Network

The growing demand for virtual reality and immersive applications has increased interest in 360° video streaming. When viewing omnidirectional content through a head-mounted display, users observe only a limited portion of the content, i.e., the viewport, at any given time. Consequently, transmitting the complete panoramic frame at uniformly high quality is bandwidth-inefficient. This review presents a system-level analysis of viewport-adaptive three-degree-of-freedom (3DoF) 360° video streaming, focusing on the coupled roles of viewport prediction, tile-based multi-rate encoding and bitrate allocation, transport mechanisms, and edge-assisted processing. The reviewed literature is examined to identify the design dependencies and trade-offs among these components. Viewport-adaptive approaches seek to reduce the bandwidth allocated to regions outside the instantaneous viewport while preserving the quality of the visible region. The analysis shows that their effectiveness cannot be attributed to prediction accuracy alone: the resulting Quality of Experience (QoE) depends jointly on tile granularity, bitrate allocation, buffer occupancy, transport delay, and whether prioritized tiles arrive before their playback deadlines. Finer tiling can improve spatial selectivity but increases coding, signaling, and request overhead. Moreover, HTTP/2, HTTP/3/QUIC, RTP/RTSP, and WebRTC present different reliability, latency, congestion-control, and scalability trade-offs across buffered video-on-demand, low-latency live streaming, and interactive immersive applications. Based on this synthesis, the review formulates a unified closed-loop cross-layer framework that coordinates prediction, tiling, bitrate allocation, request timing, transport configuration, buffering, and edge processing under bandwidth, latency, and resource constraints. Full article

Fast Reroute (FRR) after link failures is essential for carrier-grade Software-Defined Networking (SDN), yet hybrid deployments remain dominated by slow legacy routing convergence. This paper presents two port-state-driven FRR mechanisms for homogeneous and hybrid SDN networks. First, Port-State-Aware Fast Reroute (PSA-FRR) uses OpenFlow port-status events to trigger proactive, rule-based protection in the data plane. Second, Port-State-Aware Neural Fast Reroute (PSAR-FRR) formulates hybrid FRR as a controller-local multi-class classification problem and predicts the backup egress port from a port-centric state representation, enabling microsecond-scale decision latency. We evaluate the methods on the Abilene wide-area network (WAN) topology using Mininet with Open vSwitch (OVS) and a Ryu controller (homogeneous case) and Graphical Network Simulator-3 (GNS3) with Cisco IOS routers (hybrid baseline). In homogeneous SDN emulation, PSA-FRR restores connectivity within 30–100 ms under the evaluated configurations. In the hybrid baseline, conventional routing protocols converge in 13.8–256.1 s (Enhanced Interior Gateway Routing Protocol (EIGRP), Intermediate System to Intermediate System (IS-IS), Open Shortest Path First (OSPF), Border Gateway Protocol (BGP), and Routing Information Protocol (RIP)), confirming that control-plane recovery cannot meet a 50 ms target. Using the collected dataset, PSAR-FRR reduces controller decision time from 6.753 $\mathsf{\mu }$s (PSA-FRR rule evaluation) to 0.214 $\mathsf{\mu }$s (deep neural network (DNN) inference), a 31.5× speedup. These results show that port-state awareness combined with learned, controller-local policies can substantially reduce the decision-to-action latency of FRR, providing a practical path toward low-latency failure recovery in SDN migration scenarios. Full article

Civilian communication systems often fail during armed conflicts, political unrest, and large-scale Internet disruptions—precisely when reliable communication is most needed. This paper presents HERMES, a resilient hybrid communication architecture that integrates HTTP/IP networking, Bluetooth Low Energy (BLE) mesh communication, and Delay-Tolerant Networking (DTN) within a unified adaptive routing framework. Unlike conventional approaches that treat alternative transports as backup solutions, HERMES dynamically selects the most efficient transport path based on current network conditions using a transport-aware forwarding policy whose cost function combines round-trip time, transport preference, and observed link risk. The architecture is built on distributed microservices that support topology discovery, shortest-path routing, and fault-tolerant message delivery. Reliability is enhanced through acknowledgments, bounded retransmissions, duplicate suppression, and graceful degradation mechanisms, while end-to-end authenticated encryption (Noise XX with a Double Ratchet) ensures secure communication across transport changes. A prototype implementation developed in C# on .NET 9 was evaluated on a five-node testbed, and a custom Network Simulator 3 (NS-3) module was used to extend the evaluation to networks of up to 500 nodes, under multiple failure scenarios, including node crashes, network partitioning, and complete Internet outages. Experimental results show that HERMES maintains perfect or near-perfect delivery in static topologies, including during a complete Internet blackout that disables IP-only messaging. Compared with the published Delay-Tolerant Networking protocols Epidemic and PRoPHET at one hundred nodes, HERMES exceeds their delivery ratio in static and failure scenarios and remains within 0.06 of them under pedestrian mobility during blackout, while transmitting roughly 35× fewer bytes– and about 21× fewer even relative to the more bandwidth-efficient MaxProp baseline. Under coordinated drop attacks by adversarial relays, HERMES degrades gracefully where flooding-based baselines collapse. This approach demonstrates that resilient civilian communication can be effectively achieved through metric-driven adaptive multi-transport routing, making it suitable for disaster recovery, contested environments, and connectivity-limited regions. Full article

