Abstract
The rapid evolution of modern networking has led to increased research efforts in the areas of edge computing and network slicing, fueled by the rise of the Internet of Things (IoT) and the advent of 5G (and beyond) networks. The growing popularity of mobile networks has resulted in a higher demand for both conventional cloud-based services and novel cloud services, such as mobile cloud games, remote control services for automobiles, and manufacturing process management services. Nevertheless, traditional cloud computing platforms that rely on large-scale datacenters have encountered difficulties in meeting the increasing computational requirements of emerging mobile and IoT-enabled applications. As a response to these difficulties, the notion of edge computing has emerged as a viable solution. Edge computing involves the positioning of computational resources in proximity to the edge of the network. This strategy provides numerous benefits, such as lower latency, faster response times ...
The rapid evolution of modern networking has led to increased research efforts in the areas of edge computing and network slicing, fueled by the rise of the Internet of Things (IoT) and the advent of 5G (and beyond) networks. The growing popularity of mobile networks has resulted in a higher demand for both conventional cloud-based services and novel cloud services, such as mobile cloud games, remote control services for automobiles, and manufacturing process management services. Nevertheless, traditional cloud computing platforms that rely on large-scale datacenters have encountered difficulties in meeting the increasing computational requirements of emerging mobile and IoT-enabled applications. As a response to these difficulties, the notion of edge computing has emerged as a viable solution. Edge computing involves the positioning of computational resources in proximity to the edge of the network. This strategy provides numerous benefits, such as lower latency, faster response times, and the capability to offload computationally demanding tasks from mobile devices. Nevertheless, the efficient utilization and orchestration of edge computing resources, particularly in situations with limited resources, pose serious challenges that require drastic solutions. This thesis undertakes a thorough investigation of the capabilities of edge computing and its potential to improve services. In this context, we identify numerous significant obstacles and propose innovative approaches and architectural frameworks in order to lower the barrier for service orchestration at the edge. Initially, we explore the practicality and benefits of offloading resource-intensive tasks to edge servers. We primarily focus on assessing the effect of this approach on latency and throughput, with the ultimate goal of enhancing the user experience. To this end, COSMOS comprises an advanced orchestration framework specifically designed to enable intelligent compute offloading for edge clouds. Our experimentation with COSMOS showcases the versatility and efficacy of the system in improving response times and customer satisfaction. Network slicing is evolving as an indispensable feature of network infrastructures, as it provides the means for the deployment of next-generation services with performance and reliability guarantees. Based on the trends in evolving service ecosystems, services situated on separate slices inside the same edge data center can enhance their functionality through synergies. In this context, we have identified a novel aspect of network slicing, which we term as Cross-Slice Communication (CSC). To foster CSC, we introduce a new CSC orchestration framework, namely optiMized Edge Slice OrchestratioN (MESON), which facilitates secure and efficient communication between different slices co-located on the same (edge) cloud infrastructure. This can empower service providers to capitalize the added value of cross-service interactions and, thereby, render their service offerings more appealing to their customers. Resource allocation at the edge comprises a crucial challenge, given the resource-constrained nature of edge computing infrastructures. In this respect, we examine the issue of resource allocation in environments with limited resources, such as an edge data center. We circumvent this difficulty at the intra-server level, which pertains to the final stage of resource allocation the assignment of VNFs to CPU cores). In this respect, we assess the impact of different CPU core allocation strategies on the performance of service chains and identify situations at which resources can be wasted. Our findings can be of great value to resource schedulers, enhancing their ability to allocate resources for VNFs or cloud-native applications. Furthermore, we advocate for the crucial need of low-latency communication in the scope of converged IoT-cloud environments, where digital twins or other forms of IoT middleware (Virtual Objects - VOs) reside in edge clouds to augment the operation of IoT devices. Our main aim here is to sustain low latency in the communication between IoT-VO pairs. To this end, we seek to reap the benefits of Time-Sensitive Networking (TSN), by presenting the design and implementation of a TSN platform that couples a TSN bridge, compliant with Time-Aware Shaper (TAS), with Centralized Network Control (CNC). As CNC mainly deals with the computation of TSN schedules, we present a constrained programming formulation for the TSN scheduling problem at hand. In addition, we discuss the interaction between the CNC and the TSN data plane, especially for the configuration of TSN schedules based on the intervals computed by our schedule engine at the CNC. Our evaluation results corroborate the efficacy of the TSN platform in terms of delay/jitter bounds and communication overhead.
show more
