Smart home applications are growing in number and complexity, especially with advances in Artificial Intelligence. Most smart home platforms rely solely on cloud or local execution, leading to privacy concerns for the former and scalability and performance issues for the latter. Although hybrid edge–cloud solutions exist, most lack privacy considerations in offloading decision strategies or lack real-world validation. To address these challenges, we present AdaptHub, a smart hub resource and execution layer for smart home environments that adapts to increased computational loads by offloading tasks to cloud resources while prioritizing highly privacy-sensitive applications for local execution. AdaptHub employs a decision-making algorithm that models offloading and resource allocation as an optimization problem, which minimizes privacy leakage considering resource availability, application privacy sensitivity, user preferences, and application requirements. AdaptHub uses Linux control groups and multiprocessing to manage resource allocation and enable parallel execution of applications. Our evaluation shows that AdaptHub improves application completion times by at least 2X over a hybrid baseline, while executing significantly more privacy-sensitive tasks locally compared to a privacy-oblivious baseline with the same resource usage.
Add the full text or supplementary notes for the publication here using Markdown formatting.