Multi-controller deployments in Software-Defined Networking require choosing both the number of controllers and their placement on the topology. Clustering-based methods, particularly k-means, are widely used for this problem, but the topology representation that the clustering operates on is rarely chosen explicitly. We treat the representation step as a design axis: the clustering algorithm is held fixed at k-means with k-means++ initialisation, and the input is varied across three classical, training-free embeddings of the propagation-delay distance matrix: metric multidimensional scaling (MDS), Isomap, and Laplacian Eigenmaps (Spectral). The evaluation covers thirteen Internet Topology Zoo backbones grouped into three scale tiers under an effective-N definition, with the controller count K varied from 2 to 10, and reports node-to-controller latency (N2C), controller-to-controller latency (C2C), and load imbalance jointly rather than singly, with paired significance tests over ten distinct seeds. The representation choice is consequential: 60% of pairwise embedding comparisons are statistically separated (Holm-corrected Wilcoxon, $\alpha =0.05$), and the median best-versus-worst gap per configuration is 23–25% on the two latency metrics and 71% on load imbalance. Metric MDS achieves the lowest N2C in most regimes; Spectral achieves the lowest C2C on the medium and large tiers at mid-to-high K; Isomap trades single-metric wins for worst-case robustness and is non-dominated in 79% of the 63 large-tier configurations. Across all 117 configurations, each embedding is empirically non-dominated (within the embeddings compared and on this benchmark) in 67–85% of configurations, and 16% admit a single statistically separated best choice. Two baselines contextualise these results: k-means on the raw latency matrix leads in under 8% of configurations, and a Node2Vec baseline is competitive on C2C and load balance but trails the classical methods on N2C at one to two orders of magnitude higher embedding cost. Best-embedding identities transfer from k-means to a k-medoids variant in 70% of configurations. Capacity-, energy-, and reliability-aware extensions remain open. Full article

Software-defined Networking (SDN) is a promising networking architecture that separates the control and data planes to allow flexible network management. However, the SDN architecture makes networks vulnerable to various security threats, such as Distributed Denial-of-Service (DDoS) attacks. A DDoS attack is one of the most common SDN threats, aiming to exhaust a network’s computational and bandwidth resources. Self-similarity is a statistical property of time series in which data patterns repeat at different time scales. Several studies have shown that network traffic exhibits increased self-similarity during DDoS attacks, making it a promising tool for DDoS detection. Despite the effectiveness of statistical methods for detecting DDoS, some methods, such as self-similarity, are discarded due to their high computational cost, leading to detection delays. This paper proposes a lightweight Rescaled Range ($R/S$)-based scheme for effective real-time DDoS attack detection in SDN. The scheme employs the Welford online algorithm to compute statistical parameters of the $R/S$ scheme. Experimental results demonstrate that the proposed scheme efficiently captures changes in self-similarity and detects TCP/UDP DDoS attacks in real time. Moreover, it achieves high detection performance compared to other $R/S$ methods, with a False Positive Rate (FPR) below 0.5% and an average computation time of 0.047 ms. Full article

