TNSM Searchable Index

Author(s)	Title	Year	Publication	Keywords
Raffaele Carillo, Francesco Cerasuolo, Giampaolo Bovenzi, Domenico Ciuonzo, Antonio Pescapé	A Federated and Incremental Network Intrusion Detection System for IoT Emerging Threats	2026	Early Access	Training Incremental learning Adaptation models	Ensuring network security is increasingly challenging, especially in the Internet of Things (IoT) domain, where threats are diverse, rapidly evolving, and often device-specific. Hence, Network Intrusion Detection Systems (NIDSs) require (i) being trained on network traffic gathered in different collection points to cover the attack traffic heterogeneity, (ii) continuously learning emerging threats (viz., 0-day attacks), and (iii) be able to take attack countermeasures as soon as possible. In this work, we aim to improve Artificial Intelligence (AI)-based NIDS design & maintenance by integrating Federated Learning (FL) and Class Incremental Learning (CIL). Specifically, we devise a Federated Class Incremental Learning (FCIL) framework–suited for early-detection settings—that supports decentralized and continual model updates, investigating the non-trivial intersection of FL algorithms with state-of-the-art CIL techniques to enable scalable, privacy-preserving training in highly non-IID environments. We evaluate FCIL on three IoT datasets across different client scenarios to assess its ability to learn new threats and retain prior knowledge. The experiments assess potential key challenges in generalization and few-sample training, and compare NIDS performance to monolithic and centralized baselines.	10.1109/TNSM.2026.3675031
Yifei Xie, Zhi Lin, Kefeng Guo, Ruiqian Ma, Hussam Al Hamadi, Fatima Asiri, Ahlam Almusharraf	Lightweight Learning for Symbiotic Secure and Efficient ISAC in RIS-assisted Intelligent Transportation Networks	2026	Early Access		Achieving real-time processing in integrated sensing and communication (ISAC) systems presents significant challenges due to the high computational burden of conventional optimization methods, particularly within intelligent transportation networks (ITN). This paper addresses these challenges by proposing lightweight supervised and unsupervised deep learning (DL) algorithms, respectively for quasi-static and dynamic environments, aiming to improve the secrecy energy efficiency (SEE) of ITN under the constraints of the Cram´er-Rao bound (CRB) for direction-of-arrival (DOA) estimation and the transmission rate of each user. By jointly optimizing power allocation and reconfigurable intelligent surface (RIS) phase shifts, the framework ensures robust physical layer security (PLS) alongside communication efficiency, aligning with defense-in-depth strategies for securing next-generation ITN. For quasi-static environments, a supervised deep neural network (DNN) algorithm leverages offline codebook-generated labels to achieve near-optimal channel state information (CSI) mapping, explicitly minimizing signal leakage to eavesdroppers. In dynamic scenarios, an unsupervised channel attention mechanism-based residual network (CAM-ResNet) eliminates labeling overhead through direct physics-informed SEE optimization with adaptive constraint enforcement, enabling real-time adaptation to rapidly varying channels and evolving security threats. Simulation results demonstrate that both algorithms achieve comparable SEE performance with the zero-forcing (ZF) method, while significantly reducing computational complexity, with the CAM-ResNet demonstrating superior resilience to dynamic security threats. This work contributes to advancing secure and efficient ISAC solutions, reinforcing multi-layered defense mechanisms critical for future ITN.	10.1109/TNSM.2026.3679370
Xiaolong Wang, Haipeng Yao, Lin Zhu, Wenji He, Wei Zhang, Mohsen Guizani	Joint Optimization of Routing and Scheduling in Cross-Domain Deterministic Networks	2026	Early Access		Industrial Internet applications require networks to guarantee deterministic end-to-end latency and zero packet loss at both the data link and network layers. Traditional best-effort communication models in consumer networks are insufficient to meet these stringent demands. To meet these stringent demands, the IEEE 802.1 standards introduce Time-Sensitive Networking (TSN) at the data link layer, while the IETF proposes Deterministic Networking (DetNet) for the network layer. However, enabling seamless cross-domain communication between TSN and DetNet remains a significant challenge. This paper proposes a unified cross-domain network architecture and a time-slot alignment strategy that compensates for synchronization errors between the TSN and DetNet layers. We further develop a Joint Routing and Scheduling algorithm for Deterministic Cross-Domain Transmission (JRS-DCT), which simultaneously addresses routing and scheduling under cross-domain constraints. The algorithm leverages Cycle-Specified Queuing and Forwarding (CSQF) in DetNet and Cycle Queuing and Forwarding (CQF) in TSN to ensure bounded latency and deterministic transmission. Extensive simulations demonstrate that the proposed JRS-DCT algorithm significantly improves the scheduling success rate and effectively reduces network resource utilization compared to two baseline algorithms. These results validate the effectiveness and robustness of the proposed framework in supporting time-sensitive communication across heterogeneous network environments.	10.1109/TNSM.2026.3679810
