Fog–Cloud computing provides distributed computational resources close to IoT data sources, but efficient placement and resource utilization remain challenging because of heterogeneous resources and conflicting Quality of Service (QoS) requirements. In particular, simultaneously optimizing energy consumption, latency, execution cost, Service-Level Agreement (SLA) violations, and reliability leads to a constrained many-objective optimization problem in which conventional evolutionary approaches may suffer from reduced selection pressure, poor convergence, and inadequate diversity. To address these challenges, this paper proposes an Adaptive Constraint-Aware NSGA-III (I-NSGA-III) framework for many-objective QoS optimization in Fog–Cloud computing. The proposed approach extends the NSGA-III framework by integrating adaptive ϵ-constraint handling, feasibility-aware dominance, reference-direction biasing, and Angle Penalized Distance (APD)-based diversity preservation. The optimization simultaneously considers five QoS objectives—namely, energy consumption, latency, execution cost, SLA violations, and unreliability represented by (1−Reliability). Candidate solutions are represented as binary placement configurations and evaluated within a DCNSFog-based Fog–Cloud simulation environment implemented using the iFogSim2/CloudSim framework. The performance of the proposed approach is evaluated against NSGA-II and conventional NSGA-III using 30 independent runs, Hypervolume (HV), Inverted Generational Distance (IGD), statistical significance testing, and ablation analysis. The proposed I-NSGA-III achieves a mean HV of 0.8989±0.0283 and a mean IGD of 0.1677±0.0070, compared with 0.5252±0.0729 and 0.4099±0.0415, respectively, for conventional NSGA-III. The improvements are statistically significant, with a Wilcoxon test yielding p=2.8719×10−11 for both HV and IGD comparisons against the evaluated evolutionary baselines. The results demonstrate that adaptive constraint handling and diversity-oriented selection can substantially improve the quality and distribution of Pareto-optimal QoS trade-offs in many-objective Fog–Cloud optimization. The study also identifies limitations related to the current binary placement representation, the coupling between SLA and reliability measures, static simulation conditions, and the absence of large-scale real-world testbed validation, providing directions for future research.
Tiny Machine Learning (TinyML) enables resource-constrained devices to perform local inference in the Internet of Vehicular Things (IoVT), potentially reducing communication overhead, latency, and dependence on cloud services. This survey systematically reviews 57 primary studies covering vehicular applications, embedded hardware, lightweight learning models, optimization techniques, and deployment practices. The reviewed studies are organized using a seven-dimensional literature taxonomy covering deployment location, data modality, learning task, learning paradigm, optimization strategy, application domain, and system objective. Separately, their evaluation practices are critically examined through predictive performance, memory usage, inference latency, energy consumption, communication requirements, trustworthiness, and deployment evidence. The taxonomy characterizes existing research, whereas the benchmarking dimensions support comparison of reported experimental results and deployment limitations. The survey also examines computation–communication trade-offs, V2X interaction, collaborative learning, security, privacy, robustness, and functional safety. The findings reveal substantial heterogeneity in datasets, hardware platforms, evaluation protocols, and resource measurements, limiting direct cross-study comparisons. Moreover, experimentally demonstrated embedded feasibility does not necessarily establish reliability under realistic vehicular conditions or compliance with functional-safety requirements. The survey concludes by identifying research gaps in standardized benchmarking, hardware-aware optimization, adaptive learning, real-world validation, and trustworthy distributed vehicular intelligence.
Various competing wireless technologies claim superior long-range performance at low power budgets. However, it is difficult for system designers to select the appropriate technologies for their applications. Consequently, to guide such selections, this paper presents characteristic measurement data for three noted technologies: LoRaWAN, mioty, and Wi-Fi HaLow. The introduced measurement setups enable evaluation of the technologies with respect to energy consumption, sensitivity, and network performance in indoor environments. Based on the quantitative results, we discuss the individual strengths of the evaluated technologies. Supportive plots provide system designers with essential information to select the most appropriate wireless technology for their applications and avoid poor decision-making.
Decentralized Physical Infrastructure Networks (DePIN) need attestation that binds digital claims to physical devices. We present Lightweight Proof-of-Physical-Integrity (L-PoPI), a physical-to-Ethereum Virtual Machine (EVM) attestation framework combining SRAM-PUFs, BCH-based fuzzy extraction, FE-derived Groth16/Poseidon proofs, freshness-aware EVM verification, and a separate post-attestation behavioral branch. Physical validation used 30 cold-power ESP32 acquisitions and a pilot-scale six-board, 180-response campaign; BCH-C recovered 10/10 prospective and 60/60 held-out responses, and mean inter-device normalized Hamming distance was 49.22% in the homogeneous five-device cohort. The reconstructed key feeds a 1034-constraint Groth16 relation; host proving averaged 64.37 ms, verification averaged 10.11 ms, and stateful EVM attestations consumed 241,121 and 212,179 gas. A prospective ESP32-to-gateway experiment processed 904/906 PCAP windows; 798 controlled benign windows stayed below the fixed 0.512 risk threshold, with 441.59 ms mean decision latency. A scoped compositional argument prevents behavioral evidence from bypassing failed hard attestation, while [272, 128] code-offset helper exposure is bounded by 144 bits. Results support artifact-linked physical-to-EVM attestation with separate behavioral evidence. Boundary reminder: the six-device study is pilot-scale, and the online behavioral experiment used benign traffic only; neither supports manufacturing-scale population claims or online attack-recall estimates.
Digital twins are attracting growing interest in the industry. They enable the optimization of monitoring, control, and decision-making. However, many existing implementations rely on costly proprietary platforms. This limits their adoption by small and medium-sized enterprises. This paper presents a low-cost digital twin architecture. It is based on open-source platforms (Node-RED, MQTT) and affordable hardware (Raspberry Pi). This architecture was applied to the remote supervisory setpoint control of the water level in an industrial reservoir. Real-time synchronization between the physical system and its digital twin was evaluated through transport-latency measurements on clock-synchronized machines (NTP, inter-clock offset below 0.1 ms), with sequence-numbered messages paired end-to-end between the gateway and the twin. Across several measurement campaigns totaling over 33,000 messages, including a fully paired 5226-message validation session (zero loss, duplication, or reordering), the median transport latency was 44 ms, with 99.20% of messages delivered within 100 ms and occasional peaks of a few hundred milliseconds (maximum 923 ms), consistent with transport-layer retransmission stalls. Despite these peaks, performance remains acceptable for non-critical real-time applications, such as water-level supervisory control, since the resulting estimated level-deviation bound stays within the controller’s 0.2% hysteresis band for 99.0% of messages. These results indicate that a digital twin of a small industrial reservoir can be implemented in a viable and cost-effective manner at the gateway and communication layers. This approach makes the technology more accessible to small and medium-sized industries.
