arXiv 论文速递

Network traffic classification enables website fingerprinting, intrusion detection, and quality of service management. However, developing methods that capture stable and generalizable traffic patterns remains challenging. We introduce FlowCLIP, a contrastive pretraining framework for learning traffic representations using only side-channel features: packet inter-arrival times, packet sizes, and packet directions. FlowCLIP feeds traffic flow features into a traffic encoder and the corresponding domain names into a text encoder, aligning their learned representations through a CLIP-style contrastive loss. After pretraining, we freeze the traffic encoder and evaluate it through linear probing. We evaluate FlowCLIP on the CESNET-QUIC22 dataset used for pretraining under a time-based protocol, where the model is trained on earlier traffic and evaluated on later traffic. We also assess whether the pretrained representations transfer to a separate dataset, the UC Davis dataset. In the time-based evaluation, FlowCLIP outperforms XGBoost and 1-NN and performs comparably to an end-to-end supervised model with the same architecture in accuracy and weighted F1. In the cross-dataset transfer evaluation, FlowCLIP outperforms an end-to-end supervised model trained on the UC Davis training data when evaluated on human-triggered traffic but performs slightly worse on script-triggered traffic. Through these results, we show that domain names can be used directly for traffic representation learning, enabling pretraining without predefined traffic classes.

Network traffic classification enables website fingerprinting, intrusion detection, and quality of service management.

LLM deployment is migrating from data centers to edge devices, where Mixture-of-Experts (MoE) models offer a promising path: sparse expert activation allows the model to be spread across multiple low-cost edge nodes. Distributed MoE inference repeatedly dispatches embeddings from one main node to many workers - a one-to-many pattern poorly served by the sequential unicasts of mainstream stacks (NCCL, TCP), yet naturally matched by UDP broadcast. We propose a UDP broadcast method for collaborative edge MoE inference, augmented with timeout-driven retransmission exploiting near deterministic latency in distributed MoE for reliability and unordered result gathering for robustness to expert mispredictions, yielding a consistent 1.4x speedup over NCCL and TCP on a wired 8-node cluster. In wireless settings, however, we uncover a deeper, long-forgotten bottleneck: IEEE 802.11 caps broadcast rates at 54 Mbps regardless of physical-layer capacity - a legacy policy built for sparse control traffic, not edge AI. NS-3 simulations at distances 1m, 2m and 5m show that the optimal rates are much higher (64x, 43x, and 32x, respectively) than the 54 Mbps cap applied in standard. Thus, we argue that broadcast is no longer a control-plane relic: it is time for Wi-Fi standards to treat it as a high-throughput data-plane citizen.

LLM deployment is migrating from data centers to edge devices, where Mixture-of-Experts (MoE) models offer a promising path: sparse expert activation allows the model to be spread across multiple low-cost edge nodes.

The modern transport protocol QUIC is designed to enhance network performance and security, but it remains vulnerable to handshake flooding attacks. Such attacks exhaust CPU resources by forcing the server to perform expensive cryptographic operations via a large number of handshaking requests. QUIC provides a built-in defense mechanism, the Retry mechanism, to mitigate these attacks. However, our experiments reveal that it can still become a performance bottleneck under large-scale QUIC handshake floods due to substantial computational overhead. In this paper, we design and implement TurboRetry, a split design, that offloads the Retry mechanism onto DPUs to efficiently mitigate QUIC handshake floods. TurboRetry partitions the tasks of the Retry into two categories, and then assigns them to the DPUs and the host, respectively. To preserve QUIC semantics and reduce the coordination overhead, TurboRetry designs an extended Retry token format and an efficient cooperation scheme. In addition, TurboRetry offloads the connection authorization task to the on-path DPA to further improve both performance and security. Our evaluation shows that TurboRetry outperforms the host-side implementation by a wide margin, improving throughput by 10-20$\times$.

The modern transport protocol QUIC is designed to enhance network performance and security, but it remains vulnerable to handshake flooding attacks.

In Robot Operating System 2 (ROS 2), Data Distribution Service (DDS) participants must discover one another before exchanging data. In wireless environments, delayed or lost discovery messages cause reliability timers to expire, triggering retransmissions that intensify channel contention and further delay the delivery of discovery messages. This self-reinforcing feedback can escalate into a discovery storm. Existing models characterize discovery demand under fixed delivery conditions, but do not capture how shared-channel delay changes protocol state and generates further traffic. To address this issue, we present the first closed-loop analytical model of ROS 2 discovery that characterizes how delay-induced feedback amplifies retransmission overhead and leads to severe discovery storms. Our model represents channel contention as a shared service process, coupling message-delivery latency with receiver states and reliability timers. The model predicts both discovery completion time and per-class message counts. We validate the model through 1,350 experimental runs across 90 topology configurations. An open-loop airtime baseline captures only a fraction of the high-load completion time. The closed-loop model reproduces this rise and conservatively upper-bounds the observed high-load range. Guided by insights from the model, we further design a response-aware discovery policy that reduces mean discovery completion time by 25.3% to 39.7%.

In Robot Operating System 2 (ROS 2), Data Distribution Service (DDS) participants must discover one another before exchanging data.

Post-quantum cryptography (PQC) is becoming an important component of long-term trust in Internet-facing infrastructure. Publicly observable PQC support provides evidence of externally visible deployment, but does not necessarily reflect the overall progress of an organisation's post-quantum migration. This distinction matters when observable deployment is used as an indicator of organisational readiness or progress towards migration deadlines. We present a measurement study of observable PQC deployment across 4,665 UK organisations spanning ten sectors. We measure post-quantum key-exchange support across HTTPS and SMTP STARTTLS endpoints, attribute reachable endpoints to their underlying infrastructure providers where possible, and statistically examine protocol-, sector-, and provider-level deployment patterns. Among reachable endpoints, 44.0\% of HTTPS services supported at least one evaluated PQC key-exchange group, compared with 6.4\% of SMTP services. Among organisations reachable over both protocols, HTTPS support was significantly more common than SMTP support (matched odds ratio 16.89). Although deployment varied across sectors, infrastructure provider identity was substantially more predictive than organisational sector, and observable deployment was highly concentrated among a small number of providers. Only 144 organisations supported PQC across both web and email infrastructure, highlighting an uneven and fragmented migration landscape. No post-quantum certificate signatures were observed across the measured endpoints. These findings show that observable PQC deployment is currently shaped predominantly by infrastructure-provider deployment decisions and should not be interpreted as a complete measure of organisational migration readiness.

Post-quantum cryptography (PQC) is becoming an important component of long-term trust in Internet-facing infrastructure.

Evaluating congestion control is inherently challenging because performance depends on the interaction between the congestion-control algorithm, transport stack, application behaviour, measurement process, and network dynamics. This challenge is growing as state-of-the-art protocols incorporate pacing, selective loss recovery, model-based control, and, more recently, reinforcement learning. Low Earth Orbit (LEO) satellite networks are a particularly demanding setting: rapidly changing paths, handovers, non-congestive loss, RTT variation, and transient hotspots all affect transport behaviour. This paper reports the lessons learned from an extensive evaluation campaign across both simulation and emulation for LEO satellite congestion control. We compare multiple classes of congestion-control algorithms, including Cubic, BBR variants, LEO-specific protocols, and reinforcement-learning-based control, using comparable implementations across OMNeT++/INET simulation and Mininet-based emulation with the Linux transport stack. This gives us a rare opportunity to examine not only protocol performance, but also the methodological strengths and limitations of each experimental environment. Our findings show that simulation is indispensable for constellation-scale exploration, controlled parameter sweeps, and future deployment studies, but can miss behaviours caused by production transport-stack mechanisms such as pacing, SACK, RACK, kernel timing, and rate sampling. Emulation exposes these implementation-dependent effects and provides a necessary validation step, but is harder to scale and less exactly repeatable. We distil these experiences into practical lessons for combining simulation and emulation to obtain results that are scalable, reproducible, and deployment-relevant.