Julien Ali El Amine, Nour El Houda Nouar, Olivier Brun	Online Network Slice Deployment across Multiple Domains under Trust Constraints	2026	Early Access		Network slicing across multiple administrative domains raises two coupled challenges: enforcing slice-specific trust constraints while enabling fast online admission and placement decisions. This paper considers a multi-domain infrastructure where each slice request specifies a VNF chain, resource demands, and a set of (un)trusted operators, and formulates the problem as a Node–Link (NL) integer program to obtain an optimal benchmark, before proposing a Path–Link (PL) formulation that pre-generates trust and order-compliant candidate paths to enable real-time operation. To mitigate congestion, resource prices are made dynamic using a Kleinrock congestion function, which inflates marginal costs as utilization approaches capacity, steering traffic away from hotspots. Extensive simulations across different congestion levels and slice types show that: (i) PL closely tracks NL with negligible gaps at low load and moderate gaps otherwise, (ii) dynamic pricing significantly reduces blocking under scarce resources, and (iii) PL reduces computation time by about 3×–6× compared to NL, remaining within a few seconds even at high load. These results demonstrate that the proposed PL and dynamic pricing framework achieves near-optimal performance with practical runtime for online multi-domain slicing under trust constraints.	10.1109/TNSM.2026.3679794
Amin Mohajer, Abbas Mirzaei, Mostafa Darabi, Xavier Fernando	Joint SLA-Aware Task Offloading and Adaptive Service Orchestration with Graph-Attentive Multi-Agent Reinforcement Learning	2026	Early Access	Quality of service Resource management Observability	Coordinated service offloading is essential to meet Quality-of-Service (QoS) targets under non-stationary edge traffic. Yet conventional schedulers lack dynamic prioritization, causing deadline violations for delay-sensitive, lower-priority flows. We present PRONTO, a multi-agent framework with centralized training and decentralized execution (CTDE) that jointly optimizes SLA-aware offloading and adaptive service orchestration. PRONTO builds on Twin Delayed Deep Deterministic Policy Gradient (TD3) and incorporates spatiotemporal, topology-aware graph attention with top-K masking and temperature scaling to encode neighborhood influence at linear coordination cost. Gated Recurrent Units (GRUs) filter temporal features, while a hybrid reward couples task urgency, SLA satisfaction, and utilization costs. A priority-aware slicing policy divides bandwidth and compute between latency-critical and throughput-oriented flows. To improve robustness, we employ stability regularizers (temporal smoothing and confidence-weighted neighbor alignment), mitigating action jitter under bursts. Extensive evaluations show superior QoS and channel utilization, with up to 27.4% lower service delay and over 18% higher SLA Satisfaction Rate (SSR) compared with strong baselines.	10.1109/TNSM.2026.3673188
Jing Zhang, Chao Luo, Rui Shao	MTG-GAN: A Masked Temporal Graph Generative Adversarial Network for Cross-Domain System Log Anomaly Detection	2026	Early Access	Anomaly detection Adaptation models Generative adversarial networks	Anomaly detection of system logs is crucial for the service management of large-scale information systems. Nowadays, log anomaly detection faces two main challenges: 1) capturing evolving temporal dependencies between log events to adaptively tackle with emerging anomaly patterns, 2) and maintaining high detection capabilities across varies data distributions. Existing methods rely heavily on domain-specific data features, making it challenging to handle the heterogeneity and temporal dynamics of log data. This limitation restricts the deployment of anomaly detection systems in practical environments. In this article, a novel framework, Masked Temporal Graph Generative Adversarial Network (MTG-GAN), is proposed for both conventional and cross-domain log anomaly detection. The model enhances the detection capability for emerging abnormal patterns in system log data by introducing an adaptive masking mechanism that combines generative adversarial networks with graph contrastive learning. Additionally, MTG-GAN reduces dependency on specific data distribution and improves model generalization by using diffused graph adjacency information deriving from temporal relevance of event sequence, which can be conducive to improve cross-domain detection performance. Experimental results demonstrate that MTG-GAN outperforms existing methods on multiple real-world datasets in both conventional and cross-domain log anomaly detection.	10.1109/TNSM.2026.3654642
