Modeling, Simulating, and Evaluating Complex End-to-End Edge Intelligence Systems

End-to-end edge intelligence systems (EIS) at the Internet’s edge have been receiving ever-increasing attention in both academia and industry, driven by a myriad of factors, including the growing number and diversity of edge applications and their strict requirements for latency, reliability, and efficiency; the increasing capability of edge devices and infrastructure; and the push toward more privacy-preserving and less centrally dependent systems. The individual components of emerging end-to-end edge intelligence systems—the applications themselves, the distributed learning and inference strategies that drive them, and the complex and heterogeneous network infrastructures they run on—have been studied, implemented, and evaluated, providing us with each component’s theoretical and practical expectations and constraints. However, an outstanding challenge in the real-world deployment of edge intelligence systems is developing a more comprehensive and holistic understanding of how they will perform, i.e., modeling and evaluating the interactions between their components and the overall integrated system. While testing systems deployed on real hardware and on real networks is important, it is not always a practical, affordable, or accessible approach. Given the tools and platforms available to us today, we can create more comprehensive, end-to-end models and simulations that give us a better understanding of the performance of the systems we design under a wide range of scenarios in a reproducible fashion, which is a necessary initial step before such systems can be deployed and tested in the real world. In this chapter, we will: (a) define and describe the components relevant to modeling, simulating, and evaluating edge intelligence systems; (b) introduce some of the fundamental metrics used in validating and evaluating edge intelligence systems; (c) introduce tools and techniques for modeling, simulating, and evaluating the performance of edge intelligence system components; and (d) explore end-to-end and integrated solutions for holistic modeling, evaluation, and simulation of edge intelligence systems.