Pairwise evaluation of large language models (LLMs) has become the dominant paradigm for benchmarking open-ended tasks, yet non-transitive preferences—where evaluators prefer A over B, B over C, but C over A—fundamentally undermine ranking reliability. We show that this critical issue stems largely from low-quality data that contains inherently ambiguous preference pairs. To address this challenge, we propose ELSPR, a principled graph-theoretic framework that models pairwise preferences as tournament graphs and systematically identifies problematic training data. ELSPR quantifies non-transitivity through strongly connected components (SCCs) analysis and measures overall preference clarity using a novel normalized directed graph structural entropy metric. Our filtering methodology selectively removes preference data that induce non-transitivity while preserving transitive preferences. Extensive experiments on the AlpacaEval benchmark demonstrate that models fine-tuned on ELSPR-filtered data achieve substantial improvements: a 13.8% reduction in non-transitivity, a 0.088 decrease in structural entropy, and significantly enhanced discriminative power in real-world evaluation systems. Human validation confirms that discarded data exhibit dramatically lower inter-annotator agreement (34.4% vs. 52.6%) and model-human consistency (51.2% vs. 80.6%) compared to cleaned data. These findings establish ELSPR as an effective data self-purification approach for developing more robust, consistent, and human-aligned LLM evaluation systems.

Network Function Virtualization (NFV) facilitates on-demand and flexible service provisioning to meet the escalating demands of Internet of Things (IoT) applications, enabled by Service Function Chain (SFC) technique. The widespread deployment of 5G has connected a massive number of devices and users to IoT networks, accelerating the expansion of IoT scales. IoT users’ service requirements exhibit heightened diversity and dynamism. Consequently, the SFC placement problem in Next Generation Multi-domain IoT (NGMIoT) networks has garnered significant attention. How to efficiently place SFCs under uncertain resource demands to adapt to evolving service request dynamics poses substantial challenges. Therefore, this paper investigates the Robust SFC Placement (RSFCP) problem in NGMIoT networks under resource demand uncertainty. Specifically, we formulate the RSFCP problem as an integer linear programming model to minimize overall SFC placement cost while ensuring service quality. We further prove the RSFCP problem is NP-hard and propose a greedy strategy based heuristic SFC placement algorithm to solve it. Finally, extensive simulation experiments are conducted to evaluate performance, demonstrating that the proposed algorithm outperforms benchmark mechanisms in terms of service acceptance rate and placement cost.

Network measurement sketches enable efficient traffic monitoring but require careful parameter configuration to balance accuracy and memory efficiency. We presentRA-Sketch, a unified framework for generating memory-optimal sketch configurations that satisfy user-defined error constraints across diverse network measurement tasks. Unlike existing approaches that rely on computationally intensive experimental testing, RA-Sketch introduces: 1)Poisson-distributed collision modelingto construct error predictors for both frequency-independent tasks (membership query, heavy-hitter detection, and super-spreader detection) and frequency-dependent tasks (flow size distribution, frequency estimation, and cardinality estimation), eliminating the need for empirical validation; 2) Ahierarchical search strategycombining power-of-two scaling and binary search, reducing iterations through optimized parameter initialization. RA-Sketch supports 10+ sketch architectures including Bloom Filter, Elastic Sketch, HeavyKeeper, MEC Sketch, MRAC, CM Sketch, CO Sketch, gSkt, rSkt1 among others. Evaluations on real-world network traces demonstrate: 1) up to 6–7 orders-of-magnitude faster configuration than benchmark-based methods; 2) Prediction errors are within 10% for heavy-hitter detection and super-spreader detection in most evaluated settings, while prediction errors for membership query, flow size distribution, frequency estimation, and cardinality estimation are close to zero; 3) Memory utilization approaches theoretical minima. The framework’sgenerality and efficiency enable real-time reconfiguration of sketches under dynamic network conditions.

Mobile devices’ multiple network interfaces create the foundation for multipath optimization, yet widespread implementation remains challenging. Existing multipath solutions primarily rely on simulation environments or require hardware/software adaptations, limiting real-world adoption. Meanwhile, 89.8% of mobile users prefer WiFi over cellular networks due to data cost concerns, despite cellular networks often providing better stability. We propose WiCell, an application-transparent multipath enhancement system for mobile networks that addresses these challenges through: (1) a deployment-friendly approach requiring no hardware or software modifications; (2) a tunneling-based architecture with edge proxy servers that dynamically adjusts transmission based on real-time link condition feedback; and (3) two operational modes tailored to different user requirements - WiFi-Primary Adaptive Redundancy Mode for balanced performance and data conservation, and Dual Transmission Mode for maximum connection stability. Our comprehensive evaluation demonstrates WiCell’s effectiveness across 30 common applications. In WiFi-Primary Adaptive Redundancy Mode, WiCell achieves a 74.6% reduction in video stalling rate while consuming only 22.12% of the cellular data compared to full redundancy approaches. In Dual Transmission Mode, WiCell reduces maximum latency by 77.4% compared to WiFi-only transmission, with 61.36% lower average latency than WiFi and 65.84% lower than cellular-only connections. These results validate WiCell as a practical solution for enhancing wireless network performance while maintaining application transparency and addressing user concerns about cellular data consumption.

As network size continues to increase and data flows between LANs become more frequent, in-band network telemetry methods across LANs require more balanced path lengths and the privacy of telemetry data. Existing schemes have large variance in telemetry path lengths and do not focus on the riskiness of cross-LAN telemetry. In this paper, we design SRv6-ALINT to direct telemetry paths through SRv6 and upload telemetry data at boundary nodes. In the small-scale network case, SRv6-ALINT solve the path generation strategy to get the theoretical optimal value of the path length variance through a solver. In the case of larger scale networks, SRv6-ALINT propose DFS-stitch, an algorithm that guarantees that the path length variance is as small as possible when the number of paths generated is minimized with no duplicate edges and covering the entire network. For segment list compression in SRv6, SRv6-ALINT propose sequential and binary compression algorithms to compress the obtained paths, reduce the segment list depth and decrease the bandwidth. The evaluation shows that the path length variance generated by DFS-stitch is smaller compared to existing schemes. The compressed segment list length of both segment list compression algorithms is reduced to approximately 50% of the uncompressed length.

Network measurement sketches enable efficient traffic monitoring but require careful parameter configuration to balance accuracy and memory efficiency. We present RA-Sketch, a framework for generating memory-optimal sketch configurations that satisfy user-defined error constraints across diverse network measurement tasks. Unlike existing approaches that rely on computationally intensive experimental testing, RA-Sketch introduces: 1) Poisson-distributed collision modeling to construct error predictors for both frequency-independent tasks (membership query, heavy-hitter detection) and frequency-dependent tasks (frequency/cardinality estimation), eliminating the need for empirical validation; 2) A hierarchical search strategy combining power-of-two scaling and binary search, reducing iterations through optimized parameter initialization. RA-Sketch supports 10+ sketch architectures including Bloom Filter, Elastic Sketch, HeavyGuardian, HeavyKeeper, CM/CO Sketch, gSkt, rSkt1 and so on. Evaluations on real-world network traces demonstrate: 1) 6–7 orders of magnitude faster configuration than benchmark-based methods; 2) Prediction errors ≤10% for heavy-hitter detection, while prediction errors for membership query, and frequency/cardinality estimation are close to zero; 3) Memory utilization approaches theoretical minima. The framework’s generality and efficiency enable real-time reconfiguration of sketches under dynamic network conditions.

Per-Flow cardinality measurement in high-speed networks is essential for network security and traffic analysis applications. Flow cardinality refers to the number of distinct elements within a flow, such as the number of unique destination IPs associated with a given source IP. While extensive research has been conducted on single-flow cardinality estimation, achieving accurate per-flow cardinality measurement with real-time performance and low memory overhead remains challenging in large-scale network environments, particularly given the highly skewed distribution of flow cardinalities where mouse flows with smaller cardinalities dominate, and elephant flows with larger cardinalities are fewer. This paper introduces MEC-Sketch, a memory-efficient cardinality estimation data structure that leverages the inherently skewed distribution of flow cardinalities in network traffic. MEC-Sketch employs a dual-component architecture: a heavy part utilizing a majority vote algorithm for precise super-spreader detection, and a light part implementing compact cardinality estimators for memory-efficient measurement of mouse flows. We address two fundamental technical challenges: (1) adapting the majority vote algorithms to operate with cardinality estimators that lack native support for real-time queries, and (2) implementing an effective mapping strategy between large estimators in the heavy part and small estimators in the light part during elephant-mouse flow separation. Comprehensive evaluations on real-world network traces demonstrate that MEC-Sketch significantly outperforms state-of-the-art solutions in terms of estimation accuracy, memory efficiency, and computational performance for both cardinality estimation and super-spreader detection tasks.

State-of-the-art deep learning models rely on large GPU clusters and various parallelism strategies, which in turn depend on collective communication (CC) operators to synchronize data. While vendor libraries (e.g., NCCL, RCCL) provide standard CC algorithms, they often suffer from bandwidth bottlenecks in imbalanced topologies. Recent synthesis-based methods improve performance but face three key limitations: poor scalability due to the combinatorial explosion of scheduling space, lack of support for multistage execution, and suboptimal communication throughput. We propose Canvas, a scalable and near-optimal CC scheduling framework that addresses these challenges. Canvas introduces: (1) Hierarchical synthesis to decompose the global scheduling problem into tractable subproblems for scalability. (2) Collective decomposition to enable structured, multi-stage algorithm generation. (3) Cross-micro-batch pipeline scheduling to parallelize communication across micro-batches and maximize link utilization. Evaluations show that Canvas achieves up to 1.98× bandwidth speedup over TACCL and 3.56× over TE-CCL, and synthesizes algorithms for 512-GPU topologies within 1.77 hours, whereas TACCL fails to produce results within 24 hours.

Modern distributed training systems face escalating communication bottlenecks as GPU clusters scale to accommodate large models. While collective communication libraries and automated synthesizers address algorithmic efficiency, they suffer from three critical limitations including labor-intensive manual intervention requirement, overreliance on predefined input optimization, and suboptimal isolated configuration optimization. To solve these problems, we present TuCCL, a systematic framework that co-optimizes communication algorithms and runtime parameters through three innovations: Topology-Aware Sketch Generation that automatically produces high-performance primitives, Hierarchical Configuration Optimization modeling nonlinear parameter-performance relationships, and Multi-phase Resource-Aware Configuration Optimization enabling joint configuration tuning with adaptive search space pruning. Evaluations demonstrate TuCCL’s superiority over state-of-the-art systems with 1.75x–11.49x bandwidth improvements for AllGather/AllReduce on NVIDIA V100/A100 clusters, 90.2% faster configuration search than grid methods, and 1.22x-2.52x end-to-end training speedups across diverse model scales.