Evaluating congestion control is inherently challenging because performance depends on the interaction between the congestion-control algorithm, transport stack, application behaviour, measurement process, and network dynamics.

The Open Radio Access Network (O-RAN) architecture introduces flexible functional splits and open interfaces that enable distributed and centralized deployment of baseband processing. While this flexibility offers opportunities for improved resource utilization, it also introduces fundamental trade-offs between energy efficiency and latency. In this paper, we develop a throughput-based end-to-end energy consumption model for O-RAN and extend it by incorporating detailed latency modeling and application-specific Artificial Intelligence/Machine Learning inference costs. The proposed end-to-end modeling framework provides a general representation of processing, transport, and inference-related energy and delay across the access, metro, and long-haul network segments. Building on this general model, we formulate an optimization problem that selects the placement of baseband processing and AI inference tasks across candidate O-RAN configurations to analyze energy-latency tradeoffs under network load, server frequency, and energy-budget constraints. Using representative hardware platforms and realistic traffic assumptions, we evaluate multiple baseband processing placements corresponding to different O-RAN functional configurations. Our results reveal how user quality of service requirements and network load conditions jointly determine the optimal placement of baseband processing and AI inference tasks, highlighting the inherent trade-off between energy efficiency and latency. The analysis provides practical insights for latency-aware and energy-efficient O-RAN deployments supporting emerging AI-driven services.

The Open Radio Access Network (O-RAN) architecture introduces flexible functional splits and open interfaces that enable distributed and centralized deployment of baseband processing.

This paper investigates collaborative mobile edge computing (MEC) servers for large language model (LLM) inference under soft deadline constraints. In this system, to improve the quality of service, computations are expected to be completed within their deadlines. However, due to dependencies among tasks or subtasks, any missed deadline can lead to catastrophic consequences for the entire request. In this context, this work proposes an extended deadline mechanism with constrained flexibility. The main challenges lie in handling large-scale computations under strict latency constraints while limiting the number of allowable deadline extensions, especially in the presence of task dependencies within each request. To tackle these challenges, we develop a transformer-enhanced proximal policy optimization (PPO) framework that enables efficient collaboration among MEC servers. The proposed approach aims to maximize the number of tasks completed within their deadlines while minimizing the use of deadline extensions. By capturing temporal dependencies and cross-server interactions, the transformer improves decision-making for task migration. Simulation results demonstrate that the proposed method significantly outperforms conventional PPO and heuristic-based approaches in terms of task completion rate and overall system efficiency.

This paper investigates collaborative mobile edge computing (MEC) servers for large language model (LLM) inference under soft deadline constraints.

Deploying expressive learning models directly on programmable dataplanes promises line-rate, low-latency traffic analysis but remains hindered by strict hardware constraints and the need for predictable, auditable behavior. Chimera introduces a principled framework that maps attention-oriented neural computations and symbolic constraints onto dataplane primitives, enabling trustworthy inference within the match-action pipeline. Chimera combines a kernelized, linearized attention approximation with a two-layer key-selection hierarchy and a cascade fusion mechanism that enforces hard symbolic guarantees while preserving neural expressivity. The design includes a hardware-aware mapping protocol and a two-timescale update scheme that together permit stable, line-rate operation under realistic dataplane budgets. The paper presents the Chimera architecture, a hardware mapping strategy, and empirical evidence showing that neuro-symbolic attention primitives can achieve high-fidelity inference within the resource envelope of commodity programmable switches.

Deploying expressive learning models directly on programmable dataplanes promises line-rate, low-latency traffic analysis but remains hindered by strict hardware constraints and the need for predictable, auditable behavior.

Routing in Low Earth Orbit (LEO) satellite networks is challenging due to highly dynamic topologies and spatio-temporal network conditions. Reinforcement Learning (RL) has emerged as a promising approach for adaptive routing; however, its performance critically depends on reward function design, which must balance objectives such as goodput and end-to-end delay. In practice, reward design remains a complex manual process requiring significant domain expertise and extensive trial-and-error. Recent works have explored Large Language Models (LLMs) for automated reward design, but their application to highly dynamic systems such as LEO satellite networks remains largely unexplored. We propose LARGE, a framework that automates reward design for RL-based routing by combining LLM- driven generation with iterative simulator-in-the-loop evaluation. LARGE generates an initial reward from LLM prior knowledge and iteratively refines it using simulation feedback. This loop enables exploration of diverse reward formulations while aligning them with network objectives. Results show that LARGE improves reward quality within a few iterations through feedback-driven refinement. Across different backbones, the framework achieves performance comparable to an expert-designed baseline, with the best-performing configuration reaching goodput within approximately 3% of the baseline and slightly lower end-to-end delay, without manual reward engineering. These results indicate that effectiveness emerges from the iterative feedback-driven process enabled by LARGE, highlighting the potential of framework-driven LLM-in-the-loop optimization for RL-based routing in dynamic satellite networks.

Routing in Low Earth Orbit (LEO) satellite networks is challenging due to highly dynamic topologies and spatio-temporal network conditions.

Since the inception of modern communication networks, the quest for operations automation has never ceased. Yet the evolution of network automation is difficult to characterize with a single maturity ladder. Throughout this history, network control systems have expanded their capabilities for observation, decision support, routine execution, and operator interaction, but these capabilities have not advanced uniformly. Such uneven progress makes the degree of automation an unreliable proxy for trustworthy network-side actuation. The unresolved question is not simply how much automation a system provides, but under what conditions it can be entrusted to change the network state. This paper examines that question through Network Control Intelligence (NCI), a five-axis framework spanning Decision Logic, Adaptability, Knowledge, Control Delegation, and Interface. We use NCI to organize the evolution of network-control systems into three eras: rule-based and scripted automation, programmable and data-driven control, and Large Language Model (LLM)-enabled network operations. Viewed through this framework, the three eras reveal a persistent asymmetry. None of these gains, however, automatically determines when network control should be trusted to change the network state. We frame trustworthy autonomy as a governed alignment between what a system can infer, what it can verify, and what it is authorized to execute. On that basis, the paper develops a reference architecture that separates proposal generation from governed execution, identifies recurring integration patterns for LLM-enabled operations, and derives a research agenda for higher network autonomy under explicit assurance, safety, and governance constraints.

Since the inception of modern communication networks, the quest for operations automation has never ceased.

LLM inference has become a global-scale, heterogeneous workload spanning agents, retrieval, tool-use, code execution and multi-modal reasoning. These workloads naturally enable context reuse from overlapping inputs, creating a major opportunity to store and reuse the contexts' KV Caches instead of recomputing them. However, model-side advances that shrink the KV Cache and system-side advances that reduce compute, storage, and transfer costs are evolve independently within legacy cloud boundaries. We argue that future inference infrastructure should allow decoupling of compute and KV Cache storage across cloud and datacenters. The network becomes an active distribution channel; bandwidth, latency and pricing directly determines how the KV Cache should be managed. We propose a vision for an Internet for the KV Cache, with KV Cache management working as a content-distribution system. In this view, KV Cache storage and recompute decisions are driven by model, infrastructure, and application metrics, to enable adaptive, content-driven decisions for minimizing latency and cost.

LLM inference has become a global-scale, heterogeneous workload spanning agents, retrieval, tool-use, code execution and multi-modal reasoning.

Reinforcement learning (RL) has emerged as a promising paradigm for Internet congestion control, achieving higher link utilization than classical heuristics. However, RL-based controllers trained in single-flow environments are not guaranteed to share bandwidth equitably when deployed in multi-flow networks. This paper investigates the fairness properties of Aurora~\cite{jay2019aurora}, a state-of-the-art deep RL congestion controller, and evaluates three post-hoc fairness strategies that preserve Aurora's RL architecture: \emph{reward shaping} (Strategy~A), \emph{observation augmentation} (Strategy~B), and \emph{loss-sensitivity tuning} (Strategy~C). Using a custom shared-bottleneck simulator and Jain's fairness index as the primary metric, we find that modest reward shaping achieves the best fairness while preserving aggregate throughput. All strategies maintain the total bandwidth budget with fairness being achieved through redistribution, not reduction. Beyond the 2-flow homogeneous setting, an extended evaluation across mixed Aurora--CUBIC competition and dynamic flow entry/exit scenarios shows that Strategy~C's loss-sensitivity emerges as the most TCP-friendly mechanism, while Strategy~B is the most stable through dynamic flow-set changes.

