A Timeliness-based Data Computing/Gathering Offloading Model for Internet of Things Devices

Abstract

Numerous Internet of Things (IoT) applications demand accurate and timely information to enable effective actuation. Nonetheless, in computation-intensive status update systems or data gathering systems, the capabilities of IoT devices may fall short in computing/acquiring data with high accuracy, thereby necessitating multiple trials. Offloading data computation or acquisition tasks to robust units mitigates this inaccuracy. However, these units can be positioned far from the user, and latency becomes an issue for offloading due to some factors, such as propagation delays and resource sharing with other possible tasks. In this thesis, we study how to keep the information accurate and fresh, introducing a cost function representing the staleness of the recently obtained data that is accurate enough for actuation. We propose a timeliness-based model striking a balance between employing local and remote resources. We consider two settings to treat the problem, namely blind and informed decision settings. In the blind setting, we utilize a stochastic decision-making strategy where the user makes the offloading decisions without knowledge of the current value of the cost function. We conduct a steady-state analysis and solve the problem through convex optimization. We also extend this setting to multi-user scenarios sharing the same remote resources. In the informed decision setting, we address the optimization problem as a Markov Decision Problem (MDP), in which the user leverages the current cost function value. We resolve the issue using finite horizon Dynamic Program- ming (DP). We state that while the informed decision setting yields superior results, it also has several drawbacks, such as consuming extensive memory and flexibility.