Large Language Models (LLMs) have advanced rapidly, but their growing parameter scales and memory demands pose critical challenges for distributed training. Although GPU memory capacity improves steadily, model sizes expand much faster, causing frequent Out-of-Memory (OOM) errors and rising training costs. Existing memory optimization approaches, such as ZeRO-3 and offloading, alleviate per-GPU memory pressure but introduce excessive collective communication, limited computation–communication overlap, and degraded scalability. We present MAZ3, a distributed training framework that mitigates these limitations through three key techniques: (1) Collaborative CPU–GPU memory management, storing full parameters in CPU memory and broadcasting them within nodes to reduce global synchronization; (2) Fine-grained communication–computation overlap, aligning collective operations with model computation to hide latency; (3) Hierarchical aggregation operators, leveraging intra-node NVLink and inter-node NIC channels concurrently to minimize communication overhead. We implement MAZ3 on a multi-GPU cluster and evaluate it with large-scale models. Results show that MAZ3 reduces inter-node model communication (gradients and parameters) by 33%, improves training efficiency by 40.3%, and increases throughput by 67.9% compared with ZeRO-3. Moreover, MAZ3 retains the memory efficiency of ZeRO-3 while approaching the training efficiency and throughput of ZeRO-2 Offload, achieving a balanced trade-off between memory optimization and performance.

In-band Network Telemetry(INT) enables fine-grained telemetry by editing the packet header with the capability of programmable data plane to carry network status. However, INT could cause significant telemetry overhead without an effective system design. Existing measurement systems attempt to reduce this overhead by employing fixed-frequency INT sampling. Nonetheless, these methods lead to frequent measurements of network ports with large flows while neglecting ports with small flows for extended periods. In this paper, we introduce a lightweight passive telemetry system based on INT, called INTSelection. The core idea is to use a flow selection algorithm at the centralized controller so as to measure all active ports. Compared with the current method, INT-Selection reduces the bandwidth overhead by 58.2% and 2.7%.

Real-time network traffic classification plays a crucial role in ensuring Quality of Service and network security, and machine learning (ML) based methods achieve high classification accuracy but induce significant computational overhead. While SmartNIC solutions can offload classification tasks thus reducing CPU burdens, they still suffer from various limitations, manifested in limited computing capabilities, insufficiency in dynamic load handling and high latency from heterogeneous computing architectures. To solve these problems, we propose SmartTC, with three key designs: (1) SmartTC employs hardware-software co-design to optimize SmartNIC processing power, (2) SmartTC adopts a trafficaware dynamic batch submission strategy that adjusts submission policies based on real-time network load, and (3) SmartTC proposes parallel pipeline scheduling that ensures efficient task execution while minimizing communication overhead. Finally, we implement SmartTC on the BlueField-3 DPU and conduct extensive experiments for evaluations, and comparison results demonstrate that SmartTC significantly outperforms existing solutions. For example, it reduces average traffic classification time by up to$\mathbf{1 6. 8 \%}$under low loads and$\mathbf{9 0. 9 \%}$under high loads. Besides, SmartTC does not affect Bluefield-3 network services, and saves host CPU usage by at least two cores.

As network scale and complexity continue to increase, managing network configurations has become an increasingly challenging task. Existing configuration tools often depend on low-level, abstract intermediate representations, which require users to have substantial technical expertise. This reliance not only increases the learning curve but also heightens the risk of configuration errors. Recent advances in Large Language Models (LLMs) have demonstrated strong potential for automating tasks across various domains. However, their applications to network configuration generation remain limited due to several challenges, including hallucination, restricted context length, and insufficient adaptability to domain-specific requirements. To address these issues, we propose ConfAgent, an advanced network configuration generation system powered by a multi-model intelligent agent. ConfAgent comprises four key components: a conflict detector, an information extractor, a routing algorithm coder, and a formal synthesizer. These components collaborate to accurately interpret complex configuration intents, detect potential conflicts, and generate robust code and network configurations through intuitive natural language interactions. Extensive experiments conducted on the NetConfEval benchmark demonstrate that ConfAgent consistently outperforms existing state-of-the-art methods by margins ranging from 36 % to 100 %, particularly excelling in configuration tasks for large-scale network topologies.

While the P4-based programmable data plane supports flexible configuration regardless of specific hardware or protocols, potential software and hardware failures threaten its performance. However, existing tools fail to efficiently test and automatically repair faults, relying on time-consuming and error-prone manual manipulation. To address these challenges, we propose REP4, a system that automatically detects, localizes, and repairs bugs in programmable data planes. First, it specifies high-level intents to express comprehensive data plane functions and eliminate invalid paths in the control flow graph, thereby accelerating testing. Second, faulty P4 code and table rules are automatically located by employing taint tracking. Finally, REP4 updates the erroneous fragments through constraint-based repair and performs regression testing for validation. We deploy REP4 on real switches, and extensive experiments on open-source P4_16 programs show that REP4 saves nearly half of the test packets than existing approaches with 100 % function coverage and localizes different types of root causes in 55s in the switch.p4 without false positives, and automatically updates$\mathbf{P 4}$code and table rules in 1.5s, with all updates successfully passing regression testing.

The encapsulated route images provided by vendors hide detailed implementation codes of routing protocol, making it challenging to implement the same routing protocol on heterogeneous devices, as well as to locate and repair the configuration errors. While some works attempt to address the above issues, they fail in large-scale deployment, manifested in insufficiency in complex routing behaviors, lack of clear mapping between routing behavior and protocol standards, and difficulty in obtaining an open-source license. To solve all the problems, we propose RTB, a novel tool to mine the causality between network configurations and routing behaviors, in order to assist operations personnel in better pinpointing configuration errors. First, RTB generates test cases based on configuration features using a coverage-guided combinatorial testing algorithm. Then, RTB employs a differential network model to obtain differential datasets storing the correspondence between configurations and routing behaviors. Finally, RTB utilizes the differential datasets to create multilayer causal graphs, so as to reason about the implementation details and differences in routing protocols, which assists network operation and maintenance personnel in locating configuration errors. Experimental results show that RTB only needs 35.3 % test cases, and improves code coverage by 10 % over Metha. Moreover, RTB can mine the causality with 98 % accuracy, which is 5 % more efficient than existing tools.

Network device misconfigurations can lead to security vulnerabilities and operational failures. Existing validation methods primarily rely on program analysis or machine learning, which are often labor-intensive and data-sensitive. While Large Language Models (LLMs) excel in text analysis, their direct application to configuration validation faces challenges such as limited domain knowledge, hallucinations, and non-interpretability. To address these issues, we propose ConfigChecker, an automated network configuration validation system that integrates LLMassisted structured knowledge extraction with knowledge graphbased reasoning. ConfigChecker involves two key phases: (1) Knowledge Graph Construction - We construct three instructiontuning datasets (EV, PG, and RE) to fine-tune LLMs for structured command information extraction, including view transitions, parameter constraints, and inter-command dependencies. The extracted knowledge is stored in a knowledge graph, providing a foundation for configuration validation. (2) Configuration Validation - A graph-based command validator ensures syntax and semantic correctness, while knowledge graph reasoning and indexing mechanism are used to detect inter-command dependency violations. We evaluate six instruction-tuned LLMs for structured knowledge extraction, achieving up to 90% accuracy on the EV and RE datasets. Additionally, our system shows a 9%-20% improvement in validation accuracy compared to three mainstream large models. Furthermore, the average response time per configuration line remains below 15 milliseconds. Results demonstrate that ConfigChecker significantly enhances validation accuracy and efficiency, providing a novel integration of LLMs and knowledge graphs for automated configuration validation.

Network traffic measurement is critical for effective network management. Sketch has been proven to be a promising network traffic measurement solution. Considering the skewed distribution of network traffic, where low-frequency mouse flows dominate and high-frequency elephant flows are fewer, recent sketch-based solutions employ hierarchical designs to enhance memory efficiency and accuracy. However, these solutions inevitably introduce additional challenges, including increased memory access overhead, severe hash collisions between elephant and mouse flows, and limited adaptability to dynamic network environments. In this paper, we propose LA-Sketch, an adaptive level-aware data structure. First, LA-Sketch employs a level-aware classifier to intelligently map each flow to its corresponding level, thereby reducing memory access overhead caused by hierarchical designs and mitigating hash collisions between elephant and mouse flows. Second, we introduce an adaptive counter configuration method that dynamically adjusts the number of counters at each level according to diverse network traffic distributions, which theoretically minimizes overall hash collisions. Finally, to adapt to the continuously changing network traffic characteristics, we propose an adaptive online training method that enables LA-Sketch's classifier to maintain high performance using only sketch query values for training, avoiding the significant overhead of massive traffic data collection. Extensive evaluations on two real-world network traces across five measurement tasks demonstrate that LA-Sketch outperforms state-of-the-art hierarchical sketches.

Heterogeneous networks utilize devices from multiple vendors to enhance failure prevention, but also pose challenges in network device management. Network configuration translation, as a critical aspect of management, facilitates the upgrading of pairwise nodes that back up each other and device replacements across different vendors. Manually translating configuration requires a significant amount of time and demands that administrators have ample experience. To address this, we introduce the Network Configuration Translation MultiAgent System (NetMAS), designed to manage heterogeneous manuals and streamline configuration translation. By efficiently handling diverse manuals, NetMAS simplifies the translation of network configurations across different devices and vendors. NetMAS consists of three core components: a hierarchical manual retriever, a detachable network configuration translator, and a memory module. Leveraging the advanced function-calling capabilities of the Large Language Models, NetMAS eliminates the need for external training, significantly enhancing overall efficiency. Our experimental results demonstrate that NetMAS can effectively retrieves configuration manuals. The retrieval accuracy reaches 94.11% when considering the top-5 results. Furthermore, it supports manuals from various vendors and devices. When translating from Cisco to Huawei, H3C, and Juniper, NetMAS improves accuracy by an average of$25.48 \times$and$12.38 \times$, respectively, compared to directly using GPT-4o and Deepseek.