Kang Liu, Jianchen Hu, Donglai Ma, Xiaoyu Cao, Yuzhou Zhou, Lei Zhu, Li Su, Wenli Zhou, Xueqi Wu, Feng Gao	Topology-Aware Virtual Machine Placement through the Buffer Migration Mechanism	2026	Early Access	Central Processing Unit Filtering Filters	The virtual machine (VM) placement considering the topology constraints is difficult because the unpredictable topological VMs raise additional structural requirements (including the affinity, anti-affinity and fault-domain) on the resource pool. Thus, the service level agreement (SLA) can be violated even when the occupancy of the resource pool is quite modest. In order to solve this problem, we propose an efficient buffer-migration-based heuristic online algorithm. First, we build an integer programming model for the topology-aware VM placement problem. Second, we propose a hierarchical resource-preserving online approach, where the Rack and physical machine (PM) nodes are selected in the upper and lower layers respectively. Finally, we utilize the buffer to place and migrate the unfitted VMs to enhance the capacity of the resource pool. The proposed approach is tested with high proportional topological VM requests (nearly 60%) in the resource pool with the scale of 500, 1000 and 1500 PMs. The results show that our online approach (with unknown upcoming VM information) can achieve more than 85% of the performance for the offline approach (with complete upcoming VM information). The latency is lower than 5ms per VM.	10.1109/TNSM.2026.3678976
Wangqing Luo, Jinbin Hu, Hua Sun, Pradip Kumar Sharma, Jin Wang	SALB: Security-Aware Load Balancing for Large Language Model Training in Datacenter Networks	2026	Early Access	Training Load management Packet loss	To meet the massive compute and high-speed communication demands of Large Language Model (LLM) training, modern datacenters typically adopt multipath topologies such as Fat-Tree and Clos to host parallel jobs across hundreds to thousands of GPUs. However, LLM training exhibits periodic, high-bandwidth communication patterns. Existing load-balancing schemes become misaligned under dynamic congestion and anomalous surges: they struggle to promptly mitigate iteration-peak congestion and lack effective isolation of anomalous traffic. To address this, we propose Security-Aware Load Balancing (SALB) for LLM training. SALB leverages a Deep Reinforcement Learning (DRL) controller with queue and delay signals for packet-level multipath load balancing and employs path binding to confine suspicious flows. By integrating data security into load balancing, SALB simultaneously achieves high throughput and robust traffic isolation. NS-3 simulation results show that, compared with CONGA, Hermes, and ConWeave, SALB reduces the 99th-percentile flow completion time (FCT) of short flows by an average of 65% and increases the throughput of long flows by an average of 54%. It further outperforms the baselines in aggregate throughput, path utilization, and packet loss rate, thereby significantly enhancing system stability, robustness, and data security.	10.1109/TNSM.2026.3678979
Ei Theingi, Lokman Sboui, Diala Naboulsi	Adaptive and Energy-Efficient Deployment of Robotic Airborne Base Stations: A Deep Reinforcement Learning Approach	2026	Early Access	Energy efficiency Base stations Adaptation models	The increasing energy demands of future wireless networks drive the need for intelligent and adaptive deployment strategies. Traditional methods often lack the flexibility required to handle the spatio-temporal fluctuations inherent in modern communication environments. To address this challenge, we investigate the energy-efficient deployment of Robotic Airborne Base Stations (RABSs) in practical scenarios, such as managing sudden traffic surges during large-scale public events and providing emergency coverage in disaster-stricken areas where terrestrial infrastructure is compromised. We propose a novel Deep Reinforcement Learning (DRL)-based framework for an energy-efficient deployment of multiple RABSs. Unlike existing approaches, our framework features both centralized and decentralized Actor-Critic DRL, enabling scalable and adaptive decision-making. The centralized model leverages global network information to optimize the collective deployment of RABSs, while the multi-agent decentralized approach allows RABSs to make independent yet coordinated decisions based on local observations, ensuring scalability in large-scale networks. In addition, we introduce a state-action representation that captures spatio-temporal traffic variations and energy consumption dynamics. Our simulations validate the effectiveness of the proposed framework, demonstrating significant improvements in energy efficiency and adaptability compared to heuristic, Gauss-Markov, and Q-Learning models. Furthermore, comparison with an exhaustive search benchmark confirms that our approach achieves an optimal energy efficiency with significantly lower computational complexity.	10.1109/TNSM.2026.3678488
