Simulated Fog-Cloud Framework Improves Latency and Security in IoT Task Offloading

In a simulated fog-cloud computing environment, a hybrid machine learning framework combining graph attention networks and reinforcement learning achieved better latency and security success rates than representative offloading approaches. The investigation, conducted by researchers Devendran Sudalai and Thanapal Pandi at Vellore Institute of Technology in India and published in Frontiers in Big Data, addressed the operational challenges of resource-constrained Internet of Things infrastructure. The proposed system integrates a Graph Attention Network with Proximal Policy Optimization to coordinate task scheduling across interconnected edge and cloud resources, demonstrating measurable improvements in latency, processing throughput, and energy efficiency compared to established baseline offloading configurations under controlled simulation conditions.

The research team established their experimental design around critical constraints inherent to modern Internet of Things deployments, where distributed client devices regularly generate data requiring both rapid execution and robust security protections. Traditional security and scheduling paradigms in these environments frequently rely on fixed static schedules and computationally demanding cryptographic protocols. According to the authors, conventional crypto-intensive models introduce substantial processing delays, consume excessive electrical energy, and fail to adapt dynamically to fluctuating network conditions across heterogeneous IoT-fog-cloud topologies.

To address these dynamic operational obstacles, the researchers modeled the multi-tier IoT-fog-cloud infrastructure as a reconfigurable graph rather than a static computational hierarchy. Within this dynamic graph representation, individual IoT devices, local fog processing nodes, and centralized cloud servers function as distinct network vertices connected by operational communication edges. The architecture deploys a specialized Graph Attention Network designed to learn dynamic topological relationships and structural dependencies across the shifting network components. By processing the dynamic interactions among edge units and processing nodes, the attention mechanism generates topology-aware node embeddings that capture real-time resource states and structural network connectivity patterns for downstream decision-making.

The generated graph embeddings are fed directly into an adaptive offloading and scheduling agent governed by Proximal Policy Optimization. This reinforcement learning model is tasked with determining optimal execution targets for individual heterogeneous workloads, selectively assigning each task to local fog nodes or centralized cloud servers. The offloading agent optimizes scheduling decisions by considering task urgency, resource availability, network characteristics, and task sensitivities alongside the proposed graph embeddings.

Data protection is integrated into the scheduling architecture through a lightweight verification protocol designed specifically for high-sensitivity transmission tasks. To minimize computational burdens on resource-limited hardware while preserving data integrity and authenticity, the researchers combined an integrated frameproof combinatorics-based verification mechanism with SHA-256 cryptographic hashing. Rather than executing intensive cryptographic operations uniformly across all incoming traffic, this verification scheme activates exclusively when high-sensitivity tasks are offloaded.

The complete framework was implemented in Python and rigorously evaluated using network traffic derived from the publicly accessible TON_IoT benchmark dataset. The authors converted network flows into heterogeneous IoT tasks and simulated the fog-cloud infrastructure and resource scheduling environment to test adaptive offloading. In experimental evaluations, the proposed framework achieved better latency, energy efficiency, throughput, Packet Delivery Ratio, resource utilization, and security success rate than representative offloading approaches.

Despite these favorable experimental metrics, the authors emphasized the inherent boundaries of their study, noting that all reported results were collected within a simulated environment utilizing converted dataset traces rather than physical hardware testbeds. Sudalai and Pandi concluded that while the hybrid framework provides an effective, adaptive methodology for latency-sensitive and security-critical Internet of Things applications, physical hardware-based validation and broader real-world network deployment studies remain essential future research objectives.

Subject of Research: Adaptive fog-cloud task offloading and secure data transmission in IoT networks

Article Title: A hybrid graph attention network and proximal policy optimization framework for adaptive fog-cloud offloading and frameproof-based secure IoT data transmission

Article References: Sudalai, D., & Pandi, T. (2026). A hybrid graph attention network and proximal policy optimization framework for adaptive fog-cloud offloading and frameproof-based secure IoT data transmission. Frontiers in Big Data, 9, 1954144. https://doi.org/10.3389/fdata.2026.1954144

Image Credits: AI Generated

DOI: 10.3389/fdata.2026.1954144

Keywords: fog computing, cloud computing, Graph Attention Network, Proximal Policy Optimization, Internet of Things, task offloading, TON_IoT dataset, SHA-256 hashing, frameproof verification, data transmission latency, energy consumption, packet delivery ratio, cybersecurity, adaptive scheduling, resource allocation, node embeddings, Welch's t-test