This paper develops a reliability-oriented resource-allocation framework for a single-cell uplink sparse code multiple access (SCMA) system with fixed low-projection codebook (LPCB) constellation components. Link-level SCMA–message-passing-algorithm (MPA) samples calibrate a compact bit-error-rate (BER) surrogate, enabling repeated candidate evaluation without embedding MPA decoding in the search loop. The proposed two-stage method combines a subcarrier allocation whale optimization algorithm (SAWOA), equipped with stochastic binary mapping and exact degree-feasibility repair, with an exact Karush–Kuhn–Tucker (KKT) active-set power allocator. For the $J=6$, $K=4$ setting, exact enumeration over all 210 main conditions shows that the SAWOA attains the enumerated equal-power ${\mathcal{P}}_1$ oracle objective within relative tolerance ${10}^{-10}$ in every case. Across 30 paired channel/optimizer blocks, each aggregating the seven power points, the SAWOA reduces the initial-gap-normalized convergence area under the curve by 47.5% relative to the Standard Binary WOA (Holm-adjusted $p=1.19\times {10}^{-5}$); its oracle-hit rate by iteration 20 is 93.3% versus 75.7%, while no significant AUC difference is detected relative to particle swarm optimization. After 100 iterations, the three population methods approach the same oracle plateau, whereas random is significantly worse at the midpoint ($p=7.45\times {10}^{-9}$). Exact KKT refinement improves every recorded SAWOA solution and reduces the equal-power surrogate by 3.64–56.27% on average across the sweep, with a maximum relative KKT residual of $1.01\times {10}^{-16}$. A single-point direct Rayleigh SCMA–MPA check confirms the executable transfer of the selected allocation and returns the same decoded BER for the SAWOA and Standard Binary WOA; it remains a bounded transfer sanity check. The results demonstrate finite-budget stage-1 search efficiency, reliable feasible support recovery, and effective exact power refinement for the investigated quasi-static configuration. Full article

Data Centers that implement Software-Defined Networks (SDN) are not required to employ the Address Resolution Protocol (ARP), but network hosts do. Therefore, there is a need to support this protocol without modifying the intrinsic functionality of the SDN controller. In this work, four strategies for handling ARP are evaluated using an SDN and OpenFlow rules. The strategies include disabling ARP at the host level, using static MAC addresses, introducing a fake gateway, and generating ARP replies using OpenFlow flows. To our knowledge, nobody has tested and compared the main characteristics and advantages offered by these four strategies. Experimental evaluation was conducted on a real SDN network and complemented with similar experiments using Mininet. Performance was assessed using metrics such as ping response time, address resolution response time, jitter, and packet loss ratio. The results show that OpenFlow-based ARP replies provide a good balance in terms of scalability, performance, and configuration effort. This strategy achieved the lowest average ping response time (0.641 ms) and ARP response time (0.6188 ms), while avoiding the manual configuration requirements of static approaches. Full article

The convergence of optical burst switching (OBS) and elastic optical networks (EON) offers a promising pathway for 6G IoT backhaul. However, existing solutions treat OBS and EON separately and rely on heuristic resource allocation that fails to meet stringent QoS demands. This paper proposes a fully AI-driven converged OBS/EON architecture integrating a hybrid switching fabric, a multi-agent deep reinforcement learning (DRL) orchestrator, and a federated learning (FL) plane for autonomous, QoS-aware resource provisioning. The control plane implements multi-agent Proximal Policy Optimization (PPO) for joint burst scheduling, routing, modulation selection, and spectrum allocation. The orchestration plane employs q-fair FL for privacy-preserving cross-domain traffic prediction. Mathematical formulations of the optimization problem with spectrum, GSNR, and delay constraints are provided, along with pseudo-algorithms. Simulations over a 14-node NSFNET topology demonstrate a 78% reduction in blocking probability, a 42% improvement in spectral efficiency, sub-millisecond URLLC delays, and a Jain’s fairness index of 0.92, while preserving data privacy. The framework builds upon SDN principles for seamless integration with optical transport infrastructures. Full article

Vehicular Ad Hoc Networks (VANETs) are essential components of Intelligent Transportation Systems (ITS), allowing communication exchanges between vehicles and road infrastructure elements. These networks face challenges from vehicular mobility, including frequent topology changes, link instability, and variable wireless channel quality. This paper presents an extensive evaluation of Software-Defined Networking (SDN) integrated with two traffic engineering technologies, Multi-Protocol Label Switching (MPLS) and Segment Routing (SR), applied to the AODV and OLSR routing protocols. Nine incremental configurations are evaluated for each protocol, ranging from the default protocol through MPLS-enhanced forwarding, SDN-based centralized optimization, combined SDN-MPLS and SDN-SR integration, to advanced configurations using distance-based IS-IS weighted topology metrics with both Fixed and Adaptive metric computation approaches. Two distinct SDN topology construction methods are compared: a Protocol-based approach that uses routing table entries with equal hop-count metrics, and a distance-based approach using IS-IS weighted metrics. The simulation uses a realistic urban topology with 50 vehicles and 5 RSUs, evaluated across several traffic patterns, representing different application types. Results demonstrate that SR with distance-based IS-IS metrics achieves the highest Packet Delivery Ratio (PDR) and lowest delay by leveraging RSU infrastructure as reliable forwarding relays. Moreover, the proposed SDN-SR framework reduces routing overhead and control-plane signaling, improving network resource utilization and thereby indicating its potential to enhance the energy efficiency of vehicular communication infrastructures. Full article

