Design and Implementation of a Microservice-oriented Platform for Distributed Network Edge Applications

The rapid evolution of edge computing, has enabled the deployment of latency-sensitive applications at the network edge. This paradigm is essential for real-time processing scenarios, such as autonomous driving, unmanned aerial vehicles (UAVs), and computer vision applications, where deploying at the edge is not just advantageous but necessary to meet the strict latency and reliability requirements. These applications leverage distributed microservices (MSs) architectures, enabling scalability, fault tolerance, and efficient resource utilization while ensuring resilience and minimizing communication delays with the client. However, the distributed nature of MS architectures introduces challenges in efficient coordination and data sharing. Additionally, to fully exploit the benefits of edge computing, effective management and dynamic relocation of MSs are essential for maintaining service efficiency and responsiveness. Although MSs migration techniques have emerged to ensure service proximity to the end users, some challenges concerning migration latency and MSs state preservation still need to be addressed. To tackle these challenges, this thesis proposes a microservice-oriented platform to regulate collection, sharing and processing of critical distributed data, such as Network Performance Metrics (NPM) coming from the Radio Access Network (RAN), within edge computing environments. NPMs serve as quantitative indicators in AI-driven applications, assessing various aspects of network performance and providing valuable insights into the efficiency and reliability of data transmission. Initially, two reference platform architectures are presented, along with their implementation using widely adopted, off-the-shelf database solutions. To assess and compare these architectures, this work proposes PACE (Producer And Consumer Emulator), a highly configurable, scalable emulation framework that accurately emulates diverse interaction patterns and load conditions, developed and deployed within a cloud computing testbed. Through PACE an extensive evaluation of various NPM platform architectures is conducted under a range of realistic edge traffic scenarios, spanning from loosely coupled control loops to latency-sensitive and mission-critical applications. The results highlight key trade-offs in stability, availability, scalability, resource consumption, and energy efficiency, demonstrating how PACE facilitates the identification of optimal platform configurations based on the specific edge computing scenario and the required levels of reliability and data consistency. To validate the effectiveness of the proposed platform solution, a practical use case scenario is explored in the context of Multi-Object Tracking (MOT) within computer vision applications. In this domain, this work focuses on critical stateful applications, where maintaining context-based information is essential for ensuring accurate and reliable service performance. By managing tracking information as a distributed state, dynamically shared across the system, the platform enhances application consistency and service continuity, addressing key challenges for MSs relocation. The computer vision tasks were evaluated in different configurations, varying key parameters such as input video frame rate and object density. This analysis revealed trade-offs and valuable insights regarding accuracy, latency, and resource utilization, demonstrating the platform’s effectiveness in managing distributed state while ensuring optimal performance.