In communication scenarios where information is highly valuable and privacy protection is essential, such as internal networks of governments and companies, network security plays a crucial role. The emergence of programmable networks offers a new approach to achieving network security. This paper proposes an endogenous security method based on programmable networks and, for the first time, uses the programmable data plane programming language P4 to implement dynamic address hopping and key field obfuscation. This method conceals the real IP, MAC, and Port information of the data packet during the communication process, thereby increasing the difficulty of identifying and tracking the location and identity of the attacker. At the same time, the implementation of dynamic address hopping and key field obfuscation does not require communication with the control plane; instead, it is implemented within the data plane, thereby saving on communication delay and memory usage. Through simulation experiments, the additional communication delay and CPU utilization introduced by the PNEST were evaluated. The results showed that the communication delay increased to approximately 8 ms and the CPU utilization to approximately 5% during the data packet transmission process, demonstrating the feasibility and practicality of the PNEST.

As distributed deep learning training (DLT) systems scale, collective communication has become a significant performance bottleneck. While current approaches optimize bandwidth utilization and task completion time, existing communication libraries (CCLs) backends fail to efficiently manage GPU resources during algorithm execution, limiting the performance of advanced algorithms. This paper proposes ResCCL, a novel CCL backend designed for Resource-Efficient Scheduling to address key limitations in current systems. ResCCL enhances execution efficiency by optimizing scheduling at the primitive level (e.g., send and recvReduceCopy), enabling flexible thread block (TB) allocation, and generating lightweight communication kernels to minimize runtime overhead. Our approach tackles the global scheduling problem, reduces idle TB resources, and enhances communication bandwidth. Evaluation results demonstrate that ResCCL achieves up to 2.5× improvement in bandwidth performance compared to both NCCL and MSCCL. It reduces SM resource overhead by 77.8% and increases TB utilization by 41.6% while running the same algorithms. In end-to-end DLT, ResCCL boosts Megatron's throughput by up to 39%.

Segment Routing over IPv6 (SRv6) is an innovative and adaptable source routing technique that enhances interconnection services. It plays a pivotal role in next-generation networking technologies, providing crucial support for network telemetry, computing power networks, and related technologies. The end-to-end connectivity capability of SRv6 is highly regarded by ISPs, driving its widespread deployment in networks. However, configuring an SRv6 network can be challenging and prone to errors due to the complexity of low-level configuration languages and numerous protocol parameters. To address this issue, we present AutoSRv6, a system designed to synthesize SRv6 configurations for large, evolving networks using high-level abstractions of network topology and policies. AutoSRv6 leverages formal constraint-solving techniques and SMT solvers to compute protocol parameters and generate configuration files that align with network policies. Furthermore, AutoSRv6 incorporates a mechanism to overcome the constraints imposed by hardware, mapping the end-to-end path to a SID (Segment Identifier) sequence. We have developed a prototype of AutoSRv6 and conducted experiments on diverse network topologies, evaluating its performance with various network policies. The results show that autoSRv6 can generate the network configuration satisfying the policy, the time cost of IGP synthesis is better than the existing method, and the length of the segment list is optimized by more than 2 times.

Distributed machine learning over geo-distributed clouds enables joint training of data located in different regions, alleviating the burden of transferring large volumes of training datasets, which greatly saves bandwidth. However, the limited capacity of WAN links slows down the inter-cloud communications, which significantly decelerates the synchronization of distributed machine learning over geo-distributed clouds. Besides, the multi-tenancy in clouds results in multiple training tasks running simultaneously, whose synchronizations consistently compete for the limited WAN bandwidth with each other, which further aggravates the training performance of each task. While existing works optimize synchronizations through techniques like gradient compression, multi-resource interleaving and so on, none of them targets at the synchronization congestion especially due to multi-tenant learning, which results in inferior training performance.To solve these problems, we propose a simple but effective scheme, SCC, for fast and efficient multi-tenant learning via synchronization congestion control. SCC monitors the cross-cloud network conditions and evaluates the synchronization congestion level based on the round-trip transmission time for each synchronization. Then SCC alleviates synchronization congestion via controlling the synchronization frequency according to the synchronization congestion level in a probabilistic way. Extensive experiments are conducted within our testbeds consisted of 16 NVIDIA V100 GPUs to evaluate the performance of SCC, and comparison results show that SCC can reduce the average training completion time and makespan by up to 28.6% and 43.2% over SAP-SGD [1]. Targeted experiments are conducted to demonstrate the effectiveness and robustness of SCC.

Network measurement is critical for various network applications, but scaling measurement techniques to the network-wide level is challenging for existing sketch-based solutions. In software switches, centralized deployment provides low resource usage but suffers from poor load balancing. In contrast, collaborative measurement achieves load balancing through flow distribution across software switches but requires high resource usage. This paper presents a novel distributed deployment framework that overcomes the limitations above. First, our framework is lightweight such that it splits sketches into segments and allocates them across forwarding paths to minimize resource usage and achieve load balancing. This also enables per-packet load balancing by distributing computations across software switches. Second, through a novel collaborative strategy, our framework achieves finer-grained flow distribution and further optimizes load balancing. Third, we further optimize load balancing by eliminating the mutual influence among forwarding paths. We evaluate the proposed framework on various network topologies and different sketches. Results indicate our solution matches the load balancing of collaborative measurement while approaching the low resource usage of centralized deployment. Moreover, it achieves superior performance in per-packet load balancing, which is not considered in previous deployment solutions.

Advanced router configuration synthesizers aim to prevent network outages by automatically synthesizing configurations that implement routing protocols. However, the lack of interpretability makes operators uncertain about how low-level configurations are synthesized and whether the automatically generated configurations correctly align with routing intents. This limitation restricts the practical deployment of synthesizers.In this paper, we present NetKG, an interpretable configuration synthesis tool.(i) NetKG leverages a knowledge graph as the intermediate representation for configurations, reformulating the configuration synthesis problem as a configuration knowledge completion task; (ii) NetKG regards network intents as query tasks that need to be satisfied in the current configuration space, achieving this through knowledge reasoning and completion; (iii) NetKG explains the synthesis process and the consistency between configuration and intent through the configuration knowledge involved in reasoning and completion.We show that NetKG can scale to realistic networks and automatically synthesize intent-compliant configurations for static routes, OSPF, and BGP. It can explain the consistency between configuration and intent at different granularities through a visual interface. Experimental results indicate that NetKG synthesizes configurations in 2 minutes for a network with up to 197 routers, which is 7.37x faster than the SMT-based synthesizer.

With the rapid growth in data center network bandwidth far outpacing improvements in CPU performance, traditional software middleboxes running on servers have become inefficient. The emerging data processing units aim to address this by offloading network functions from the CPU. However, as DPUs are still a new technology, there lacks comprehensive evaluation of their capabilities for accelerating middleboxes. This paper benchmarks and analyzes the performance of offloading middleboxes onto the NVIDIA BlueField-2 DPU. Three key DPU capabilities are explored: flow tables offloading, ARM subsystem packet processing, and connection tracking hardware offload. By applying these to implement representative middleboxes for firewall, packet scheduling, and load balancing, their performance is characterized and compared to conventional CPU-based versions. Results reveal the high throughput of flow tables offloading for stateless firewalls, but limitations as pipeline depth increases. Packet scheduling using ARM cores is shown to currently reduce performance versus CPU-based scheduling. Finally, while connection tracking hardware offload boosts load balancer bandwidth, it also weakens connection creation abilities. Key lessons on efficient middleboxes offloading strategies with DPUs are provided to guide further research and development. Overall, this paper offers useful benchmarking and analysis of emerging DPUs for accelerating middleboxes in modern data centers.

In modern data centers and enterprise networks, software switches have become critical components for achieving flexible and efficient network management. Due to resource constraints in software switches, sketches have emerged as a promising approach for network traffic measurement. However, their accuracy is often impacted by hash collisions. Existing hash functions treat all collisions equally, failing to account for the differing impacts of collisions involving elephant flows versus mouse flows. We propose FSA-Hash, a novel flow-size-aware hashing scheme that separates elephant flows from each other and from mouse flows, minimizing the most detrimental collisions. FSA-Hash is designed based on two insights: separating elephant flows from mouse flows avoids overestimating mouse flows, while separating elephant flows from each other enables accurate heavy-hitter detection. We implement FSA-Hash using machine learning models trained on network traffic data (LFSA-Hash), and also design a lightweight online variant (OLFSA-Hash) that learns the hash model solely from sketch queries on the software switch, obviating traffic collection overheads. Evaluations across four sketches and two tasks demonstrate FSA-Hash’s superior accuracy over standard hash functions. Moreover, OLFSA-Hash closely matches LFSA-Hash’s performance, making it an attractive option for adaptively refining the hash model without monitoring traffic.

As an emerging architecture, edge computing enables resource limited terminal devices to offload their computation tasks to edge servers in the vicinity, to efficiently reduce delay and energy consumption. However, the continuous expansion of network scale and rapid growth of network traffic in recent years have brought huge challenges to task offloading and resource allocation. To tackle the challenges, by integrating Knowledge Defined Networking (KDN) and edge computing technologies, we design a novel Knowledge defined Edge Computing (KEC) architecture, to achieve intelligent resource allocation and task offloading in dynamic large-scale edge computing networks. We formulate the task offloading and resource allocation optimization problem, to minimize delay and energy consumption, by considering resource requirements and controller deployment. To solve it, we present an intelligent Resource Allocation based Task Offloading (TORA) mechanism, where a Multi-Agent SD3 based resource allocation (MASD3) algorithm is devised to perform efficient resource allocation. To adapt to the rapid expansion of network scale, we design a resource Allocation based Controller Deployment and task offloading Decision (DACD) algorithm, to perform the optimal controller deployment and task offloading. Extensive simulation experiments demonstrate the effectiveness and efficiency of our proposed solution, and TORA mechanism outperforms comparison mechanisms on delay and energy consumption.