Reinforcement learning (RL) has emerged as a promising paradigm for Internet congestion control, achieving higher link utilization than classical heuristics.

Underwater acoustic networks face a fundamental rate--concurrency tradeoff: high-rate waveforms (e.g., OFDM, OTFS) are designed for point-to-point links and rely on orthogonal MAC protocols (e.g., TDMA) to avoid collisions, sacrificing concurrency; conversely, collision-resilient waveforms (e.g., CDMA, ZCMod) support uncoordinated access but are inherently rate-limited by spreading or sparse index modulation. We present \system, a cross-layer concurrent random-access system that combines two new components: (i) \textbf{EZCDM}, an equidistant ZC division-multiplexing waveform that activates multiple cyclic shifts of a ZC root as parallel sub-channels with a tunable rate--robustness tradeoff, and an intra-symbol differential receiver that eliminates the shared multipath channel response without explicit CIR estimation; and (ii) a \textbf{cross-layer link adaptation (LA) framework} featuring beacon-framed random access, user-specific closed-loop power control, and overlap- and CIR-aware common-MS selection. Channel-trace- and signal-trace-driven physical-layer experiments combined with PHY-in-the-loop network simulations demonstrate that \system\ achieves significant BER and throughput gains over conventional waveforms and MAC protocols by converting traditionally destructive collisions into decodable concurrent streams.

Underwater acoustic networks face a fundamental rate--concurrency tradeoff: high-rate waveforms (e.g., OFDM, OTFS) are designed for point-to-point links and rely on orthogonal MAC protocols (e.g., TDMA) to avoid collisions, sacrificing concurrency; conversely, collision-resilient waveforms (e.g., CDMA, ZCMod) support uncoordinated access but are inherently rate-limited by spreading or sparse index modulation.

The threat of quantum computers to classical public-key cryptography has created an urgent need to evolve secure communication protocols with post-quantum cryptographic (PQC) primitives. The Noise Protocol Framework, widely used in systems such as WireGuard and WhatsApp, traditionally relies on the Elliptic Curve Diffie-Hellman (ECDH) public-key exchange scheme, which is vulnerable to quantum threats. In this paper, we present NoisePQC++, a unified C++23 implementation of the Noise Protocol framework augmented with post-quantum Key Encapsulation Mechanisms and Hybrid Forward Secrecy. Our design integrates the National Institute of Standards and Technology (NIST) standardized ML-KEM algorithm alongside classical ECDH, enabling full PQC, hybrid ECDH+PQC handshakes, and unified support for all 57 classical Noise handshake pattern variants, 13 post-quantum Noise handshakes, and their hybrid variants. Compared with prior work, NoisePQC++ offers broader protocol coverage, more complete implementation support, and greater flexibility. Our evaluation shows minimal overhead under normal network conditions and acceptable overhead in adverse cases, while significantly improving resistance against quantum adversaries. These results indicate that NIST-standardized post-quantum and hybrid Noise handshakes are practical and provide a credible basis for future deployment.

The threat of quantum computers to classical public-key cryptography has created an urgent need to evolve secure communication protocols with post-quantum cryptographic (PQC) primitives.

Cooperative perception (CP) enables connected autonomous vehicles (CAVs) to share complementary observations for safer navigation, but practical deployment is limited by bandwidth constraints, unreliable links, and redundant information exchange. Existing CP methods often assume predefined participants and merely focus on collective perception. Likewise, recent LLM-based cooperative driving frameworks facilitate multi-vehicle reasoning but do not regulate participation criteria to select more beneficial vehicles. To bridge this gap, we propose an LLM-assisted coalition formation framework that selects the most informative helper vehicles before LLM reasoning. The approach jointly optimizes perceptual diversity using a determinantal point process (DPP) over multimodal vehicle embeddings and communication-aware reliability. This leads to a joint coalition selection and power allocation problem, which we solve efficiently via a relaxed convex reformulation and an ADMM-based optimization strategy that decouples diversity-aware selection from network-aware resource allocation. The selected coalition is then summarized and provided with an LLM reasoning module for efficient and less redundant multi-vehicle decision support. Experimental results show that our approach outperforms other baselines in overall coalition value, while maintaining high diversity and improved networking efficiency. The framework achieves a better balance between task performance and safety across OPV2V and V2V4Real datasets, demonstrating its effectiveness for cooperative autonomous driving with communication constraints.

Cooperative perception (CP) enables connected autonomous vehicles (CAVs) to share complementary observations for safer navigation, but practical deployment is limited by bandwidth constraints, unreliable links, and redundant information exchange.

Serving Mixture-of-Experts (MoE) large language models across distributed edge servers is bottlenecked by the cross-server expert transmission. The existing approaches mainly focus on how to reach a remote expert faster. However, in this paper, we instead consider whether a given layer, and the layers after it, need to be executed at all. To this end, a communication-aware adaptive-depth framework is proposed in this paper, termed TrimMoE, which couples layer skipping and confidence-based early exit with substitute execution and server-expert selection under a unified quality budget. Specifically, in the offline stage, TrimMoE freezes the backbone, trains the lightweight per-layer exit heads, calibrates the per-layer importance thresholds, and allocates the expert replicas by a skip/exit-aware redundancy benefit. In the online stage, a transition-aware look-ahead anticipates the token movement, so that the depth reduction targets the costliest transmissions, and besides, two feedback rules adapt the delay-quality weights and the exit threshold. Moreover, we prove that the substitution-and-skipping proxy degradation never exceeds the configured budget, and that the early exit is admitted only under a calibrated confidence gate. On a heterogeneous 10-server testbed with Switch-Base-8E, Qwen-MoE-A2.7B, and Mixtral-8x7B, TrimMoE reduces the average latency by up to 62.8%, lowers the cross-server traffic and the remote-execution ratio, and sustains high throughput under load, while keeping the task-quality degradation within a 2% bound.

Serving Mixture-of-Experts (MoE) large language models across distributed edge servers is bottlenecked by the cross-server expert transmission.

Traffic shockwaves are stop-and-go waves that propagate upstream through the streams of vehicles and are one of the major causes of traffic congestion, fuel inefficiency, and increased accident rates in modern transportation systems. Although Connected and Autonomous Vehicles (CAVs) offer a promising opportunity to mitigate such shockwaves, most existing control strategies rely on global traffic state information, making them impractical for early-stage deployment of Vehicular Ad-hoc Networks (VANETs). In this paper, we propose a decentralized Multi-Agent Reinforcement Learning (MARL) framework that integrates a Graph Neural Network (GNN) to enhance the control architecture of connected and autonomous vehicles. The proposed approach enables vehicles to learn cooperative control policies using locally available information and interaction with neighboring vehicles. The effectiveness of the proposed scheme is evaluated using a scalable simulation environment under realistic highway traffic conditions. Simulation results show that the proposed GNN-based MARL framework can reduce the propagation of traffic shockwaves by up to 80%, even when only 10% of the vehicles are connected.

Traffic shockwaves are stop-and-go waves that propagate upstream through the streams of vehicles and are one of the major causes of traffic congestion, fuel inefficiency, and increased accident rates in modern transportation systems.