Archana Ojha, Om Jee Pandey, Prasenjit Chanak	Energy-Efficient Network Cut Detection and Recovery Mechanism for Cluster-Based IoT Networks	2026	Early Access	Wireless sensor networks Data collection Energy consumption	Recently, the Internet of Things (IoT) has found widespread applications in diverse fields, including environmental monitoring, Industry 4.0, smart cities, and smart agriculture. In these applications, sensor nodes form Wireless Sensor Networks (WSNs) and collect data from the monitoring environment. Sensor nodes are vulnerable to various faults, including battery depletion and hardware malfunctions. These faulty nodes cut/partition the network into several isolated segments. Therefore, several non-faulty nodes become disconnected from the Base Station (BS)/Sink and are unable to transmit their data to the BS. It is subject to the early demise of the network. Network cuts also significantly degrade overall network performance. Once the network is divided into isolated segments, it is very difficult to detect and collect data from them. Therefore, this paper proposes a Mobile Data Collector (MDC)-based data-gathering approach for WSNs to collect data from isolated segments. This paper proposes a novel MDC-based network cut detection algorithm that identifies the formation of network cuts in WSNs. A network recovery algorithm is also proposed to enable data collection from the isolated segment. Furthermore, this paper proposes a Reinforcement learning Brain Storm Optimization (RLBSO) algorithm for optimal selection of Rendezvous Points (RPs) and optimal MDC path design. It significantly reduces data-gathering time across isolated network segments. The simulation and testbed results show that the proposed approach outperforms existing state-of-the-art approaches in terms of network lifetime, data collection ratio, energy consumption, and latency.	10.1109/TNSM.2026.3677868
Jianwei Zhang, Bowen Cui	Bandwidth-Delay Optimal Segment Routing: Upper-Bound and Lower-Bound Algorithms	2026	Early Access	Routing Optimization Quality of service	Segment routing (SR) is a novel source routing paradigm that enables network programmability. However, existing research rarely considers multicriteria optimization problems in SR networks. Given the critical role of bandwidth and delay in quality-of-service (QoS) routing, we formally define the bandwidth-delay optimal SR (BDoSR) problem for the first time and prove its NP-hardness. By leveraging the label correcting algorithm schema, we design a suite of polynomial-time algorithms, including an upper-bound algorithm (BDoSR-UB) and a lower-bound algorithm (BDoSR-LB). BDoSR-UB enables rapid estimation of the optimal solution while BDoSR-LB is accuracy-adjustable and delivers (near-)optimal feasible solutions. We rigorously analyze their performance gap through carefully constructed network examples, providing deep insights into the adjustable parameters of BDoSR-LB. Finally, we validate our algorithms on realistic network topologies, demonstrating that both BDoSR-UB and BDoSR-LB frequently converge to the optimal solution in practice while offering superior computational efficiency compared to existing approaches.	10.1109/TNSM.2026.3678190
Basharat Ali, Guihai Chen	MIRAGE-DoH: Metamorphic Intelligence and Resilient AI Grid for Autonomous Governance of Encrypted DNS	2026	Early Access	Cryptography Domain Name System Fingerprint recognition	Existing DNS over HTTPS defenses have demonstrated limited resilience against polymorphic traffic shaping, staged tunneling, and adaptive mimicry, largely because they rely on static learning pipelines and rigid cryptographic configurations. MIRAGE-DoH was designed to examine whether adaptive inference, persistent structural encoding, and calibrated cryptographic agility could be integrated into a deployable and measurable encrypted DNS control architecture. The framework combined flow-level Cognitive MetaAgents capable of internal reconfiguration, Topological Memory Networks that preserved stable geometric irregularities across temporal windows, and Metamorphic Cryptographic Shards that adjusted key encapsulation policies