. Initially, a heuristic undertakes the association of VNFs of a Virtual Network Embedding (VNE) request; namely mapping, to a set of candidate ECs that can be hosted, while focusing on the minimization of the resource utilization within the EC infrastructure and increased co-location ratio between adjacent VNFs of the request. Given this initial mapping, an efficient path-based algorithm is proposed to construct the DVNE solution, in polynomial time. Based on graph theory, an augmented graph is constructed, on which several candidate paths for the DVNE are generated, using the k-shortest-paths algorithm. The embedding policy of the algorithm constructs optimal, or near-optimal solutions in polynomial time. Moreover, to ensure the compatibility and interoperability of some of the proposed mechanisms with existing industry standards and practices, the integration of the proposed mechanisms and algorithms into a MANO architecture is considered, to support Industry 4.0 application deployments. Extending the standard ETSI NFV reference architecture, a deployment and orchestration mechanism for enabling CSC in industrial environments is proposed. For evaluation purposes, a Warehouse Robotics use case, which requires the collaboration of mobile robotic agents in an automated storage and retrieval system, is used. Emphasizing the dynamic characteristics of CC, this thesis, finally, delves into the intricacies of dynamic resource allocation and autoscaling techniques. To this end, a multi-objective optimization mechanism for autoscaling has been developed. This mechanism leverages a multifaceted approach by integrating time series forecasting methods, specifically Auto-regressive Integrated Moving Average (ARIMA), to predict the incoming workload to applications' virtualized resources (containers) in the forthcoming time periods. Additionally, it incorporates the element of resource profiling per application, enabling the selection of the optimal application deployment, in terms of the number of identical replicas of containers. The primary objective of this mechanism is to strike a balance between the QoS guarantees for a given application, and, simultaneously, the mitigation of the administrative overhead for infrastructure providers by concurrently minimizing the overall energy consumption and resource utilization.