Software-Defined Data Center Networks (SDDCNs) utilizes Software Defined Networking (SDN) as a network architecture to achieve highly flexible, programmable, and automated management of Data Center Networks (DCNs). The high energy consumption of DCNs remains a persistent and significant challenge. Thus, the energy saving is crucial and imperative for DCNs. Current energy-efficient solutions primarily rely on flow consolidation and dynamic device sleeping techniques to reduce energy consumption. However, these approaches often yield long Flow Completion Time (FCT), potentially resulting in violations of service-level agreements, particularly for delay-sensitive applications. In this paper, we formulate the problem of minimizing the power consumption in SDDCNs as key objective while ensuring timely FCT for delay-sensitive applications. To solve this problem, we first introduce the Active Network Generation (ANG) approach, which generates a minimal active subnet with the least number of active devices while meeting the current traffic demand. Subsequently, we propose two algorithms based on the type of applications: the Delay-Tolerant Flow Route (DTFR) algorithm for delay-tolerant applications and the Delay-Sensitive Flow Route (DSFR) algorithm for delay-sensitive applications. Simulation results demonstrate that our propose solution achieves an energy-saving rate of up to 67.77% and significantly reduces FCT compared to benchmark solutions.

Configurations deployed in routers govern the routing and addressing of IP-based networks. Automated tools for synthesizing and verifying configurations have been deployed to replace manual configurations, preventing network outages from misconfigurations. However, these tools offer limited assistance in real production networks as manual configuration remains the primary and preferred approach for daily operations. To address this issue, we present NetGenius, a configuration recommendation tool that assists operators in manually editing network configurations like a code editor. NetGenius employs a graph neural network to estimate the importance of neighboring commands to the center command, providing recommendations based on importance scores when the center command is used as input. First, NetGenius leverages a generic graph model named Knowledge Graph to model existing configurations and derives configuration features for each center command by composing it with its neighbors. Subsequently, NetGenius conducts a general approach based on the graph neural network to estimate and calculate corresponding feature importance scores. Finally, NetGenius uses a recommendation mechanism to recommend configurations based on importance scores while adhering to specific constraints. We extensively evaluate the performance of NetGenius using real-world configurations from different vendors. The experimental results demonstrate that NetGenius achieves the highest recommendation accuracy of 98.59% across various datasets. Moreover, when applied to a large-scale network comprising$152,475$configuration commands, NetGenius completes its training within 6 seconds and generates recommendations in 5 milliseconds.

In-band Network Telemetry (INT) technology enables fine-grained network monitoring by encapsulating intra-switch network status into INT probes, which is essential in data center networks for ensuring Quality of Service (QoS). Existing INT-based telemetry systems leverage centralized controllers to compute non-overlapping probe paths, thereby facilitating lightweight and network-wide measurements. However, these systems fail to adapt effectively to network topology changes caused by link or device failures, primarily due to inflexible path planning under dynamic conditions. To address this problem, we propose INT-Source, a unified policy-based network-wide telemetry system for probing and forwarding. First, we design a data plane forwarding mechanism for INT probes to ensure telemetry coverage during topology changes and reduce telemetry overhead. Second, we design a probe packet structure and introduce a switch-based probe verification and discard mechanism to prevent redundant link probing. Third, we introduce two algorithms for INT-Source: a Single-Source algorithm to facilitate deployment and a Multi-Source algorithm to enable lightweight and scalable telemetry. Our evaluation shows that INT-Source reduces bandwidth overhead to 12.5% compared to existing methods across three network topologies. Even with a 10% link failure rate, INT-Source is able to monitor 93.1% of network ports, demonstrating strong robustness.

With the expansion of production networks, new challenges arise in scaling telemetry systems to accommodate the massive number of network devices. In-band Network Telemetry (INT) is widely adopted for its fine-grained and accurate measurements. However, system robustness and performance scalability remain key challenges for INT-based network-wide telemetry systems. Existing INT-based measurements manage the network as a whole and rely on a centralized controller for path planning and telemetry data collection. As networks scale and the probability of failures increases, frequent re-planning leads to prolonged telemetry interruptions. In this work, we propose INT-Partition, a hierarchical and fault-tolerant in-band network telemetry system with a divide-and-conquer paradigm. INT-Partition conducts telemetry in two stages: network partitioning and telemetry within each partition, enabling scalability for mega-scale networks. Our evaluations, including mega-scale simulations using BMv2 software switches and small-scale validations on Tofino hardware switches, demonstrate the effectiveness of our approach. With 1% of network equipment out of order, INT-Partition covers over 89.67% of the area and accurately locates faults. As the network scales, telemetry planning and deployment time is reduced by 75.62% to 96.02%, and hot reloading enables seamless switching of telemetry deployment.

Damage or relocation of network devices necessitates the replacement of devices. Additionally, manufacturers of devices may also change due to factors such as price, policies, and functional limitations. This brings great needs of converting existing configurations to fix new devices and ensuring consistent network policy behaviors, which is configuration translation. Manually translating requires network administrators to possess significant configuration expertise and to spend extreme time translating thousands of commands for a core device. Therefore, we propose ConfigTrans, a novel configuration translation method for commands with and without parameters based on constraint solving and large language models (LLMs), respectively. For commands with parameters, we explore constraints about parameters, commands, and format limitations, and design a heuristic algorithm to solve them. For commands without parameters, we utilize an LLM to find the translation. The most likely several commands are selected based on semantic similarity, and the final result and keywords in it are chosen by the LLM. To meet the requirements of the view, the result commands are arranged while supplementing missing higherview commands. Experimental results show that our method achieves an accuracy rate of 82.47 % for translating commands with parameters and 75.5 % for commands without parameters.

Network measurement is critical for various network applications, but scaling measurement techniques to the network-wide level is challenging for existing sketch-based solutions. Centralized sketch deployment provides low resource usage but suffers from poor load balancing. In contrast, collaborative measurement achieves load balancing through flow distribution across switches but requires high resource usage. This paper presents a novel lightweight distributed deployment framework that overcomes the limitations above. First, our framework is lightweight such that it splits sketches into segments and allocates them across forwarding paths to minimize resource usage and achieve load balancing. This also enables per-packet load balancing by distributing computations across switches. Second, our framework is also optimized for load balancing by coordinating between flows and enabling finer-grained flow distribution. We evaluate the proposed framework on various network topologies and different sketch deployments. Results indicate our solution matches the load balancing of collaborative measurement while approaching the low resource usage of centralized deployment. Moreover, it achieves superior performance in per-packet load balancing, which is not considered in previous deployment policies. Our work provides efficient distributed sketch deployment to strike a balance between load balancing and resource usage enabling effective network-wide measurement.

Network measurement plays a critical role in numerous network applications that rely on fundamental flow processing tasks such as frequency estimation, heavy hitter detection, and distribution estimation. Sketch has emerged as an efficient approach for network measurement due to its low overhead. However, most sketch-based solutions target static windows while enabling sliding window-based measurement remains an open challenge. This paper introduces two novel general frameworks applicable to diverse sketch models for sliding window-based network measurement: a traditional sliding window framework and a fine-grained flow-level framework. The traditional framework divides the window into parts and uses centralized flushing to remove expired parts. The flow-level framework tracks timestamps to maintain exact flow characteristics over one period, preventing truncation. To optimize memory usage, a bit-wise adaptive allocation algorithm allows dynamic borrowing of unused counter bits. The frameworks are evaluated on sketches for different flow processing tasks. Results show the frameworks are widely generalizable, reduce error substantially compared to existing approaches, and provide more efficient memory usage.

Network configuration plays a vital role in quality assurance of network services, requiring considerable effort and time. Automatic network configuration approaches are promising due to their capacity to automatically generate and verify configurations. However, these methods suffer from drawbacks, such as generated configuration content being largely unknown to network operators and inefficient for large-scale networks. Manual configuration is thus still the primary way of managing networks. To facilitate editing processes of manual configuration, a network-wide tool for recommending custom keywords is in urgent need. In this article, we propose a keyword recommendation tool that recommends custom keywords across various network devices. We observe that network devices of the same type and role tend to have a unified template and similar configurations, which enables recommending custom configurations between them. However, the vision entails the following three challenges. First, configurations need to be modeled accurately. Second, a wide variety of network protocols need to be supported. Third, relationships between custom keywords might be implicit and difficult to find. To address the challenges, we first built a configuration knowledge graph that could accurately model configurations, extract latent relationships between keywords, and generate explainable recommendations. Then we applied a recommendation framework to the graph for appropriate keyword recommendations. Lastly, to validate the performance, we conduct recommendations on real configurations over 26 000 times. Experimental results indicate that the overall coverage rate for matching expected configurations reaches 79.396%, and the redundancy rate is less than 20%.

Cellular networks are vital for emerging applications like the Metaverse, which impose demanding quality and quantity requirements. This necessitates frequent reconfiguration of both new and existing base stations to balance network service quality (e.g., ultra-low latency and high bandwidth) and resource consumption. Existing data-driven configuration methods learn from historical data, but have two key limitations. First, they yield only approximate solutions, lacking precision. Second, poor bootstrapping for new base stations with previously unobserved attributes. In this paper, we pioneer intent-driven configuration synthesis by designing an intent language and utilizing satisfiability modulo theory (SMT) for cellular networks to enable exact and precise solutions. We formulate synthesis as an SMT problem, permitting verification of precision. First, we cast configuration generation as a program synthesis problem via novel modeling to bridge the intent-configuration gap. Second, we extend SMT synthesis to scale to large networks. However, vanilla SMT approaches have poor scalability. Hence, we propose an optimization using sampling for constraint verification instead of exhaustive forward solving. We also design a domain-specific optimization to prune the sample space and improve efficiency. Experiments on various network scales demonstrate the effectiveness of our proposed SMT-based cellular network configuration synthesis.

Programmable switches allow data plane to program how packets are processed, which enables flexibility for network management tasks, e.g., packet scheduling and flow measurement. Existing studies focus on program deployment at a single switch, while deployment across the whole data plane is still a challenging issue, especially manifested in the difficulty in joint correct implementation of P4 programs, resource load balancing of network devices, and optimization of network performance. In this paper, we present RED, a Resource-Efficient and Distributed program deployment solution for programmable switches. First of all, we analyze data plane programs to estimate the resource utilization and divide them into two categories for further processing. Then, the proposed merging and splitting algorithms are selectively applied to merge or split the pending programs. Finally, we consolidate the scarce resources of the whole data plane for distributed program deployment. Extensive experiments with both testbed and large-scale simulations are conducted and comparison results show that 1) RED achieves network-wide resource balancing in a distributed way and the latency of processing packets within the switch was reduced by 16.7%. 2) RED improves the speedup by two orders of magnitude compared to P4Visor in merging program and merges more 18% tables than SPEED; 3) If the resources required to run a P4 program exceed the resource limit of the switch, it cannot be deployed on the switch. RED makes the overwhelmed programs to be deployed on switches and switch throughput increased by 10.7%.