according to empirically calibrated threat severity. A Causal Counterfactual Environment modeled constrained attacker decision pathways, while Spectral Game Intelligence analyzed flow interaction graphs to anticipate structural attack transitions.Evaluation on extended CIC-DoH2023 and Gen-C-DDD-2022 datasets was conducted under fixed flow-level decision intervals with explicit accounting for synchronization overhead, spectral graph construction cost, and cryptographic rotation latency. Cross-dataset experiments yielded a mean detection accuracy of 97.8% with a 0.41% false positive rate, sustaining median inference latency of 62μs and cryptographic morph latency of 3.7 ms under load. Quantum-assisted inference was assessed through bounded simulations, indicating constrained information gain within the adopted lattice-based configuration, without asserting unconditional post-quantum immunity. These results demonstrated that adaptive encrypted DNS governance can be empirically grounded, operationally bounded, and stress-evaluated without reliance on unqualified claims of perfect security.	10.1109/TNSM.2026.3677474
Deemah H. Tashman, Soumaya Cherkaoui	Trustworthy AI-Driven Dynamic Hybrid RIS: Joint Optimization and Reward Poisoning-Resilient Control in Cognitive MISO Networks	2026	Early Access	Reconfigurable intelligent surfaces Reliability Optimization	Cognitive radio networks (CRNs) are a key mechanism for alleviating spectrum scarcity by enabling secondary users (SUs) to opportunistically access licensed frequency bands without harmful interference to primary users (PUs). To address unreliable direct SU links and energy constraints common in next-generation wireless networks, this work introduces an adaptive, energy-aware hybrid reconfigurable intelligent surface (RIS) for underlay multiple-input single-output (MISO) CRNs. Distinct from prior approaches relying on static RIS architectures, our proposed RIS dynamically alternates between passive and active operation modes in real time according to harvested energy availability. We also model our scenario under practical hardware impairments and cascaded fading channels. We formulate and solve a joint transmit beamforming and RIS phase optimization problem via the soft actor-critic (SAC) deep reinforcement learning (DRL) method, leveraging its robustness in continuous and highly dynamic environments. Notably, we conduct the first systematic study of reward poisoning attacks on DRL agents in RIS-enhanced CRNs, and propose a lightweight, real-time defense based on reward clipping and statistical anomaly filtering. Numerical results demonstrate that the SAC-based approach consistently outperforms established DRL base-lines, and that the dynamic hybrid RIS strikes a superior trade-off between throughput and energy consumption compared to fully passive and fully active alternatives. We further show the effectiveness of our defense in maintaining SU performance even under adversarial conditions. Our results advance the practical and secure deployment of RIS-assisted CRNs, and highlight crucial design insights for energy-constrained wireless systems.	10.1109/TNSM.2026.3660728
Henghua Zhang, Jue Chen, Haidong Peng, Junru Chen	MAT4PM: Machine Learning-Guided Adaptive Threshold Control for P4-based Monitoring in SDNs	2026	Early Access	Monitoring Switches Accuracy	This paper presents MAT4PM, a P4-based proactive monitoring framework designed for Software-Defined Networking (SDN). This is the first monitoring framework that combines Programmable Data Plane (PDP) capabilities for event-driven data collection with control plane intelligence for real-time threshold optimization. The architecture consists of a lightweight P4-based monitoring module deployed at the switch, a Machine Learning (ML) inference engine running at the controller, and a P4Runtime feedback channel for real-time threshold updates. Traffic features are leveraged to predict optimal monitoring thresholds, which are then synchronized with the data plane. A composite cost function is introduced to jointly consider monitoring error and communication overhead, guiding the model toward a balanced trade-off between accuracy and efficiency. Experimental evaluation on BMv2 software switches demonstrates that, compared to static threshold strategies, MAT4PM reduces monitoring error to 7.0% and achieves a 5.6% reduction in overall cost, while maintaining sub-millisecond inference latency and minimal resource consumption. These results demonstrate the practical viability and scalability of MAT4PM in SDN environments.	10.1109/TNSM.2026.3677416