Agentic AI networking (AgentNet) systems rely heavily on third-party skillset implementations and distributed multi-agent collaboration, yet they face major claim-to-capability inconsistencies and security vulnerabilities under trust-by-declaration assumptions. To bridge this gap, this paper proposes TrustAgentNet, a dual-tier blockchain-secured zero-trust framework. Specifically, a global Chain of Skillsets (CoS) governs the lifecycle of skillset metadata with protocols empowered by specialized agents to enforce off-chain auditing while maintaining lightweight on-chain cryptographic consensus. Furthermore, transient, task-oriented Chains of Collaboration (CoC) are dynamically established to enable trustless distributed multi-agent collaboration. Theoretical analysis of the three-way trade-off among security level, task performance, and resource overhead is provided and empirically validated. Experimental results on a hardware prototype demonstrate that compared with no-blockchain trust-by-default baselines, the zero-trust overhead of TrustAgentNet is dominated by off-chain inference, while the blockchain layer incurs minor ledger costs via the ledger-IPFS storage and on/off-chain integration design. Crucially, the proposed verification pipeline achieves a flawless 100% accuracy across 50 AI models, correctly validating 40 honest skillsets and intercepting 10 adversarial ones, and generalizes to non-AI domains with an 83.91% accuracy and a 0.85 F1-score across 1478 features from 171 ClawHub skills. Adversarial experiments further show that TrustAgentNet enables autonomous skillset self-recovery against various malicious attacks.

Agentic AI networking (AgentNet) systems rely heavily on third-party skillset implementations and distributed multi-agent collaboration, yet they face major claim-to-capability inconsistencies and security vulnerabilities under trust-by-declaration assumptions.

Industrial and autonomous systems increasingly depend on AI, automation, and real-time coordination to act on operational data as it is generated. Yet conventional architectures often require that data to pass through centralized platforms before decisions can be made. Cloud systems remain valuable for training, reporting, and long-term analytics, but they add latency and external dependencies to the critical decision path and become harder to scale as each new site adds additional edge devices and data. As intelligence spreads across machines, sites, facilities, and vehicles, continued dependence on centralization will constrain response time, resilience, scalability, and autonomous operation. This paper presents the AnyLog Edge Data Fabric, an agent- and edge-based platform that manages operational data at its source while presenting distributed data, assets, compute resources, and services as one logical system. Through its Distributed Metadata Layer, Virtual Data Lake, Unified Namespace, Single System Image, and Model Context Protocol, authorized users, applications, automation services, and AI agents can discover, query, process, and act on distributed resources without knowing where they are hosted. Queries and computation execute at the agents holding the relevant data, so only requests and results traverse the network. This preserves local ownership, reduces data movement, supports continued operation during connectivity disruptions, and enables repeatable deployment from validated digital-twin configurations. AnyLog provides a cloud-like operating model for distributed SQL, real-time automation, Edge AI, federated learning, and resilient decision-making without a single point of failure or any dependence on centralized infrastructure.

Industrial and autonomous systems increasingly depend on AI, automation, and real-time coordination to act on operational data as it is generated.

Large language models (LLMs) offer powerful reasoning capabilities for network traffic analysis, but standard capture formats and their textual equivalents are prohibitively verbose, overflowing LLM context windows by two orders of magnitude. We present PCAP-LM, a flow-centric, LLM-native text representation that acts as a lossy knowledge extraction step rather than a standard compression tool: raw captures are transcoded into semantic summaries using PacketGlyphs - a novel ASCII alphabet coined in this paper that encodes packet direction, TCP/TLS state, log-scale size, and inter-packet delay. Combined with a constrained PMI-BPE tokenizer and motif run-length encoding, repetitive behavioural patterns are aggressively collapsed. A @REFS side-index preserves lossless drill-down into the original packets. Evaluated on a homogeneous corpus of 5G/4G TLS 1.3 bulk-download traffic, the BPE vocabulary fully saturates at 159 tokens, achieving an 812x size reduction over tshark -V and fitting entire captures within a single LLM context window. In a forensic question-answering evaluation over 30 held-out files, a frontier LLM achieves 99.3% accuracy from PCAP-LM documents versus 51.0% from a token-budget-matched tshark -V prefix. The lossy design introduces known blind spots - most notably a 24% false-negative rate for TCP retransmissions - and extending to heterogeneous mixed-protocol environments will require vocabulary retraining.

Large language models (LLMs) offer powerful reasoning capabilities for network traffic analysis, but standard capture formats and their textual equivalents are prohibitively verbose, overflowing LLM context windows by two orders of magnitude.

Iran conducted two nationwide Internet shutdowns in 2026, on January 8-25 and March 1-May 26, the latter lasting 86 days. We characterize both using three complementary measurement planes: six months of daily Censys scan data, BGP analysis of RIPE RIS snapshots spanning 2019-2026, and continuous per-prefix TCP probing from five vantage points. Each plane captures a different aspect of Iranian connectivity, and interpreting any one in isolation can be misleading. Unlike the partial BGP withdrawal of 2019, the 2022 and 2026 shutdowns were enforced by forwarding-plane discard while 80-88% of Iranian prefixes remained announced, leaving control-plane monitors blind. Restoration is similarly invisible to BGP, appearing in our forwarding-plane measurements as a centrally coordinated step. Censys host counts overshoot to approximately 3.6 times their pre-shutdown baseline after both restorations, rather than returning to baseline. Active probing reveals this inflation to be an artifact: most of the 3M apparent hosts are injected UDP/5353 responses synthesized by an on-path element at Iran's international gateway. Finally, AS-path classification shows that some apparent shutdown survivors were routed through foreign upstreams and never traversed the enforcement point. Together, these results show that measuring shutdowns requires reading multiple planes against one another, as no single signal reliably distinguishes genuine connectivity from its absence.

Iran conducted two nationwide Internet shutdowns in 2026, on January 8-25 and March 1-May 26, the latter lasting 86 days.

5G and B5G/6G foundations heavily rely on virtualization technologies, and virtualized Radio Access Networks (vRANs) are one of their major keystones. However, while vRANs have been traditionally suffering from significant hardware/software coupling, next generation vRANs aim for open, standardized interfaces and multi-vendor, interoperable components to enable truly flexible deployments following the cloud-native principles. In this line, the O-RAN Alliance is promoting a novel Open RAN architecture to further boost flexibility and cost efficiency. In order to reduce costs and effectively achieve the promised disaggregation levels, O-RAN must ensure shared, integrated transport networks in opposition to dedicated, over-provisioned links from traditional approaches. However, keeping deterministic performance requirements in such cost-effective networks (i.e., general-purpose Ethernet networks), especially in those interfaces that are time-critical, is a challenge. In this article, we review the most relevant Time Sensitive Networking (TSN) standards that may bring compelling benefits to O-RAN (i.e., IEEE 802.1CM, IEEE 802.1Qbu and IEEE 802.1Qbv) for providing determinism over cost-efficient networks. We explore the design space for a TSN-enabled O-RAN architecture, reporting on the requirements and deployment options and finally, we discuss on the opportunities and challenges that O-RAN will face when adopting TSN technologies to fully open the vRAN ecosystem.

5G and B5G/6G foundations heavily rely on virtualization technologies, and virtualized Radio Access Networks (vRANs) are one of their major keystones.

Increasingly, LLM inference services proxy client requests to engine replicas distributed globally. Load-balancing policies must jointly account for factors including KV-cache locality, replica load, and variable network latency when optimizing for metrics like latency and TTFT. However, existing systems only evaluate a subset of these factors in their cost model, leading to uneven concentrations of load and KV-cache across replicas. We present GORGO, a proxy architecture that holistically factors network latency, prefill cost, and queueing delay using tunable parameters. Since open-source chat datasets such as LMSYS-Chat1M and WildChat-4.8M lack long-context, high prefix-reuse data, we release a synthetic dataset, ART-Chat-2.5M, from long-context production metadata. On a tuning window from ART-Chat-2.5M, evolutionary strategies guide the GORGO policy's parameters to directly optimize p95 TTFT. During held-out evaluation windows, we fix the parameter values learned from tuning and improve p95 TTFT by 6.9-15.5% and p95 end-to-end (E2E) latency by 14.3-30.9% over baseline load-balancing policies such as simple session affinity and prefix-cache. The code and ART-Chat-2.5M dataset can be found at https://github.com/Arcadia-Research-Team/GORGO.

Increasingly, LLM inference services proxy client requests to engine replicas distributed globally.