Traditional routing algorithms cannot dynamically change network environments due to the limited information for routing decisions. Meanwhile, they are prone to performance bottlenecks in the face of increasingly complex business requirements. Some approaches, such as deep reinforcement learning (DRL) have been proposed to address the routing problems. However, they hardly utilize the information about the network environment fully. The Knowledge Defined Networking (KDN) architecture inspires us to develop new learning mechanisms adapted to the dynamic characteristics of the network topology. In this paper, we propose an effective scheme to solve the routing optimization problem by adding a graph neural network (GNN) structure to DRL, called Message Passing Deep Reinforcement Learning (MPDRL). MPDRL uses the characteristics of GNN to interact with the network topology environment and extracts exploitable knowledge through the message passing process of information between links in the topology. The goal is to achieve the load balance of network traffic and improve network performance. We have conducted experiments on three Internet Service Provider (ISP) network topologies. The evaluation results show that MPDRL obtains better network performance than the baseline algorithms.

Network verification has recently made strides, focusing on the satisfiability of configurations and policies or the performance and versatility of their methods. However, they generally ignore explainability, which is the ability to explain why a network violates or satisfies a certain forwarding policy. In this paper, we propose an explainable network verification framework XNV, which uses a novel interpretable fault analysis method to construct an effective explainable network verifier using knowledge graph (KG). XNV provides appropriate explanations to help operators understand the verification results, improving the transparency and trustworthiness of the verification system. First, XNV uses the KG as an intermediate representation of the configuration semantic level, storing the configuration semantics and routing protocol states. Then, XNV constructs human-logical fault trees for policies and implements root-cause analysis of policy violations based on KG queries and minimum cut set matching. Experiments and case evaluations show that our system provides good interpretability while balancing performance, accelerated understanding, and handling of misconfigurations.

With the development of network measurement technologies, a hybrid measurement architecture can effectively optimize the sketch structure in switches, making it more adaptable to the current complex and volatile network environment. However, current optimization technologies based on hybrid measurement architectures generally suffer from insufficient automation, difficulty of learning effective numerical features, and lack of generality, resulting in poor scalability in real deployment. To solve these problems, we propose theTalentSketchframework, based on which we further developDeepSketchfor effective sketch optimization. First, we useSeq2Seqto automatically identify target flows instead of relying on manual thresholds. Second, we propose a new training strategy that extracts low-precision flows for models with weak learning capabilities. Last, we develop a new sketch optimization framework that can optimize different kinds of sketches only by changing the training data for generality. A large number of experimental results show thatDeepSketchexhibits superior performance. For example: (1) the accuracy of optimized sketches has increased by 20% to 73%, (2) Without replacing the model structure, the accuracy of the optimized sketches can generally reach over 80%. (3) The impact of low sampling rates on accuracy is less than 1% on various sketches.

In-band network telemetry (INT) allows for fine-grained network monitoring, without requiring communication with the controller at each hop. Existing INT-based network-wide telemetry systems achieve low-overhead monitoring with non-overlapping path planning algorithms. However, these systems do not constrain the length of the generated probing paths, which will lead to packet loss when the size of the packet with collected telemetry data exceeds the MTU limit. To address this issue, we propose MTU-adaptive path segmentation algorithms step by step in this paper. Initially, we present two single-path planning algorithms: the INT-optimize algorithm, which produces a single path that covers the entire network with the lowest southbound communication overhead, and the INT-low-cost algorithm, which further accelerates the INT-optimize. Next, to consider the MTU limit, we propose the single-MTU adaptive INT-Segment algorithm to divide the single long path generated in the previous step into multiple path segments. In addition, we generalize the MTU-adaptive network telemetry problem and propose a multi-MTU adaptive INT-Segment solution to achieve high-performance network telemetry in networks with multiple MTU settings. Extensive evaluations demonstrate that our proposed MTU-adaptive solutions can achieve sub-second network-wide telemetry for large-scale networks, with less than 2.9ms to calculate the probing paths for an 18-pod FatTree. Furthermore, our multi-MTU adaptive INT-Segment solution significantly reduces the number of INT Sinks and INT Sources by 13.25%-42.39% when deployed in multi-MTU networks while maintaining stable telemetry data collection time. Compared with the state-of-the-art INT-path, our solution adapts the probing path to the network MTU limit, producing a telemetry data collection efficiency improvement of 10%-94%.

Federated unlearning (FUL) is an emerging distributed machine learning paradigm which enables the removal or unlearning of specific training data effects from trained Federated Learning (FL) models. While current studies mostly focus on client-side FUL to address the “right to be forgotten”, and ignore the server's right to remove local models from the global model, particularly when clients are trained with low-quality data. In this paper, we introduce the Server-Initiated Federated Unlearning (SIFU) algorithm, devised to eliminate low-quality data from the global model. SIFU consists of two main components: (i) Identifying low-quality data: we develop a category-based method for quantifying low-quality data for each client and filter out clients containing such data. Datasets are then divided accordingly. (ii) Unlearning low-quality data: we employ gradient ascent training to counteract the adverse effects of low-quality data on local models. To minimize any bias introduced, we concurrently perform several batches of boosting training with good-quality data. SIFU could identify and promptly eliminate the impact of low-quality data on the FL global model while still preserving the benefits of good-quality data. Finally, extensive evaluations are conducted to verify the performance of SIFU with four different kinds of datasets and models. Results show that, compared to retraining from scratch, SIFU accelerates the speed of unlearning by 15× for small datasets (i.e., MNIST and FMNIST) and 20× for large datasets (i.e., CIFAR-10 and CelebA) without any degradation in accuracy, which also outperforms the state of the arts.

Enforcing scheduling policies with software schedulers at end-hosts leads to high CPU consumption, low throughput, and inaccuracies. To address these issues, offloading packet schedulers to the network interface card presents a promising research direction. However, existing attempts suffer from inflexible on-NIC scheduling that cannot execute complex hierarchies of network policies. In this article, we propose FlowValve+, a general framework for multi-queue packet scheduling on SoC-based SmartNICs. The key insight behind FlowValve+is to abstract the inherent queueing system as a single FIFO queue and perform specialized tail drop to mix the FIFO queue with expected flow proportions. FlowValve+leverages hardware accelerations to produce high throughput while substantially reducing CPU and memory usage on end-hosts. We prototype FlowValve+on Netronome Agilio 40GbE and NVIDIA Bluefield-2 100GbE SmartNICs to demonstrate its ability to accurately enforce network policies while driving TCP traffic at 40 Gbps and 80 Gbps on both platforms, respectively. Moreover, FlowValve+can save two CPU cores compared to DPDK packet schedulers.

In recent years, probabilistic network analysis has received significant attention as a crucial aspect of network management. With the increasing use of distributed routing protocols in networks, computing the probabilities of multiple properties becomes a challenging task, as different property violations derive from different failure scenarios. While some previous work has attempted to prune the space of failure scenarios by identifying a set of links whose failure does not change whether a property holds, this approach often requires multiple calls to the system, leading to redundant computation when verifying multiple properties. To solve this problem, we propose NetDiceSyn, a scalable multi-property probabilistic network configuration analyzer. As verifying each property requires exploring a state tree, NetDiceSyn proposes to explore all state trees synchronously and compute a shared data plane for multiple states in different state trees, thus reducing the redundant computation between different state trees and verifying multiple properties simultaneously. Besides, we design a state sharing method to reduce the redundant storage of states during synchronous exploration. Additionally, we propose abstract topology to merge equivalent states within a state tree to speed up probabilistic verification in series link networks. Extensive experiments are conducted with real-world network topologies, and the results show that NetDiceSyn outperforms state-of-the-art methods, providing up to an order of magnitude speedup when verifying hundreds of properties for networks with hundreds of links.

Traffic simulation provides interactive data for the optimization of traffic control policies. However, existing traffic simulators are limited by their lack of scalability and shortage in input data, which prevents them from generating interactive data from traffic simulation in the scenarios of real large-scale city road networks.

The rapid growth of mobile device (e.g., smart phone and bracelet) has spawned a lot of new applications, during which the requirements of applications are increasing, while the capacities of some mobile devices are still limited. Such contradiction drives the emergency of computation migration among mobile edge devices, which is a lack of research currently. In this article, we focus on addressing the computation migration oriented resource allocation problem among mobile edge devices. Specifically, we first construct a framework for Mobile Social Cloud(MSC), in which the mobile devices with rich resources are abstracted as resource suppliers and those resource-lacking devices are abstracted as resource demanders. Then, a mathematical model is formulated and an evolutionary algorithm is proposed to effectively solve this model based on decomposition, dominance, and genetic operations. Moreover, the parallel computing is introduced to further improve the efficiency of the proposed algorithm. The experimental results indicate that the proposed algorithm outperforms the other state-of-the-art methods and it improves the calculation efficiency by about 178 percent (2 cores) and 262 percent (3 cores) by introducing parallel computing.

Signal phase optimization at signalized intersection is of great importance to urban traffic control and management yet is very challenging. The traditional approaches for signal phase optimization heavily rely on the traffic engineering practitioners’ experience. To tackle these challenges, a novel data-driven method is proposed to realize signal phase optimization and recommendation solely using limited amount of real signalized intersection samples. Firstly, all of discrete features related to signal phase design, encoded by one-hot representation, are sampled by the Gumbel-SoftMax distribution, which is a continuous approximation to a multinomial distribution. With this approximation distribution, the generative adversarial network (GAN) is applied to produce the most acceptable signal phase samples among all acceptable choices, dealing with the problem of insufficient samples and uneven sample distribution in real word. Thirdly, a decision-tree based classifier is established to realize signal phase recommendation automatically. We conducted extensive experiments to evaluate our proposed method on three cities in China, including Beijing, Tongxiang and Chaozhou. The experimental results showed that the proposed method could effectively improve the signalized intersection operation efficiency. Moreover, the proposed method has already been deployed in several cities, and it successfully keeps serving hundreds of signalized intersections. This confirms that SPSRec is a practical and robust solution for large-scale real-world signal control services.