Dongyi Han, Qiang Zhi	DGDPFL: Dynamic Grouping and Privacy Budget Adjustment for Federated Learning in Networked Service Management	2026	Vol. 23, Issue	Privacy Data privacy Adaptation models	In federated learning (FL), effective client and privacy management are crucial for maintaining system efficiency and model performance. However, existing FL frameworks face challenges such as imbalanced client contributions, inefficient resource allocation, and static privacy mechanisms, making scalable client management and adaptive privacy control essential. To address these issues, this paper proposes DGDPFL, a novel FL framework that enhances client selection, resource management, and privacy control through dynamic client grouping and adaptive privacy budgeting. The framework optimizes client management by clustering participants based on device capabilities, bandwidth, and data quality, enabling efficient resource allocation. A contribution-aware selection mechanism ensures fair participation, while a privacy-aware control strategy dynamically adjusts privacy budgets based on model similarity, improving both privacy guarantees and learning performance. We evaluate DGDPFL in real-world and simulated environments. On CIFAR-10 and Fashion-MNIST, DGDPFL achieves 77.83% and 88.35% test accuracy respectively with only 10–20 clients and 40 training rounds, outperforming state-of-the-art baselines by up to 12.36%. On audio datasets FSDD and SAD, the accuracy reaches up to 97%, validating the method’s robustness across modalities. Experimental results demonstrate that DGDPFL outperforms existing approaches by achieving higher model accuracy, improved system efficiency, and better privacy-utility balance. These findings highlight DGDPFL’s effectiveness in managing clients and privacy in FL environments.	10.1109/TNSM.2025.3640713
Long Chen, Yukang Jiang, Zishang Qiu, Donglin Zhu, Zhiquan Liu, Zhenzhou Tang	Toward Energy-Saving Deployment in Large-Scale Heterogeneous Wireless Sensor Networks for Q-Coverage and C-Connectivity: An Efficient Parallel Framework	2026	Vol. 23, Issue	Wireless sensor networks Metaheuristics Sensors	Efficient deployment of thousands of energy-constrained sensor nodes (SNs) in large-scale wireless sensor networks (WSNs) is critical for reliable data transmission and target sensing. This study addresses the Minimum Energy Q-Coverage and C-Connectivity (MinEQC) problem for heterogeneous SNs in three-dimensional environments. MnPF (Metaheuristic–Neural Network Parallel Framework), a two-phase method that can embed most metaheuristic algorithms (MAs) and neural networks (NNs), is proposed to address the above problem. Phase-I partitions the monitoring region via divide-and-conquer and applies NN-based dimensionality reduction to accelerate parallel optimization of local Q-coverage and C-connectivity. Phase-II employs an MA-based adaptive restoration strategy to restore connectivity among subregions and systematically assess how different partitioning strategies affect the number of restoration steps. Experiments with four NNs and twelve MAs demonstrate efficiency, scalability, and adaptability of MnPF, while ablation studies confirm the necessity of both phases. MnPF bridges scalability and energy efficiency, providing a generalizable approach to SN deployment in large-scale WSNs.	10.1109/TNSM.2025.3640070
Eduardo Castilho Rosa, Daniel Nunes Corujo, Flávio de Oliveira Silva	CoFIB: A Memory-Efficient NDN FIB Design for Programmable Edge Switches	2026	Vol. 23, Issue	Data structures Random access memory Pipelines	Designing high-performance and memory-efficient data structures for the Forwarding Information Base (FIB) in Named-Data Networking (NDN) is a challenging task. Since the FIB size is orders of magnitude larger than IP routing tables, scaling it to store millions of prefixes in SRAM/TCAM memory in programmable switches is an open problem. To address this issue, we propose a Compressed FIB data structure called CoFIB, designed to run on edge programmable switches in an SDN-based environment. The CoFIB is a collection of exact-match tables placed in an optimized manner in both ingress and egress pipelines. We propose a LNPM algorithm that carefully recirculates packets in the pipeline. We also introduce the concept of canonical name prefixes to reduce memory footprint and propose an algorithm to extract canonical prefixes from the Routing Information Base (RIB). Experimental results show that CoFIB can compress memory up to $16.58{\times }$ compared to the state-of-the-art, with no significant impact on throughput compared to hardware-based solutions. Additionally, our proposed table placement optimization for LNPM increases the number of packets processed at a line rate by 23.17% compared to the linear table placement approach using a large NDN name dataset.	10.1109/TNSM.2025.3641145
