The rapid growth of the Internet of Things (IoT) demands efficient system architectures and protocols to ensure consistent performance at scale. This paper explores the scalability of IoT systems across three key layers: sensing, network, and control. IoT scalability is the ability of a system to maintain consistent and reliable performance despite a continuous increase in connected devices. To evaluate scalability, we introduce the scalability indicator (SI), a metric designed to assess an IoT system’s scalability capability. Through extensive research and real-world deployments, we identify key challenges in data sensing, routing, and system control. Our study presents a model to understand these challenges and proposes strategies to optimize resource utilization, ensuring efficient data collection. The findings also emphasize the key influencing factors for the stable performance of large-scale IoT systems, providing valuable insights for how to design scalable systems that can meet the growing demand for interconnected devices.

Through deploying computing resources close to users, edge computing is regarded as a promising complement to cloud computing to provide low-latency computational services. Meanwhile, edge platforms also play the role of competitors of the cloud platforms in a non-cooperative game, which sets prices for computational resources to attract users with different real-time requirements. In this paper, we propose the edge pricing game under competition (EPGC) and investigate the truthful pricing mechanisms of the edge platform with the objective of maximizing its revenue under three different settings. When all user information is available, the optimal mechanism (OM) can be achieved based on a knapsack problem oracle. With partial information, where users’ resource demand is given but their preference information to the edge platform is private, we propose a random sampling mechanism (RSM) that achieves a constant approximation with probability approaching one. We also propose an efficient heuristic greedy mechanism, and we call it GM. Both mechanisms are truthful, GM is directly applicable, while RSM requires minor modifications (RSM⁺) for deployment in the prior-free setting where all user information is private. Finally, extensive simulations are conducted on the Google cluster dataset. The results validate our theoretical analysis that RSM⁺ works well in the market where edge resources are scarce, while GM performs better when the edge platform has a larger capacity constraint.