Enforcing scheduling policies at end-hosts with software schedulers suffers from high CPU consumption, low throughput, and inaccuracy. Offloading scheduling functions to the network interface card (NIC) provides a promising direction to address these problems. However, existing efforts in scheduling offloading suffer from inflexible on-NIC packet schedulers, which cannot execute complex hierarchies of network policies. In this paper, we present FlowValve, the first parallel packet scheduler for Network Processor (NP)-based SmartNICs that offloads critical functions of Linux traffic control, including packet classifying and scheduling. The key insight behind FlowValve is to abstract inherent queues attached to the NIC interface (wire side) as a single FIFO queue and perform specialized tail drop to mix the FIFO queue with expected flow proportions. FlowValve takes advantage of on-chip multi-core parallelism and hardware accelerations to produce high throughput. Meanwhile, it substantially reduces CPU and memory burdens on end-hosts. We prototype FlowValve on a Netronome Agilio SmartNIC and demonstrate its effectiveness against non-offloaded kernel schedulers and DPDK QoS Scheduler. We find that FlowValve outperforms both in accurately enforcing network policies while driving line rate performance (i.e., 40Gbps), which contributes to saving at least two CPU cores.

With the massive and rapid growth of network traffic, sketching techniques are widely used to estimate a variety of network flow-level metrics, e.g., flow size, top-k flows, and the number of flows. However, how to cope with the constantly changing network environment and achieve high accuracy with limited switch resources is still a challenging problem. Existing studies do not well address the issues of identifying and correcting the inaccurately estimated flows. Therefore, we propose TalentSketch, an adaptive and high-precision hybrid measurement framework based on LSTM. TalentSketch uses LSTM to learn the sketch features with the captured traffic information. It could identify the inaccurately estimated flows with an error-prone flow model and correct them with a well-trained regression model. We conduct extensive experiments to verify the performance of TalentSketch. Experimental results show that, without increasing switch resource overhead, TalentSketch improves the measurement accuracy of different kinds of sketches by 12%∼23%. More importantly, it can track network fluctuations and provide feedback on the overall accuracy of a sketch in real-time.

In-band network telemetry (INT) enables hop-by-hop fine-grained network monitoring without interacting with the controller at every hop. Existing INT-based network-wide telemetry systems achieve low-overhead monitoring with the non-overlapped path planning algorithms. However, they do not bound the length of the generated probing paths, which may lead to packet loss when the collected telemetry data exceeds the MTU limit. In this paper, we propose an MTU-adaptive path segmentation algorithm to solve this problem. First, we provide two single-path planning algorithms: the INT-optimize algorithm produces a single path that covers the entire network with the lowest southbound communication overhead, and the INT-low-cost algorithm further improves the path planning efficiency of the INT-optimize. By taking the MTU limit into account, we further propose INT-Segment, a novel path segmentation algorithm to split the single long path produced from the previous step into multiple path segments. Extensive evaluations show that the proposed INT-Segment can realize sub-second network-wide telemetry for large-scale networks. It takes less than 2.9ms to calculate the probing paths for an 18-pod FatTree. Compared with the state-of-the-art INT-path, our solution makes the probing path well adapted to the network MTU limit and improves the telemetry efficiency by 10%-94%.

In-band network telemetry (INT) delivers high-precision network monitoring by collecting device-internal states entirely on the data plane. For rapid congestion awareness and network troubleshooting, it is necessary to conduct network-wide telemetry by generating multiple monitoring paths covering the entire network graph. Solving the optimal path planning problem used the eulerian trail initially at a time complexity of$O(k(3E+V-15k/2))$. For mega-scale data center networks, such a high complexity is unacceptable because the algorithm cannot adapt well to occasional topology changes. In this work, we propose improved INT-path and INT-react, two refined path planning algorithms with a much reduced time complexity of only$O(E)$. Furthermore, INT-react also considers balanced path generation to reduce the longest path length for synchronized collection of telemetry data from each monitoring path. The evaluation shows that on average it costs 2.10s for the improved INT-path to solve the optimal path planning for a network of 9500 switches, while the computation is completed within only 0.283s on average for INT-react. In addition, INT-react reduces the longest path length. INT-react's path planning is so fast that it promptly reacts to topology changes and is ready to be deployed in mega-scale production networks.

Time-Sensitive Networking (TSN) is an emerging network architecture that provides bounded latency and reliable network services for time-sensitive applications. Since time-triggered flows in TSN are typically periodic, a concept of network cycle is widely used in both standards and academic researches. However, although network cycle has gained popularity, its rationale has not yet been analyzed systematically.In this paper, we mathematically evaluate the performance of several flow scheduling algorithms in terms of flow schedulability with and without employing network cycle. We observe that only when the network cycle is set to a proper value can the performance of flow scheduling be significantly improved. To better evaluate the scheduling effect, a novel assessment metric and a goal-based optimization algorithm are introduced. Our experiment results show that the network cycle-based algorithm can achieve a considerable improvement (40% - 170% improvement in the number of scheduled flows) compared to the ones with network cycle disabled.

Abstract 5G cellular IoT has several advantages compared to other access technologies, enabling operators to serve a wider area and more IoT devices. However, in the urban transportation system, a massive number of vehicles exhaust the available resources in the cell, resulting in excessive load in the 5G cellular network. This article proposes a novel reinforcement learning based V2V routing (RLbR) framework, which offloads non‐realtime traffic into the V2V network and significantly relieves the load of 5G cellular network. Meanwhile, we propose a V2V routing algorithm. Specifically, the Q ‐values of neighbouring vehicles are firstly calculated according to the cache factor CF and energy factor EF and evaluate the quality of neighbouring vehicles. Then, the position factor PF is calculated, based on which, the vehicle forwards the data packet. In addition, an environment model is designed to accelerate the convergence of Q ‐table. The results show that the RLbR framework brings the highest offload rate compared to the other three frameworks, and simultaneously, the proposed algorithm improves the lifetime of the V2V network and performs well in terms of delivery ratio and average delay.

Mobile-edge computing (MEC) sinks computation and storage capacities to network edge, where it is close to users to support delay-sensitive services. However, due to the dynamic and stochastic properties of MEC networks, the deployed services may be frequently migrated among edge servers to follow the mobility of users, which greatly increases the network operational cost. In this article, considering the service migration cost brought by user mobility, we study the joint optimization problem of service deployment and request routing decisions to maximize the long-term network utility of MEC networks. First, we propose a Lyapunov optimization-based online service migration algorithm to decompose the continuous optimization problem into a number of one-slot online optimization problems. Then, to address the NP-hard issue of one-slot optimization, we use a randomized rounding technique to implement service migration and request routing. Furthermore, through a closed-form theoretical analysis, we prove that the proposed algorithm not only greatly meets the local user requests and enables approximate performance guarantees but also adaptively balances the service migration cost and system performance online. Finally, extensive simulations are conducted, which demonstrate that our algorithm can efficiently utilize the storage and computation resources of edge servers, and maximize the long-term network utility while ensuring the stability of service migration cost.

The rapid development and application of 5G/B5G generate tremendous amount of traffic which in turn cause great burden for the corresponding transmission network. One typical way to address such challenge is to sink the content (e.g., 4K and 8K videos) from the remote cloud to the edge servers. In this case, how to efficiently visiting and getting these contents becomes a new problem, in which the cooperation between cloud and edge should be taken into consideration. In this regard, this work builds an edge and cloud cooperative routing and caching system which consists of three main modules of content popularity prediction, cooperative caching and cooperative routing. Specifically, the content prediction is designed by jointly leveraging the technologies of Long Short-Term Memory (LSTM) and Temporal Pattern Attention (TPA) to dig the traffic features and predict the future content popularity. Based on the prediction results and the technology of reinforce learning, the cooperative caching module designs both a reactive content replacement and an active content caching strategies. After that, the cooperative routing is carried out to help customers visiting and obtaining these content efficiently with the objective of minimizing the overhead. The experimental results indicate that the proposed methods outperform the state-of-the-art benchmarks in terms of the caching hit rate, the average throughput, the successful content delivery rate and the average routing overhead.

With the development of communication and transmission technologies, more and more applications, like Internet of vehicles and tele-medicine, become more sensitive to network latency and accuracy, which requires routing schemes to be more efficient. In order to meet such urgent need, learning-based routing strategies emerges, with the advantages of high flexibility and accuracy. These strategies can be divided into two categories, centralized and distributed, enjoying the advantages of high precision and high efficiency, respectively. However, routing become more complex in dynamic network, where the link connections and access states are time-varying, so these learning-based routing mechanisms are required to be able to adapt to network changes in real time. In this paper, we designed and implemented both two of centralized and distributed reinforcement learning-based routing schemes (RLR-T). By conducting a series of experiments, we deeply analyzed the results and gave the conclusion that the centralized is better to cope with dynamic networks due to its faster reconvergence, while the distributed is better to handle with large-scale networks by its high scalability.

With the rapid growth of the number of IoT devices on the Internet, security problems of IoT devices are becoming more and more serious, which bring more challenges to network administrators. The first task to solve these problems for network administrators is being aware of IoT devices in the network. Previous IoT device identification methods typically use supervised machine learning methods, which require a large amount of labeled sample data. However, it is difficult to obtain a large number of labeled samples effectively. In order to address this problem, we propose an unsupervised IoT device fingerprinting method at the network level, which can effectively cluster IoT devices without labeled samples. We deeply analyze the temporal and spatial dimension characteristics of network traffic, which can adequately reflect the differences between different IoT devices. By using these features, we develop a clustering framework based on variational autoencoder and K-means algorithms. We conduct evaluation experiments on a public dataset including 24 different IoT devices. The experimental results show that our clustering algorithm can achieve accuracy of 86.7% outperforming a k-NN based state-of-art supervised approach.

Visibility is essential for operating and troubleshooting intricate networks. In-band Network Telemetry (INT) has been embedded in the latest merchant silicons to offer high-precision device and traffic state visibility. INT is actually an underlying technique and each INT instance covers only one monitoring path. The network-wide measurement coverage therefore requires a high-level orchestration to provision multiple INT paths. An optimal path planning is expected to produce a minimum number of paths with a minimum number of overlapping links. Eulerian trail has been used to solve the general problem. However, in production networks, the vantage points where one can deploy probes to start and terminate INT paths are constrained. In this work, we propose an optimal path planning algorithm, INT-probe, which achieves the network-wide telemetry coverage under the constraint of stationary probes. INT-probe formulates the constrained path planning into an extended multi-depot k-Chinese postman problem (MDCPP-set) and then reduces it to a solvable minimum weight perfect matching problem. We analyze algorithm's theoretical bound and the complexity. Extensive evaluation on both wide area networks and data center networks with different scales and topologies are conducted. We show INT-probe is efficient, high-performance, and practical for real-world deployment. For a large-scale data center networks with 1125 switches, INT-probe can generate 112 monitoring paths (reduced by 50.4 %) by allowing only 1.79% increase of the total path length, promptly resolving link failures within 744.71ms.

