A learning-based data and task placement mechanism for IoT applications in fog computing: a context-aware approach

In recent years, with the technological advancement and rapid growth in the Internet of Things (IoT), many physical devices and real-time applications have generated an enormous amount of data. Fog computing, as a decentralized computing infrastructure, facilitates the distribution of compute and storage resources nearer to IoT devices, where data is generated and processed. Despite the emergence of efficient fog computing models to address challenges like communication delays, network congestion, and energy consumption, the fog-cloud architecture is still in its early stages. Due to computationally intensive tasks and the huge amount of generated data on IoT infrastructure, task placement strategy has been considered an essential solution to low-memory IoT infrastructure and IoT-sensitive applications. This research proposes a novel context-aware framework for data and task placement within fog computing. The proposed framework is focused on separating data from tasks and employing context-based scheduling to enhance performance and efficiency. To address the limitations of static data placement, a high-performance reinforcement learning algorithm, federated learning, is utilized to dynamically determine the number of data replicas. Furthermore, an autonomous framework is designed to demonstrate the replica transferring between devices and fog nodes. Through extensive simulations, the proposed dynamic data replica framework has been shown cost-effectiveness and superiority by reducing total latency, average runtime, and enhancing system availability and reliability.