Smart and sustainable mobility increasingly relies on distributed sensing, low-power communication technologies, and cloud-based ICT platforms. This article presents a comprehensive scientific analysis of the technological foundations of sensitized mobility, reviewing the state of the art in embedded sensing, distributed systems, and communication paradigms for future mobility challenges. The research project “MOST” and in particular its subgroup “Spoke 5” falls within this framework of sustainable and sensorized mobility, with numerous activities in data collection, analysis, and field experimentation. In order to allow data collection, retention and analysis, one of the challenges that we must address is the definition of an adequate ICT architecture. The core contribution of this work is the presentation of the MOST ICT architecture, designed as a containerized, scalable, and resilient infrastructure capable of integrating heterogeneous data coming from field-deployed systems. In addition, it discusses the primary research challenges encountered in the definition and development of the presented architecture by examining two representative case studies within the MOST-Spoke 5 research project: renewable energy charging stations for light electric vehicles and cyclists monitoring systems. Full article

Blockchain technology has introduced a rapidly expanding range of consensus mechanisms, each designed to satisfy different operational requirements related to security, scalability, decentralization, transaction throughput, and energy efficiency. Selecting an appropriate consensus mechanism has consequently become a complex multi-criteria decision problem, particularly for developers and organizations without extensive expertise in distributed systems and blockchain architectures. Existing tools primarily address protocol benchmarking and static documentation, leaving the decision-support dimension largely unaddressed. This paper presents a web-based expert system designed to support the selection of blockchain consensus mechanisms according to specific user-defined operational requirements. The proposed solution was implemented using the MERN technology stack, consisting of MongoDB, Express.js, React, and Node.js, enabling a modular and scalable architecture suitable for future expansion and maintenance. The recommendation process is based on a two-phase filtering and scoring algorithm. In the first phase, mechanisms incompatible with mandatory user-defined constraints, including network type and key resource type, are systematically eliminated. In the second phase, the remaining mechanisms are ranked using attribute matching across criteria encompassing energy efficiency, scalability, security, decentralization, and transaction speed. The system returns the three most suitable consensus mechanisms for the given operational scenario together with their key characteristics. In addition to recommendation functionality, the application supports user authentication, recommendation history management, and administrative maintenance of the consensus mechanism database, which currently contains 38 distinct blockchain consensus protocols. Experimental evaluation through twelve representative usage scenarios demonstrated that the system consistently produces contextually relevant recommendations aligned with user-specified requirements. A comparative analysis with existing tools confirms that the proposed system occupies a distinct decision-support role currently absent from the available tooling landscape. A comparative analysis against four representative existing tools indicates that the proposed system combines a set of decision-support capabilities not jointly offered by any one of them. The presented approach contributes a transparent and extensible decision-support framework intended to simplify architectural planning and management of blockchain-based distributed systems. Full article

IPv4 fragmentation attacks, including tiny fragment injection, teardrop offset manipulation, and fragment flooding exploit the RFC 791 reassembly process to evade firewalls and network intrusion detection systems (NIDSs). Detection is challenging because widely used flow-level datasets (UNSW-NB15, CIC-IDS2017), lack the packet-level fragment header information required for direct RFC 791 validation. This study investigates whether fragmentation-related traffic can be identified using only flow-level statistical features. The proposed framework introduces four contributions: (1) a direction-corrected proxy labeling scheme, where flows are labeled as fragmented when ${CV}_L$ < 0.70, validated on a controlled 3000-flow Scapy dataset with $F_1$ = 0.873 and ROC-AUC = 0.895 against verified packet-level ground truth; (2) a dual-model Random Forest architecture with feature separation to prevent circular self-prediction; (3) RFC 791-inspired statistical heuristics applied as a post-inference filter; and (4) a six-configuration ablation study with a reproducible protocol. The study distinguishes fragmentation-like statistical signatures from confirmed packet-level fragmentation. The benchmark model $M_{frag}$ achieves $F_1$ ≈ 0.998 on synthetic data, while external PCAP validation yields $F_1$ = 0.873. Unlike most flow-level NIDS research, the framework is validated against both controlled Scapy-generated traffic and the MAWI real-world backbone trace, establishing a practical performance bound ($F_1$ = 0.82 for the full hybrid framework). The attack classification model $M_{attack}$ achieves $F_1$ = 0.974 on dataset-provided labels. Results show that flow-level analysis can provide useful indicators of fragmentation-related activity when packet-level evidence is unavailable, while highlighting the limitations of statistical detection. Full article

