An empirical study on edge-to-cloud continuum for smart applications: performance, design Patterns, and key factors

The rapid evolution of cloud-native technologies has facilitated seamless application deployment and execution across the entire edge-to-cloud continuum. This continuum offers a myriad of benefits, including reduced latency, optimized bandwidth utilization, enhanced data privacy, improved reliability, scalability, and flexibility. However, realizing a coherent edge-to-cloud continuum poses challenges especially in resource management, due to the heterogeneous and dynamic nature of computing resources such as resource scheduling and load balancing. This paper focuses on the Container-as-a-Service model enabling independent execution of functions/microservices anywhere on the continuum. We propose an architectural design for constructing a practical edge-to-cloud infrastructure and conduct comprehensive performance evaluations using a real edge-to-cloud testbed. Through an empirical study, we aim to identify key factors impacting application performance and resource management within the continuum, with a specific focus on AI-based IoT applications. Our experiments explore various design patterns including load balancing techniques, scheduling algorithms, invocation methods, gateway and data source location, and factors such as bandwidth and delay, providing practical insights for practitioners and researchers alike.