Semantic communication (SemCom) promises to reduce transmitted payloads by conveying task-relevant meaning instead of raw bits. However, practical SemCom also incurs semantic metadata, control signaling, feedback, model or knowledge-base synchronization, and neural computation costs, which may offset semantic compression gains. This paper develops an overhead-aware analytical framework for quantifying the spectral-resource and energy costs of SemCom under equal task utility. The framework covers point-to-point transmission, user equipment (UE)-to-next-generation NodeB (gNB) uplink, and UE-to-UE communication under a single gNB, and derives closed-form break-even conditions with respect to payload size, semantic compression factor, model reuse, protocol overhead, and computation energy. Simulation results show that SemCom becomes spectrally beneficial only for sufficiently large payloads, while energy gains require larger payloads due to processing and synchronization overheads. The results also show that multi-user downlink is particularly favorable, as shared semantic overheads can be amortized across multiple UEs. These findings provide design guidance for realistic SemCom evaluation and standardization-oriented deployment.

Semantic communication (SemCom) promises to reduce transmitted payloads by conveying task-relevant meaning instead of raw bits.

The Transmission Control Protocol (TCP) relies on deterministic state machines and sequence-number arithmetic to ensure reliable communication. However, hard-coded protocol logic is difficult to adapt to increasingly complex and dynamic network conditions. This paper proposes Smart-TCP, an agentic AI-based transport protocol framework that organizes TCP control logic as a fast/slow model-assisted decision process. Specifically, a feature-aware classifier routes routine segments to a lightweight Small Language Model (SLM) fast path and anomalous or boundary cases to a Large Language Model (LLM) slow path, while an Arithmetic Logic Unit (ALU) handles deterministic sequence- and acknowledgement-number computation. A state module maintains connection state, packet history, and key control variables, enabling protocol decisions over the full session context. We evaluate Smart-TCP through path classification, atomic packet processing, slow-path anomaly response, and end-to-end session interaction. Experimental results show that Smart-TCP achieves 99.14% action accuracy on 700 slow-path anomaly and fallback samples, and a 100% full-lifecycle success rate over 300 ideal fast-path end-to-end session windows. These results suggest that decoupling model-based protocol reasoning from deterministic arithmetic improves the reliability of model-driven TCP control for network service operation.

The Transmission Control Protocol (TCP) relies on deterministic state machines and sequence-number arithmetic to ensure reliable communication.

The rapid development of the Internet of Everything (IoE) is accelerating the adoption of intelligent applications. However, the massive number of connected devices generates diverse and heterogeneous tasks, which pose increasing challenges for dynamic resource scheduling in IoE environments. Using their superior semantic understanding and reasoning capabilities, Large Artificial Intelligence Models (LAIMs) demonstrate significant potential to handle complex scheduling scenarios and improve resource utilization efficiency. This paper investigates a task-oriented LAIM-driven resource scheduling mechanism, which constructs a multidimensional scheduling decision model by integrating task semantics, network states, and constraint conditions. Furthermore, a task-oriented prompt generation method is designed to establish a deep association between task requirements and network state. In the proposed resource allocation scheme, an external evaluation and feedback module is incorporated to conduct real-time feasibility verification and performance evaluation of scheduling strategies, thus enhancing the robustness and adaptability of scheduling. Simulation results demonstrate that the proposed Large Language Model (LLM)-driven network architecture and resource allocation scheme achieve significant improvements in convergence speed, processing latency, and energy consumption, effectively enhancing IoE task responsiveness and resource utilization.

The rapid development of the Internet of Everything (IoE) is accelerating the adoption of intelligent applications.

Cellular networks are integrating Artificial Intelli- gence (AI) into radio access network control. The MAC scheduler is a promising target because it allocates a limited resource, spectrum, at every slot, under competing latency, throughput, and reliability requirements. However, most learning-based sched- ulers are evaluated only in simulation. Production schedulers are difficult to modify, and realistic stress tests require more radio hardware than most laboratories can provide. We present MAC-Gyver, an open-source framework for developing and evaluating scheduling applications that execute directly inside the OpenAirInterface scheduler. It exposes scheduler observations and controls through typed interfaces while preserving the underlying protocol and real-time execution paths. The same applications run over the air and in mac-emu, a PHY-less emulator that executes the unmodified OpenAirInterface Layer 2 stack for up to 90 users on one host at real-time slot pace, with a 3GPP-compliant channel model. To showcase the flexibility of MAC-Gyver, we evaluate two use cases. A proactive uplink scheduler predicts packet arrivals and roughly halves median round-trip latency. A frequency-selective uplink scheduler selects contiguous sub-bands from per-PRB sounding observations and is evaluated across mobility and power-limited operating points against an offline scheduling ceiling. Together, they show how the same production stack can be an AI playground that supports implementation, controlled evaluation, and over-the-air validation through complementary scheduling use cases.

Cellular networks are integrating Artificial Intelli- gence (AI) into radio access network control.

Mobile network operators are increasingly exploring the use of artificial intelligence (AI) to automate complex network tasks, such as cell selection and mobility management. A fundamental problem arises: there is currently no way to verify that an AI function is making the right decisions or for the right reasons, rather than arriving at correct-looking answers through unreliable shortcuts. In safety-critical and resilience-focused infrastructure, this lack of transparency poses a significant challenge to the widespread adoption of AI technologies in wireless networks. In this paper, we propose a mechanical auditing approach: inspecting a function's internal representations and checking them against machine-verifiable 3GPP specifications. Specifically, we set out a general three-step auditing principle that locates protocol-relevant features, verifies their causal role, and diagnoses how adaptation reshapes their use, grounding it throughout publicly available interpretability and telecommunications research. We present an audit-native network architecture in which a dedicated verification agent continuously checks the reasoning of AI functions in networks, supporting both predeployment certification and runtime auditing. We also discuss how it could be realised, the data and benchmarks, as well as the open challenges that remain before mechanistic auditing can enter telecommunications practice and standardisation.

Mobile network operators are increasingly exploring the use of artificial intelligence (AI) to automate complex network tasks, such as cell selection and mobility management.

Mixture of Experts (MoE) architectures have become key to large language models; however, their typical round-robin (RR) scheduling introduces significant bottlenecks. In this paper, we demonstrate that RR causes a previously-undiscovered exponential incast phenomenon with MoE traffic. We propose an alternative proactive fair scheduling framework tailored for MoE workloads, which effectively prevents fabric oversubscription. We also outline how it can be implemented in NICs. Finally, through extensive simulations with real and synthetic workloads, we demonstrate that this framework consistently eliminates incast, maintains a near-100% link utilization, and reduces Collective Completion Time (CCT).

Mixture of Experts (MoE) architectures have become key to large language models; however, their typical round-robin (RR) scheduling introduces significant bottlenecks.

Autonomous Network Levels 4-5 require AI agents to invoke tools across vendor boundaries without human oversight, yet existing management standards lack a standardized mechanism for cross-vendor trust visibility. When a tool from Vendor B is compromised, agents from Vendor A continue invoking it -- unaware of the trust degradation -- causing cascading service impact. We present AgentToolMO, a proposed 3GPP NRM information model for agent tool trust management. The model comprises: a formally defined trust state machine with provable graduated enforcement, damped cascade propagation with bounded convergence, cross-vendor trust notifications via existing Management Services (MnS) interfaces, and retroactive impact assessment through NRM dependency graph traversal. Simulation-based evaluation across multi-vendor topologies shows that standardized cross-vendor notifications reduce blast radius from hours-scale undetected propagation to near-real-time containment bounded by MnS notification delivery, with cascade convergence guaranteed in bounded iterations and sub-linear notification scaling across vendor domains. The framework operates within existing 3GPP management infrastructure, leverages existing protocols, and provides a standardization pathway for trustworthy multi-vendor autonomous network management.

Autonomous Network Levels 4-5 require AI agents to invoke tools across vendor boundaries without human oversight, yet existing management standards lack a standardized mechanism for cross-vendor trust visibility.