As climate and societal challenges intensify, achieving sustainable development requires a holistic approach that balances economic growth with environmental preservation and social equity. The United Nations’ 2030 Agenda outlines 17 Sustainable Development Goals with corresponding global indicators. However, adapting these high-level goals to the specific context of telecommunications networks, which are more and more governed by AI-based components, presents major challenges. This paper proposes an extensive framework for assessing the adequacy of Key Value Indicators, used to evaluate the Key Values brought by network services and applications. The framework encompasses the phases of indicator elicitation, analysis, technical realization, and assessment. Applying this framework ensures that Key Value Indicators remain relevant, actionable, and pertinent to the objectives throughout the system’s evolution. Case studies from leading projects in the telecommunications sector illustrate how the proposed methodology effectively identifies deficiencies and helps refine the framework. Full article

The recent advances in Vehicular Ad Hoc Networks (VANETs) can have a tremendous positive impact on vehicle safety and traffic flow. In VANETs, vehicles communicate wirelessly with each other and with roadside infrastructure nodes to improve awareness of neighboring vehicles and traffic conditions. However, such communication also increases the potential for various safety and security challenges, such as the threat of false reporting attacks. In these attacks, malicious or compromised nodes inject alert notifications that report fictitious traffic incidents that may trigger unnecessary evasive actions and increase the risk of collisions. This research addresses false alert attacks in VANETs by developing a machine learning-based detection framework that leverages innovative feature engineering and model assessment strategies. The proposed framework designs new features that capture vehicle kinematics to improve the detection of malicious alerts. The performance of multiple ML models is then analyzed in terms of both detection effectiveness and computational requirements to determine their suitability for different deployment scenarios. Our simulation results demonstrate that the proposed framework can achieve significant improvements compared to existing techniques for false alert detection. In addition, the study highlights the critical role of feature engineering in improving detection performance. Full article

Modern communication networks require routing protocols that can adapt to dynamic traffic conditions while accounting for topology-based structural risk. Conventional open shortest path first (OSPF) relies on static or linear link cost metrics, which are often inadequate for capturing the nonlinear behavior of network dynamics and structural risk. This paper proposes sigmoid-enhanced OSPF (SE-OSPF), which integrates topology-based structural risk into the OSPF routing metric through a nonlinear sigmoid function. The proposed framework employs two configurable sigmoid parameters, the midpoint ($x_0$) and the steepness (k), to provide smooth cost transitions and adaptive routing decisions under varying network conditions. Simulation results on a Barabási–Albert scale-free topology demonstrate that SE-OSPF reduces the average end-to-end delay by 19.7% and packet jitter by 8.6% compared with Standard OSPF. In addition, SE-OSPF increases the average number of successfully delivered packets by up to 16.6% compared with Linear-OSPF while reducing maximum link utilization (MLU), indicating more balanced traffic distribution, improved load balancing, and reduced congestion. These results demonstrate that the proposed sigmoid-based routing metric effectively balances routing efficiency, packet delivery reliability, and network load distribution, establishing SE-OSPF as an effective framework for topology-based structural risk-aware adaptive routing in modern communication networks. Full article

The post-quantum signature scheme SQIsign achieves extremely compact public keys and signatures, making it attractive for bandwidth-constrained environments. However, its signing efficiency is limited by the high random failure rate of the ideal-to-isogeny mapping procedure and the substantial cost of each retry. Existing optimizations mainly reduce the number or cost of isogeny computations, while overlooking how to schedule commitment retries when multiple candidate ideals are available. We formulate commitment-stage scheduling as a lightweight learning-to-rank problem and provide an instrumented scheduling framework for SQIsign signing only. The pipeline uses two features, trains a weighted logistic regression scorer offline by maximum likelihood with class weighting, and deploys the same scorer online in Rank-ML mode. Live instrumentation on Apple M2 (n = 20,000 candidate attempts at NIST-I) quantifies the commitment bottleneck (86.4% failure; 7.36 mean attempts per session) and shows constant features at a fixed commitment degree (live AUC $=0.50$). Synthetic training supports the scorer when feature variance is present (test AUC $\approx 0.634$). A remeasured four-way ablation with Batch-only control ($n=100$, seed 42) separates batch overhead from learned ordering: Rank-ML is indistinguishable from Batch-only at deployment, while Baseline remains fastest for its wall-clock signing time at batch size 10. These results clarify when lightweight ML scheduling applies in SQIsign and provide a reproducible evaluation template separating live, synthetic, remeasured, and proxy evidence. Full article