Many cloud applications in modern datacenters have very demanding latency requirements, making flow completion time (FCT) an important metric for evaluating the network performance. Existing network flow scheduling methods either base on pre-known information or have poor performance. Therefore, we present LAFS, an efficient learning-based flow scheduling approach which minimizes the FCT with estimated information of flows. LAFS combines system call monitoring and learning methods to learn the flow size and implements the Shortest Remaining Processing Time (SRPT) principle with in-network priorities. Moreover, LAFS adopts flowlets to alleviate the packets disorder problem in fine-grained flow scheduling. Our theoretical analysis and extensive simulations show that LAFS is a practical design and significantly outperforms other information-agnostic designs like DCTCP and PIAS under diverse workloads.

Power-aware traffic engineering via coordinated sleeping is usually formulated into Integer Programming problems, which are generally NP-hard with unbounded computation time for large-scale networks. This results in delayed control decision making in dynamic network environments. Motivated by advances in deep Reinforcement Learning, we consider building intelligent systems that learn to adaptively change router/switch’s power state according to changing network conditions. Neural network’s forward propagation can greatly speed up power on/off decision making. Generally, conducting RL requires a learning agent to iteratively explore and perform the "good" actions based on the feedback from the environment. By coupling Software-Defined Networking for performing centrally calculated actions to the environment and In-band Network Telemetry for collecting feedback from the environment, we develop GreenTE.ai, a closed-loop control/training system to automate power-aware traffic engineering. Furthermore, we propose novel techniques to enhance the learning ability and reduce the learning complexity. With both energy efficiency and traffic load balancing considered, GreenTE.ai can generate reasonable power saving actions within 276ms under a network testbed of 11 software P4 switches.

Software-Defined-Networking (SDN) is progressively dominating the dynamic management for timely network trouble shooting and fine grained traffic scheduling in data center networks. One critical issue in SDN is to reduce the communication overhead between the switches and the controller. Such overhead is mainly caused by handling miss-match packets, because for each miss-match packet, a switch will send a request to the controller asking for forwarding rule. Existing approaches to address this problem generally need to deploy intermediate proxy or authority switches to hold rule copies, so as to reduce the number of requests sent to the controller. In this paper, we argue that using the intrinsic buffer in a SDN switch can also greatly reduce the communication overhead without using additional devices. If a switch buffers each miss-match packet, only a few header fields instead of the entire packet are required to be sent to the controller. Experiment results show that this can reduce 78.7 percent control traffic and 37 percent controller overhead at the cost of increasing only 5.6 percent switch overhead on average. If the proposed flow-granularity buffer mechanism is adopted, only one request message needs to be sent to the controller for a new flow with many arrival packets. Thus the control traffic and controller overhead can be further reduced by 64 percent and 35.7 percent respectively on average without increasing the switch overhead.

With the explosive growth of mobile devices at the edge of mobile social networks (MSNs), the amount of the content that needs to be transmitted is exploded. Traditional content delivery mechanisms leverage only local information to make routing decisions, which results in both high latency and low delivery rate. Software-defined networking (SDN) is a novel network paradigm, the design philosophy of which could be applied to MSN for improving the content delivery performance. In this article, the centralized control thought of SDN is introduced into MSN to efficiently process social information. The classical routing algorithm of BubbleRap is improved from the perspective of network density, which is the basis of designing the sparse and dense routing mechanisms for MSNs. In addition, flexibly switching between these two routing mechanisms is implemented by a discriminating scheme, achieving efficient yet adaptive routing. The experimental results show that the delivery ratio of sparse routing is up to 83%, and the dense routing could reach up to 93%.

Wireless communication demands increase sharply with the explosive growth of mobile devices. The communications mainly depend on the infrastructure-based networks, e.g., WLANs and cellular networks. However, such wireless connections may be unavailable in crowded areas (e.g., concert and conference hall) or interrupted by infrastructure failures caused by earthquake or tsunami. These promote the evolution of local communication systems over device-to-device communication, such as Bluetooth and Wi-Fi Direct (WFD). However, none of the existing studies construct a full-featured local communication system, and they do not consider how to support the user mobility either. In this article, we implement and evaluate the performance of a WFD-based local communication system. First, we improve the intragroup communication by the native implementation of WFD on the Android platform, and propose an application-layer forwarding solution for the intergroup communication, which can be applied to three or more connected groups. Then, we put forward a self-adaptive handover mechanism taking user mobility and node failures into account. To deal with the uncertainty in the handover decision procedure, a fuzzy-logic-based normalized quantitative decision algorithm (FNQD) with the weights derived from the fuzzy analytic hierarchy process (FAHP) is utilized. Finally, we evaluate the performance of the system through both simulation and experiment analysis. Results show that we can get a maximum throughput of 31.7 Mb/s for the intragroup communication and a maximum goodput of 4.76 Mb/s for the intergroup communication. What is more, mobile devices could perform various types of handover according to their roles and status, which could improve the robustness of the local communication system.

With the rapid increase in service requirements driven by Internet of Things (IoT) networks, mobile crowdsourcing has become a compelling paradigm that can efficiently solve complex tasks in the physical world. Nevertheless, we found that most IoT tasks have constraints on deadline, location, and resource consumption, which limit the application of crowdsourcing platforms in the IoT networks. In this article, we innovatively propose a reliable fog-based temporal-spatial crowdsourcing for serving the above tasks. In this scenario, the key point is to achieve the best match of the attributes among tasks, fog nodes, and workers. As the bridge of the other two parts, fog nodes determine the orientation of tasks. Therefore, we present a temporal-spatial task allocation (TS-TA) scheme in the fog layer, aiming to make task results more reliable. In this scheme, we build a temporal-spatial attribute learning model based on the user behaviors. Then, we use the users' interest attribute matching model to identify the candidate fog nodes that satisfy the requirements of temporal-spatial tasks. We choose the fog nodes with low spatial correlation that is benefit to defense the attack on the nodes in the intensive area. Meanwhile, we assign the redundancy nodes for intrusion response through replacing the attacked/negative node. Both theoretical and real-topology simulation results validate that the proposed scheme can get better performance in system resource consumption and system robustness compared with other benchmark schemes.

The cycle-based volume is critical for traffic state estimation and signal control optimization at signalized intersections. Traditional volume estimation mainly depends on fixed detectors represented by loop detectors, but limited spatial coverage and detection failure are also prominent. With the development of vehicle positioning, smartphone-based navigation, and connected-vehicle technologies, massive high-resolution trajectory data have recently become available, which can provide rich and timely information on the traffic arrival and departure processes at signalized intersections. Hence, the studies utilizing trajectory data for estimating the queue length and traffic volume at intersections has received increasing attention in the past few years. However, the most existing studies have demanded a comparatively high penetration rate and adopted site-specific assumptions for unsteady arrival patterns. In contrast, this paper solely used trajectory data for cycle-based flow estimation through a generic hybrid method that combined a probabilistic model and shockwave theory to maximize the utilization of limited captured trajectories, especially under a low penetration rate. In this method, within each cycle, the volume of stopped vehicles is estimated based on the shockwave theory, while the volume of non-stopped vehicles is modeled as a parameter estimation problem of a time-dependent constrained Poisson distribution, where the time headway correspondingly obeys an M3 distribution. The cycle-based volume is solved by a maximum likelihood estimation using an expectation-maximization procedure. An empirical case study was conducted with various signal timing schemes and the results showed satisfactory robustness with an accuracy of more than 90% under a penetration rate of 7.6%.

With the drastic increase in the number and capabilities of mobile devices, user demands for bandwidth-hungry applications such as video streaming and multimedia file sharing are pushing the limits of current cellular systems. Offloading resources sharing services from the cellular infrastructure to device-to-device (D2D) networks can improve spectrum efficiency and expand system capacity. This paper designs and implements new D2D networks formation mechanism based on commercial off-the-shelf smartphones, and data transmission mechanism among these devices using Wi-Fi Direct(WFD). In this system, we realize the single-hop and multi-hop real-time video streaming transmissions, and compare the performance between the multicast and the unicast transmission mechanisms. To demonstrate and evaluate the system, we implement a prototype system using various brand handsets with Andorid OS and conduct experiments. The results show that the system can transmit real-time videos at the bitrate of 2200kbps and frame rate (fps) of 25 for 4-5 receivers simultaneously within 75 meters in case of single-hop transmission. As for performing multi-hop transmission, the real-time videos can be transmitted to 1 recevier at the bitrate of 5000kbps and fps of 25.

Device-to-Device (D2D) communication is considered as an efficient approach to improve current wireless networks. However, for inter-group communication, since devices are using the same IP address segment, IP conflicts can be induced into D2D networks. To address this problem, we propose an Information-Centric resource sharing system for D2D communication network. In the system, both P2P and WLAN network adaptor interfaces are adopted simultaneously for establishing a D2D network. A socket reusing mechanism is proposed for inter-group communication. In addition, to improve content distribution and retrieval, we base our system on Information-Centric networking (ICN) paradigm. We design ICN flow tables on network nodes to realize request forwarding and data backtracking. To demonstrate the feasibility and effectiveness, we have implemented the system on commercial Android devices using Wi-Fi Direct technology. Experimental results show that the proposed Information-Centric resource sharing system is feasible for D2D network in terms of data transmission and energy consumption. The average latency for transmitting 1MB data is 257.23ms; i.e., the throughput is up to 31.1 Mbps. The average energy consumption for transmitting 100MB data is only about 0.66 mAh.

The infrastructure-based networks will be unavailable in the case of infrastructure failures (e.g., earthquake and tsunami) or in crowded areas (e.g., concert and conference hall). This promotes the evolution of location communication systems, which also benefit offloading computing, mobile edge computing and mobile crowdsourcing. In this paper, we utilize Wi-Fi Direct (WFD) to develop a local communication system. We not only present an intra-group communication solution by the native characteristics of WFD, but also propose a general solution for the bidirectional inter-group communication, which is beyond the scope of WFD specifications. Experimental results show that the maximum throughput of intra-group communication could reach up to 31.7 Mbps, and the maximum throughput of inter-group communication is 8.26 Mbps.