The shift to O-RAN architectures marks a turning point in cellular security, where increased openness and modularity directly translate into a broader attack surface. Among the security threats cataloged by the O-RAN Alliance Working Group 11, performance-degradation attacks constitute the largest class. These attacks induce packet losses and latency spikes that are hard to distinguish from operational events such as misconfigurations, transient congestion, or software regressions. Consequently, upon an adverse incident detection, support engineers must rapidly determine whether to route the corresponding incident ticket to network maintenance or escalate it to security operations. This triage phase represents a critical human-in-the-loop bottleneck in the incident response lifecycle. To address this vulnerability, we introduce C-RE-ACT (Causal RE-ACTing agent), an automated agentic triage framework designed to generate actionable incident reports. C-RE-ACT starts constructing a Weighted Directed Acyclic Graph (WDAG) over O-RAN metrics using the Structural Agnostic Model (SAM). The resulting causal topology is encoded into a continuous soft token via a Graph Isomorphism Network (GIN) aligned with the language space of the Large Language Model (LLM) powering a ReAct agent. We evaluate C-RE-ACT on a physical, O-RAN-compliant testbed across 140 distinct performance-degradation experiments. Empirical results demonstrate the causal ranking isolates the correct root cause within the top three candidates in 89% of instances. Furthermore, graph soft-prompting improves LLM accuracy on causal-topology queries from 0.22 (text-only baseline) to 0.72. The autonomous agent achieves anomaly classification accuracies of 83% for delay anomalies and 84% for packet-loss anomalies.

The shift to O-RAN architectures marks a turning point in cellular security, where increased openness and modularity directly translate into a broader attack surface.

Full-duplex omni models are transforming human--AI interaction from turn-based exchanges into continuous multimodal conversations in which speaking, listening, and reasoning unfold concurrently. Rather than viewing the model as a replacement for a human endpoint, we argue for a new perspective: the model is a stateful computational middlebox inside a human-centered feedback loop, with network transport, model serving, and user playback jointly shaping how the interaction evolves. This perspective breaks the traditional boundaries among stages designed around local objectives. Rather than optimizing them in isolation, an AI-native real-time stack should allow the state of each stage to shape the actions of the others. We explore three cross-stage coordination opportunities: network-aware inference scheduling, execution-aware transport prioritization, and playback control that accounts for both network and model variability. We are building Conflux to explore these ideas, and preliminary results show substantial improvements in response latency and playback deadline adherence under network degradation. More broadly, we call for an AI-native real-time communication stack that resolve the joint control problem spanning communication, computation, and playback.

Full-duplex omni models are transforming human--AI interaction from turn-based exchanges into continuous multimodal conversations in which speaking, listening, and reasoning unfold concurrently.

Translating user intent into physical radio signals is the last critical step in wireless prototyping. It chains protocol planning, baseband synthesis, and hardware configuration. Large language models and multi-agent systems have reshaped software engineering, raising the question of whether they can solve this problem. Yet current models fail at this task, even when augmented with domain tools. Because the stages run sequentially, an error at any stage propagates downstream, so the end-to-end success rate collapses toward zero even when each stage looks locally competent. We introduce RadioMaster, a fully autonomous multi-agent framework that drives user input to verified emissions transmitted over the air. It rests on three synergistic pillars. RadioWiki grounds generation in domain knowledge to suppress hallucination. RadioAgent decomposes the fragile pipeline into independently executable and locally recoverable stages. RadioEmulator gates deployment behind closed-loop physical-layer verification. We further build RadioBench, the first benchmark for autonomous radio signal generation. Extensive real-world evaluations show that RadioMaster substantially outperforms state-of-the-art baselines in configuration viability and signal fidelity, while reducing configuration time by up to 28x.

Translating user intent into physical radio signals is the last critical step in wireless prototyping.

In graph theory and its applications to networking, such as telecommunications or transportation, path-finding is a central problem. While single-path algorithms are well established, methods for handling sets of multiple paths are less developed. A companion paper introduced a formal model for defining attributes over sets of paths based on their structural properties; this paper addresses that model's practical implementation. We present an optimized algorithm for computing cut sets of a path set, a nontrivial task that can be infeasible without efficient methods, and validate its performance via systematic benchmarks on network simulations of varying complexity. Additionally, we introduce a vectorized computational framework that expresses property calculations as matrix operations, enabling concise implementations in array-oriented languages. Together, these contributions establish practical foundations for the companion model, demonstrating that its implementation is both feasible and characterized by predictable, acceptable execution times.

In graph theory and its applications to networking, such as telecommunications or transportation, path-finding is a central problem.

Quantum networks can enable distributed computing and sensing. To realize these capabilities securely, quantum identity authentication is essential. Without authentication at the quantum layer, malicious repeaters may retain entanglement instead of performing swapping, enabling man-in-the-middle attacks (MitM) between communicating parties. Authentication mitigates this threat by embedding authentication qubits within data qubits at positions and bases based on a secret key shared a priori. While prior work analyzes security and MitM detection guarantees, physical layer side channel analysis remains unexplored. If an attacker infers protocol stages, it can avoid authentication qubits and extract data qubits, rendering authentication ineffective. To this end, we carry out experimental studies using a quantum communication testbed. A beam splitter is used to tap a portion of the optical signal, allowing the observer to collect side channel data without disrupting the quantum state. We evaluate two sampling settings, where 30% or 10% of the signal is diverted. The collected side channel data includes photon arrival timing and optical power data obtained using a single-photon detector and a power meter. Using this dataset, we extract and engineer features that capture both timing dynamics and signal intensity variations. We then train machine learning models to classify protocol stages based solely on side channel observations. Our results show that protocol-stage inference is feasible with high accuracy, reaching 98% (F1-score 97%) at 30% sampling and 96% (F1-score 94%) at 10% sampling. These findings reveal an overlooked vulnerability and highlight the need for robust designs against side channel inference attacks.

Quantum networks can enable distributed computing and sensing.

Quantum communication is a key enabler of next-generation networks, leveraging quantum entanglement to enable a new class of information exchange. While prior work has focused on the theoretical analysis of communication protocols, their exposure to physical layer side channel analysis remains largely unexplored. In classical systems, side channel analysis has been shown to reveal sensitive information without accessing the underlying data, raising the question of whether similar risks exist in quantum networks. In this work, we investigate whether different quantum communication protocols exhibit distinguishable signatures that can be inferred through passive side channel observations. We consider a threat model in which an observer accesses only a fraction of the optical signal without directly measuring the encoded quantum states. Under this setting, we experimentally examine four representative protocols, namely entanglement distribution, quantum gate sequences, heralded quantum key distribution, and quantum identity authentication, realized on a polarization entangled photon link. Observable physical layer features, including single photon detection statistics and optical power measurements, are collected and used to construct protocol fingerprints. We develop a data-driven framework for protocol identification based on these observations. Our results show that protocol identity can be inferred with accuracy reaching up to 96% under 30:70 sampling configuration/optical tapping, while remaining distinguishable at 10:90 with accuracy ranging from 70-89%. Bell inequality measurements confirm that the sampling/tapping process preserves entanglement, validating the non-destructive nature of the observation model. These findings demonstrate that side channel analysis can expose protocol-level information without disrupting quantum correlations, introducing new security considerations.

Quantum communication is a key enabler of next-generation networks, leveraging quantum entanglement to enable a new class of information exchange.

Unmanned Aerial Vehicle (UAV)-assisted networks are increasingly foreseen as a promising approach for emergency response, providing rapid, flexible, and resilient communications in environments where terrestrial infrastructure is degraded or unavailable. In such scenarios, voice radio communications remain essential for first responders due to their robustness; however, their unstructured nature prevents direct integration with automated UAV-assisted network management. This paper proposes SIREN, an AI-driven framework that enables voice-driven perception for UAV-assisted networks. By integrating Automatic Speech Recognition (ASR) with Large Language Model (LLM)-based semantic extraction and Natural Language Processing (NLP) validation, SIREN converts emergency voice traffic into structured, machine-readable information, including responding units, location references, emergency severity, and Quality-of-Service (QoS) requirements. SIREN is evaluated using synthetic emergency scenarios with controlled variations in language, speaker count, background noise, and message complexity. The results demonstrate robust transcription and reliable semantic extraction across diverse operating conditions, while highlighting speaker diarization and geographic ambiguity as the main limiting factors. These findings establish the feasibility of voice-driven situational awareness for UAV-assisted networks and show a practical foundation for human-in-the-loop decision support and adaptive network management in emergency response operations.