Federated learning (FL) has emerged as a promising paradigm for privacy-preserving edge intelligence because it enables geographically distributed devices to collaboratively train a shared model without transferring raw data to a central cloud. This capability is particularly valuable for 5G and emerging 6G networks, where edge-native services are required to satisfy stringent latency, bandwidth, and privacy constraints while operating on highly heterogeneous devices and time-varying wireless channels. In practice, however, synchronous FL is often constrained by straggling clients with limited computation capability or unfavorable communication conditions, which increases round latency and reduces overall resource efficiency. To address this challenge, this study develops a rigorously structured framework for dynamic client selection and radio resource allocation in heterogeneous wireless edge environments. Each FL round is formulated as a latency-aware scheduling problem that jointly captures local computation time, uplink transmission time, minimum participation constraints, and resource block assignment. On this basis, we propose a Dynamic Client Selection and Resource Allocation (DCS-RA) method that integrates computation-aware, channel-aware, and fairness-aware scoring with greedy resource block allocation guided by marginal completion time reduction. The study further provides a clear methodological structure, workflow visualization, literature-grounded justification, dataset documentation, and uncertainty-aware result reporting. Under the reported simulation setting with 100 clients and 20 resource blocks, DCS-RA reduces the average round completion time from 1.92 s to 1.55 s on MNIST and from 2.02 s to 1.57 s on CIFAR-10, corresponding to improvements of 19.39% and 22.47%, respectively. Standard deviation reductions of 70.59% and 80.77% further indicate improved round-to-round stability and more reliable training behavior. These results support the central conclusion that lightweight joint scheduling can materially improve wall-clock FL efficiency in heterogeneous 5G/6G edge networks. Full article

The internet of things (IoT) has rapidly evolved into a ubiquitous communication paradigm for enabling the deployment of autonomous wireless networks across diverse application domains. However, the limited energy storage capacity and computational resources of IoT devices (IoTDs) pose a serious concern to their long-term sustainability and the expected quality of service delivery. Moreover, in the foreseeable era of the internet of everything, centralised network resource management is likely to constrain network scalability. To tackle these challenges in the current and next-generation communication networks, the adoption of adaptive and lightweight computational frameworks coupled with energy-efficient transmission strategies is essential. To demonstrate this, we exploit the concept of cooperative communication and radio frequency-based energy-harvesting to improve the network throughput while maintaining power supply to the IoTDs. Furthermore, to intelligently and autonomously perform resource allocation, we employ the reinforcement learning frameworks, particularly state–action–reward–state–action (SARSA) and Q-learning. Based on key performance evaluation metrics, we compare our findings with the baseline methods, including the equal, random, and greedy power level selection schemes, with SARSA exhibiting the most favourable performance trade-offs. Full article

IoT networks now handle traffic from billions of devices, and edge nodes are under constant pressure to classify that traffic and dispatch tasks within tight latency deadlines. Most existing systems treat classification and scheduling as two separate steps that run one after the other. This sequence adds unnecessary delay and breaks the feedback between the two tasks: the scheduler never sees the traffic type, and the classifier never sees the queue state. We propose ParallelEdge-AI, a system built around a shared flow encoder that feeds two task-specific heads in parallel, one for multi-class traffic classification and one for task-urgency scoring. Both heads are trained end-to-end using a joint loss that combines cross-entropy and pairwise ranking. A load-balance controller then reads the urgency scores alongside live queue lengths to decide, every 200 ms, whether a task stays local or moves to a less-loaded edge node. No global synchronisation is needed. We test the system on three real IoT datasets: RT-IoT2022, N-BaIoT, and CICIoT2023. ParallelEdge-AI reaches 97.63% accuracy and an F1-score of 97.34%, which is 3.16 percentage points above the best baseline. Inference latency is 19.62 ms per batch, the deadline-miss rate is 2.34%, and the load-imbalance index is 0.083, all three are the best results in our comparison. These numbers show that running classification and scheduling together on a shared representation is both faster and more accurate than treating them as separate problems. Full article