Xi Xu, Yang Yang, Wei Huang, Songtao Guo, Guiyan Liu	VNF-FG Placement and Admission Control in SDN and NFV-Enabled IoT Networks: A Hierarchical Deep Reinforcement Learning Method	2026	Vol. 23, Issue	Admission control Feature extraction Virtual links	Software Defined Networking (SDN) and Network Function Virtualization (NFV) are expected to provide greater flexibility and manageability for next-generation IoT networks. In this context, network services should be modeled as Virtual Network Function Forwarding Graphs (VNF-FGs). A key challenge is efficient allocation of resources for sequentially arriving network service requests, a process known as VNF-FG placement. Most existing algorithms either manually or partially extract features from the physical network and VNF-FG or adopt a greedy approach, allocating resources as long as a feasible solution exists, which may over-allocate resources to VNF-FG requests, ultimately harming infrastructure providers’ long-term revenue. In this paper, we propose a VNF-FG placement and admission control algorithm based on hierarchical reinforcement learning, called EAC. It consists two levels of agents: a coarse-level agent that generates placement strategies and rejects requests with no feasible placement strategies, and a refine-level agent that implements admission control and rejects requests that are detrimental to long-term revenue. To fully capture the topological features of both the physical network and the VNF-FG, we employ a customized Graph Attention Network (GAT) that incorporates link feature awareness and enables deeper exploration. To fully explore historical temporal information for admission control, we construct state triples and feed them into a Recurrent Neural Network (RNN). Using Proximal Policy Optimization (PPO) as the foundational training algorithm, the corresponding agents are trained hierarchically. Extensive experimental results demonstrate that the proposed EAC algorithm outperforms existing state-of-the-art solutions in terms of acceptance rate, revenue-to-cost ratio, and long-term average revenue.	10.1109/TNSM.2025.3640927
Junfeng Tian, Junyi Wang	D-Chain: A Load-Balancing Blockchain Sharding Protocol Based on Account State Partitioning	2026	Vol. 23, Issue	Sharding Blockchains Load modeling	Sharding has become one of the key technologies for improve the performance of blockchain systems. However, the imbalance of transaction load between shards caused by extremely hot accounts leads to an imbalance in the utilization of system resources as well as the increase of cross-shard transactions with the number of shards limits the expansion of sharding systems, and sharding systems do not achieve the desired performance improvement. We propose a new blockchain sharding system called D-Chain. D-Chain splits and distributes the state of extremely hot accounts into multiple shards, allowing transactions for an account can be processed in multiple shards, thus balancing the load between shards and reducing the number of cross-shard transactions. We have implemented a prototype of D-Chain, and evaluated its performance using real-world Ethereum transactions. Experimental results show that the proposed system achieves a more balanced shard load and outperforms other baselines in terms of throughput, transaction latency, and cross-shard transaction ratio.	10.1109/TNSM.2025.3640097
Qiang Zou, Yuhui Deng, Yifeng Zhu, Yi Zhou, Jianghe Cai, Shuibing He, Lina Ge	Analyzing Request Volatility in Cloud-Based Machine Learning: Insights From Alibaba’s Machine Learning as a Service Platform	2026	Vol. 23, Issue	Machine learning Training Correlation	With advancements in machine learning (ML) technology and the deployment of large ML-as-a-Service (MLaaS) clouds, accurately understanding request behaviors in an MLaaS cloud platform is paramount for resource scheduling and optimization. This paper sheds light on the correlation of request arrivals in a representative and dynamic MLaaS workload – Alibaba PAI (an ML platform for artificial intelligence). For requests in the PAI workloads at the job, task, instance, and machine levels, our burstiness diagnosis reveals that the request arrival processes at all levels are significantly bursty. Additionally, our Gaussianity test indicates that the bursty activities in PAI consistently appear to be non-Gaussian. Our findings show that there exists a certain degree of correlation between request arrivals at each level over long-term time scales. Moreover, we reveal the self-similar nature of request activities in the various-level wild MLaaS workloads on Alibaba PAI through visual evidence, the auto-correlation structure of the aggregated process of request sequences, and Hurst parameter estimates. Furthermore, we implement a versatile workload synthetic model to synthesize request series based on the inputs measured from the PAI trace. Experimental results demonstrate that our model outperforms typical self-similar workload models, and can improve accuracy by up to 99% compared to them.	10.1109/TNSM.2025.3640771