Unmanned Aerial Vehicle (UAV)-assisted networks are increasingly foreseen as a promising approach for emergency response, providing rapid, flexible, and resilient communications in environments where terrestrial infrastructure is degraded or unavailable.

The rapid growth of UAV applications necessitates a robust communication and networking system architecture capable of addressing the diverse requirements of various applications concurrently, rather than relying on application-specific solutions. This paper proposes a generic and reliable multi-UAV communication and networking system architecture designed to support the varying demands of heterogeneous applications, including short-range and long-range communication, star and mesh topologies, different data rates, and multiple wireless standards. Our architecture is designed for both ad hoc and infrastructure networks, ensuring seamless connectivity throughout the network. Additionally, we present the design of a multi-protocol UAV gateway that enables interoperability among various communication protocols to enhance connectivity. Furthermore, we introduce a data processing and service layer framework with a graphical user interface of a ground control station that facilitates remote control and monitoring from any location at any time. We practically implemented the proposed architecture and evaluated its performance using different metrics, demonstrating its effectiveness.

The rapid growth of UAV applications necessitates a robust communication and networking system architecture capable of addressing the diverse requirements of various applications concurrently, rather than relying on application-specific solutions.

Low Latency, Low Loss, and Scalable Throughput (L4S) is an emerging paradigm for latency control based on DualPI2 active queue management and scalable congestion control. While a Linux kernel implementation of DualPI2 is available, controlled and reproducible experimentation on L4S mechanisms can be facilitated by a modular, user-space alternative. In this paper, we present a DualPI2 module for the Mahimahi network emulator, designed to support extensible, component-level experimentation without kernel modification. We conduct a statistical behavioral characterization of the Mahimahi implementation by examining key metrics across diverse traffic patterns and network conditions, using the Linux kernel implementation as a reference baseline. Our analysis shows that behavioral alignment across execution environments is not automatic: identical DualPI2 parameterization does not guarantee identical dynamics. Instead, key control parameters exhibit environment-dependent sensitivity, leading to regime-dependent discrepancies across bandwidth-delay product (BDP) conditions. Through targeted parameter exploration, we identify configurations that improve cross-platform alignment in low BDP regimes, while revealing structural differences that persist under higher load. This work provides both a practical tool for experimental L4S research and empirical insight into cross-platform behavioral differences, highlighting the importance of systematic characterization and environment-aware parameter selection in emulation-based AQM studies.

Low Latency, Low Loss, and Scalable Throughput (L4S) is an emerging paradigm for latency control based on DualPI2 active queue management and scalable congestion control.

The growing interest in Low Latency, Low Loss, and Scalable Throughput (L4S) reflects the need for lower latency in interactive multimedia applications. In this paper, we use an open-source DualPI2 implementation over the Mahimahi emulator to evaluate the impact of L4S on SCReAM congestion controlled video traffic. To do so, we augment the SCReAM BW tool with a video codec, enabling the generation of video traffic in addition to its original synthetic RTP mode. We evaluate both network-level and Quality of Experience (QoE) metrics on a mobile network trace, under random packet loss, and with different motion-complexity levels. In our baseline scenario, L4S reduces the median per-run $95^{th}$ percentile queue delay by 35%, at the cost of a 42% drop in sender throughput. Under 1% packet loss, L4S yields more pronounced QoE gains compared to the lossless scenario, despite narrower network-level benefits. Across video content complexities, L4S also maintains more stable QoE than Classic. These results underscore the importance of evaluating QoE alongside network-level metrics when assessing the effect of L4S on end-user application performance.

The growing interest in Low Latency, Low Loss, and Scalable Throughput (L4S) reflects the need for lower latency in interactive multimedia applications.

A formal model enables verifying reachability, localizing an outage, or anticipating the blast radius of a change. Yet, virtually no production network has one, since writing a model by hand demands rare expertise and is hard to keep current as the network changes frequently. At its core, network modeling is a typographical exercise: it translates network artifacts (e.g., configurations, topology, and routing state) into rules in formal logic. Translation of this kind is what large language models (LLMs) nowadays do well. Unlike free-form AI reasoning, such translation can be formally verified. Once modeling is no longer the bottleneck, trusting AI to reason over large, complex networks no longer makes sense. Our position therefore cuts against the prevailing race to put autonomous AI agents in charge end-to-end. We instead confine AI to translation and rely on a solver for reliable long-horizon reasoning, building a reusable formal model of general network behavior that can then be specialized to specific tasks, e.g., root-cause analysis (RCA). We build TypoNet that constructs and validates a symbolic model of an emulated production-scale WAN from the network's own artifacts. Our preliminary evaluation shows TypoNet helps in two ways. On its own, TypoNet answers operational questions (e.g., reachability verification and change-impact analysis) faster, more cheaply, and more reliably than an LLM. As a tool for an AI agent, TypoNet boosts fault localization at lower cost. The result makes the case for AI that builds verifiable network models and relies on a solver for reliable long-horizon reasoning.

A formal model enables verifying reachability, localizing an outage, or anticipating the blast radius of a change.

Invariants, the relations expected to hold among measured signals of a network, underpin applications from verification to traffic generation, telemetry imputation, and input validation, yet writing them by hand demands rare expertise in both formal logic and networking. Automatic miners can help but fall short on two fronts: they still require the hardest input (the grammar of admissible invariants) and they learn only exact, ``hard'' rules, struggling with real-world approximation caused by inherent noise in data. LLMs are tools that can provide semantic reasoning over data, but are non-deterministic and opaque in their learning. Our key idea is to partition the invariant search problem into an AI-driven grammar ``discovery'' problem, followed by a statistics-driven ``search'' problem within the learned grammar. Taken together, this allows non-deterministic, hallucination-prone AI to help produce auditable invariants with formal guarantees. We design and implement such a system, Autogram, and evaluate it on both public and production telemetry data, recovering expert-derived invariants with high coverage and low false positives. We close with discussion on open problems on the path toward fully open-ended discovery.

Invariants, the relations expected to hold among measured signals of a network, underpin applications from verification to traffic generation, telemetry imputation, and input validation, yet writing them by hand demands rare expertise in both formal logic and networking.

Data-center congestion control targets high throughput, fair bandwidth allocation, and low latency. Modern transports couple rate computation and packet scheduling into a single feedback loop, converging to near-optimal rates but leaving standing queues that can scale with the number of flows. We argue that separating the two reveals a simpler design point. Given stable feasible rates, the residual queue problem reduces to a timing problem: if every flow's packets arrive at the bottleneck in the correct slot, the link stays busy and the queue stays empty. Clocked ACK-Paced Synchronization CAPS is a lightweight distributed scheduling layer that achieves this by phase-locking each sender's transmissions to ACK-clocked bottleneck slots, with a per-flow correction that compensates for heterogeneous RTTs. We characterize the phase-locked steady state for dumbbell topologies under equal RTT, heterogeneous RTT, and bidirectional traffic, and validate the mechanism on a fat-tree under incast, permutation, and all-to-all traffic. CAPS reduces worst-case queue occupancy by 5-10x across all tested scenarios without throughput loss.

Data-center congestion control targets high throughput, fair bandwidth allocation, and low latency.

Large Language Models (LLMs) are increasingly deployed as autonomous agents, transitioning from static conversational interfaces to dynamic systems capable of complex reasoning, tool execution, and decision-making. However, the operational reliability of these agentic AI systems is fundamentally challenged by the absence of reliable ground truth in open-ended environments and the risk of increasing operational drift over time. To address this challenge, we propose and experimentally evaluate an agentic AI framework, designed to enforce autonomous integrity within LLM-driven systems. We design a self-calibration mechanism that mitigates drift and dynamically approximates ground truth by incorporating an ARIMA forecaster, without requiring continuous human oversight. To demonstrate the effectiveness and reliability of our methodology, we apply it to the complex domain of profiling the resource usage of zero-knowledge workloads in edge computing networks. Experimental results show that the proposed self-calibrating agentic framework successfully profiles the zero-knowledge workloads, achieving a higher accuracy than baseline LLM agents by 91.7% for resource usage prediction and improving the prediction speed by 71.7% compared to pure profiling, establishing a robust foundation for deploying autonomous AI in decentralized infrastructures. Furthermore, the ground truth generation using the proposed ARIMA leaping algorithm is 52% faster than a standard ARIMA forecasting algorithm, while achieving the same accuracy.

Large Language Models (LLMs) are increasingly deployed as autonomous agents, transitioning from static conversational interfaces to dynamic systems capable of complex reasoning, tool execution, and decision-making.

The emergence of 6G-enabled vehicular metaverses enables Autonomous Vehicles (AVs) to operate across physical and virtual spaces through space-air-ground-sea integrated networks. The AVs can deploy AI agents powered by large AI models as personalized assistants, on edge servers to support intelligent driving decision making and enhanced on-board experiences. However, such cross-reality interactions may cause serious location privacy risks, as adversaries can infer AV trajectories by correlating the location reported when AVs request LBS in reality with the location of the edge servers on which their corresponding AI agents are deployed in virtuality. To address this challenge, we design a cross-reality location privacy protection framework based on hybrid actions, including continuous location perturbation in reality and discrete privacy-aware AI agent migration in virtuality. In this framework, a new privacy metric, termed cross-reality location entropy, is proposed to effectively quantify the privacy levels of AVs. Based on this metric, we formulate an optimization problem to optimize the hybrid action, focusing on achieving a balance between location protection, service latency reduction, and quality of service maintenance. To solve the complex mixed-integer problem, we develop a novel LLM-enhanced Hybrid Diffusion Proximal Policy Optimization (LHDPPO) algorithm, which integrates LLM-driven informative reward design to enhance environment understanding with double Generative Diffusion Models-based policy exploration to handle high-dimensional action spaces, thereby enabling reliable determination of optimal hybrid actions. Extensive experiments on real-world datasets demonstrate that the proposed framework effectively mitigates cross-reality location privacy leakage for AVs while maintaining strong user immersion within 6G-enabled vehicular metaverse scenarios.

The emergence of 6G-enabled vehicular metaverses enables Autonomous Vehicles (AVs) to operate across physical and virtual spaces through space-air-ground-sea integrated networks.

Counting Wi-Fi devices within access point (AP) coverage by listening to their probe request frames (PRFs) is a well-established research problem, fundamental to many Internet of Things (IoT) applications such as crowd management and public transportation scheduling. While commendable counting accuracy has been reported, existing approaches fall short in deployment convenience due to their reliance on machine learning, which necessitates 1) extensive data collection and training efforts for system setup, and 2) specialized model fine-tuning for operational maintenance. We propose RateCount, an accurate, lightweight, and learning-free counting approach to lower deployment costs. RateCount employs a provably unbiased closed-form expression to estimate the device count based on the rate at which APs receive PRFs, along with an error model to compute the estimation variance. We also demonstrate its application in people counting by incorporating a device-to-person calibration scheme. Through extensive real-world experiments conducted at multiple sites spanning a wide range of counts, we show that RateCount, without any deployment costs for machine learning, achieves comparable counting accuracy to the state-of-the-art (SOTA) learning-based device counting and improves previous people counting schemes by a large margin.

Counting Wi-Fi devices within access point (AP) coverage by listening to their probe request frames (PRFs) is a well-established research problem, fundamental to many Internet of Things (IoT) applications such as crowd management and public transportation scheduling.

Datacenter networks follow a multipath doctrine: provision many paths between endpoints, hash flows across them, and let redundancy absorb both failures and load imbalance. The ZCube topology violates this doctrine. It removes the Spine layer, eliminates path multiplicity, and cuts one third of switching hardware, yet delivers better performance for both large model training and inference. We explain this anomaly through a structural connection to Braess's paradox, first observed in 1968: both phenomena trace to congestion-oblivious routing over competing paths. Braess showed that adding paths under this condition can hurt; ZCube shows that removing paths under the same condition can help. First, we show that multipath fabrics under structured LLM traffic operate in Braess's shadow: static ECMP hashing is strictly more fragile than greedy routing. Greedy routing reaches an equilibrium within 4/3 of optimal for affine latencies; static hashing admits unbounded imbalance in the worst case. Second, we prove that ZCube is immune to Braess's paradox and that its orthogonal dual partition provably balances load for arbitrary traffic matrices; AllReduce in training and KV cache transfers in disaggregated inference fall out as two corollaries. Third, we quantify the price of this immunity: ZCube trades microsecond hash recovery for millisecond control plane recovery, a trade that upper-layer resilience in LLM serving makes favorable. Production measurements from a cluster serving GLM-5.1 coding inference report 33% lower network cost, 15% higher GPU throughput, and 40.6% lower P99 time to first token. Our analysis suggests that for workloads driven by model structure, matching topology to traffic matters more than path multiplicity.

Datacenter networks follow a multipath doctrine: provision many paths between endpoints, hash flows across them, and let redundancy absorb both failures and load imbalance.

Rural areas exhibit low population density and highly variable connectivity needs shaped by both household usage and field operations such as planting, harvesting, and mining. These field activities often occur in isolated locations requiring temporary connectivity, whereas rural households depend on fixed broadband. During intensive outdoor activities, household fixed networks may remain underutilized, resulting in inefficient resource use and unnecessary energy consumption. The coexistence of residential and field-based communication demands creates substantial spatial and temporal fluctuations that the current rural network cannot effectively adapt to. Limited visibility into user mobility, activity patterns, and intent makes it difficult for operators to coordinate temporary and fixed networks. To address these underexplored challenges, we propose an AI driven Intent Aware Satellite Integrated Access and Backhaul (IAB) approach to connect rural areas. In our proposal, a large language model (LLM) translates users' intents into explicit network requirements. Guided by these inferred requirements, we develop a dynamic satellite IAB based Fixed Wireless Access (FWA) network approach that jointly optimizes temporary field connectivity and fixed broadband access to maximize energy efficiency while satisfying the data rate requirement. The formulated optimization problem is solved using a two stage Benders decomposition approach. The simulation results show that our approach significantly reduces energy consumption while maximizing energy efficiency.

Rural areas exhibit low population density and highly variable connectivity needs shaped by both household usage and field operations such as planting, harvesting, and mining.

Federated parameter-efficient fine-tuning (PEFT) enables communication-efficient adaptation of large pretrained models on decentralized edge data, but it remains fragile under non-IID client heterogeneity. In low-rank adaptation (LoRA), different clients may learn locally useful but spectrally misaligned update subspaces, causing high-variance aggregation and poor global transfer. We propose TRISHUL, a spectral-control framework for robust federated PEFT. TRISHUL follows the FL no-raw-data-sharing setting but does not itself provide formal privacy guarantees. TRISHUL uses shared frozen multi-head low-rank bases to obtain algebraically exact aggregation of compact core updates, applies nuclear norm proximal shrinkage to suppress client-specific high-rank spectral components before upload, and allocates adaptation heads non-uniformly across layers using a concave water filling budget rule derived from pretrained layer capacity. Because shrinkage is performed only on small core matrices, TRISHUL adds negligible computation and no extra per-round communication over the underlying multi-head PEFT protocol. Across vision and language benchmarks, including CIFAR-100, SVHN, 20 Newsgroups, MRQA, and GLUE with LLaMA3.2-1B, TRISHUL improves convergence, stability, and final performance over federated LoRA baselines, with greater gains under stronger heterogeneity.

Federated parameter-efficient fine-tuning (PEFT) enables communication-efficient adaptation of large pretrained models on decentralized edge data, but it remains fragile under non-IID client heterogeneity.