Dynamic Host Configuration Protocol (DHCP) provides dynamic use of IP addresses, but it presents challenges to meet smart terminals with great mobility and transient network access patterns. Existing studies have tried to solve this problem through adjusting DHCP lease, which controls how long a host owns an address. However, few studies clearly express the relations among the lease, address utilization and DHCP overhead. In this paper, we uncover how the leases affect address utilization and DHCP overhead with two methods, based on which, we can set the leases for the smart terminals flexibly and judiciously. First of all, we present an emulation technique to evaluate address utilization and DHCP overhead under different leases. It provides an experimental basis for setting the lease for the whole WLAN. Evaluation results show that if the lease is set to 120 min instead of 60 min by default, it can reduce 41.78% DHCP overhead on average and still reserve at least 9.2% address space for the possibly emerging terminals. Then, we model the relationship between the lease and address utilization, as well as the relationship between the lease and DHCP overhead. According to these models, we propose a load-aware DHCP lease time optimization algorithm, which helps to set different leases for each area of the WLAN based on theoretical analysis. Evaluation results show that compared with the default lease for the whole WLAN, a lease combination of {15, 120, 120} for different areas can reduce 36.85% DHCP overhead on average and guarantee there is always 10% available address space.

Due to the centralized control, network-wide monitoring and flow-level scheduling of Software-Defined-Networking (SDN), it can be utilized to achieve Quality of Service (QoS) for cloud applications and services, such as voice over IP, video conference and online games, etc. However, most existing approaches stay at the QoS framework design and test level, while few works focus on studying the basic QoS techniques supported by SDN. In this paper, we enable SDN with QoS guaranteed abilities, which could provide end-to-end QoS routing for each cloud user service. First of all, we implement an application identification technique on SDN controller to determine required QoS levels for each application type. Then, we implement a queue scheduling technique on SDN switch. It queues the application flows into different queues and schedules the flows out of the queues with different priorities. At last, we evaluate the effectiveness of the proposed SDN-based QoS technique through an experimental analysis. Results show that when the output interface has sufficiently available bandwidth, the delay can be reduced by 28% on average. In addition, for the application flow with the highest priority, our methods can reduce 99.99% delay and increase 90.17% throughput on average when the output interface utilization approaches to the maximum bandwidth limitation.

One critical issue in SDN is to reduce the communication overhead between the switches and the controller. Such overhead is mainly caused by handling miss-match packets, because for each miss-match packet, a switch will send a request to the controller asking for forwarding rule. Existing approaches to address this problem generally need to deploy intermediate proxy or authority switches to hold rule copies, so as to reduce the number of requests sent to the controller. In this paper, we argue that using the intrinsic buffer in a SDN switch can also greatly reduce the communication overhead without using additional devices. If a switch buffers each miss-match packet, only a few header fields instead of the entire packet are required to be sent to the controller. Experiment results show that this can reduce 78.7% control traffic and 37% controller overhead at the cost of increasing only 5.6% switch overhead on average. If the proposed flow-granularity buffer mechanism is adopted, only one request message needs to be sent to the controller for a new flow with many arrival packets. Thus the control traffic and controller overhead can be further reduced by 64% and 35.7% respectively on average without increasing the switch overhead.

Pedestrian detection has become one of the hottest topics in intelligent traffic system because of its potential applications in driver assistance and automatic driving. In this study, a fast pedestrian detection and dynamic tracking method within vehicle‐to‐vehicle (V2V) cooperative environment is proposed. A dynamic tracking‐by‐detection framework for real‐time pedestrian detection is developed. First, a cascade classifiers, based on selected Haar‐like features, is trained to detect pedestrian. Then, CamShift algorithm combined with extended Kalman filtering is used to pedestrian dynamic tracking. Finally, with the crowdsourcing detected information, a smartphone‐based V2V cooperative warning system is developed to share useful detection results within blind spots. The experiment results show that the proposed method has a real‐time and accurate performance, which can provide a reference for road traffic safety monitoring technology.

In this paper, we design an LAP based routing algorithm in General Networks (GN) to solve the problem of information consistency of the full network under OSPF with the following operations: (i) decomposing GN into one or more Single-link Networks (SNs) with the approach of depth-first walk, (ii) re-composting the SNs to a network with regular topology structure by adding links, (iii) searching the undirected complete graph of three nodes round by round until it converges to a simple network topology based on region binding, and (iv) processing different converged network topologies with different LAP based routing algorithms. The proposed algorithm is compared with Dijkstra algorithm over some random network topologies. Simulation results show that the proposed algorithm can solve the problem of information consistency of the full network under OSPF and has better performance than Dijkstra algorithm.

With the prosperity of network applications, traffic classification serves as a crucial role in network management and malicious attack detection. The widely used encryption transmission protocols, such as the Secure Socket Layer/Transport Layer Security (SSL/TLS) protocols, leads to the failure of traditional payload-based classification methods. Existing methods for encrypted traffic classification suffer from low accuracy. In this paper, we propose a certificate-aware encrypted traffic classification method based on the Second-Order Markov Chain. We start by exploring reasons why existing methods not perform well, and make a novel observation that certificate packet length in SSL/TLS sessions contributes to application discrimination. To increase the diversity of application fingerprints, we develop a new model by incorporating the certificate packet length clustering into the Second-Order homogeneous Markov chains. Extensive evaluation results show that the proposed method lead to a 30% improvement on average compared with the state-of-the-art method, in terms of classification accuracy.

As the network infrastructure has been consuming more and more power, various schemes have been proposed to improve the power efficiency of network devices. Many schemes put links to sleep when idle and wake them up when needed. A presumption in these schemes, though, is that router's line cards can be waken up very quickly. However, through systematic measurement of a major vendor's high-end routers, we find that it takes minutes to get a line card ready under the current design. To address this issue, we propose a new line card design that 1) keeps the host processor in a line card standby, which only consumes a small fraction of power but will save considerable wakeup time, and 2) downloads a slim slot of popular prefixes with higher priority, so that the line card will be ready for forwarding most of the traffic much earlier. We design algorithms as well as architecture that ensure fast and correct longest prefix match during prioritized routing prefix download. Experiments on an FPGA-based prototype show that the customized hardware can be ready to forward packets in 127.27 ms, which is 0.3% of the time the original design takes. This can better support numerous power-saving schemes based on the sleep/wakeup mechanism.

Due to network dynamics, performance tuning is often indispensable in network management. In this paper, we propose MOE, a framework integrating network performance monitoring, optimization and evaluation. This is a trial towards the top-down and systematic management of network performance. We validate MOE based on a typical scenario in the real network environment. Results show that MOE can collect many kinds of network information, based on which it could conduct performance tuning automatically. In addition, MOE has the ability of evaluating the effect during and after performance tuning. Evaluation results are further analyzed and could fed back to provide positive advices to minimize the influence caused by network adjustments and maximize the performance profits.

DDoS flooding attack is one of the top threats to the Internet. However, due to the fast development of the Internet, current detection algorithms are already inadequate to meet the growth of network traffic. In this paper, we propose a lightweight algorithm. We first observe the real Internet traffic, and find that flows of DDoS flooding attack traffic are persistent and synchronous while most flows of normal traffic are short-lived and non- synchronous. According to this difference, we propose our detection algorithm. We label the alarms firstly and then confirm the attack. Our algorithm is lightweight and sensitive to the ongoing attack. However, randomly spoofing the IP address of the attack source to different IP addresses can hide the synchronization of attack flows. Thus, we add a spoofing IP detection algorithm called hop-count filter (HCF) to our algorithm to strengthen the robustness. At last, we evaluate our detection algorithm based on the real Internet traffic from CAIDA. Results show that our detection algorithm has a high accuracy (93.3%), no false positive in attack confirmation and just 1.1% false positive rate in labeling alarms. In addition, we analyze the challenges we may face when dealing with distributed LDoS attack.

In this paper, we gather the weekly reports of an operational IP backbone network from January 2006 to January 2013, according to which, we can restore the truth and uncover the evolution of network configurations of the studied network. Our high-level analyses illustrate that rate limiting and launching routes for new customers are most frequently configured. We can identify and construct configuration templates by correlating each task to a certain set of commands in configuration files, based on which, automated configuration provisioning for an operational backbone network is feasible. In addition, we can configure redundant links for those with higher rate of failures according to our detailed analyses of link failures, which will enhance the stability and reliability of data transmission.

Anomaly detection has been a hot topic in recent years due to its capability of detecting zero attacks. In this paper, we propose a new on-line anomaly detection method based on LMS algorithm. The basic idea of the LMS-based detector is to predict IGTE using IGFE, given the high linear correlation between them. Using the artificial synthetic data, it is shown that the LMS-based detector possesses strong detection capability, and its false positive rate is within acceptable scope.

As the network infrastructure has been consuming more and more power, various schemes have been proposed to improve power efficiency of network devices. Many schemes put links to sleep when idle and wake them up when needed. A presumption in these schemes, though, is that router's line cards can be waken up quickly. However, through systematic measurement of a major vender's high-end router, we find that it takes minutes to get a line card ready under the current implementation. To address this issue, we propose a new line card design that (1) keeps the host processor in a line card always up, which only consumes a small fraction of power, and (2) downloads a slim slot of popular prefixes with higher priority, so that the line card will be ready for forwarding most of the traffic much earlier. We design algorithms that ensure fast and correct longest prefix match lookup during prioritized routing prefix download. Experiments on real hardware show that the wakeup time can be reduced to 127.27ms, which is 0.3% of the original line card wakeup time, well supporting many power-saving schemes.

Anomaly detection has been a hot topic in recent years due to its capability of detecting zero day attacks. In this paper, we propose a new metric called Entropy-Ratio. We validate that the Entropy-Ratio is stationary. Making use of this observation, we combine the Least Mean Square algorithm and the Forward Linear Predictor to propose a new on-line detector called LMS-FLP detector. Using the two synthetic data sets - CEGI-6IX synthetic data and CERNET2 synthetic data, we validate that the LMS-FLP detector is very effective in detecting both anomalies involving many small IP flows and anomalies involving a few large IP flows.

As a crucial function of network management, network topology discovery provides a basis for lots of network analysis, such as network monitoring and performance management, etc. With the undergoing deployment of IPv6, the importance of precise topology discovery method in IPv6 networks becomes more and more evident. However, IPv6 network topology discovery faces new challenges due to different characteristics between IPv4 and IPv6, and the lack of well support of IPv6 related MIBs from device manufacturers in current state. At present, there are no well-accepted topology discovery methods for pure IPv6 networks with high accuracy, high coverage and less reliance on network configuration and device support. In this paper, we propose an IPv6 network topology discovery solution combining the advantages of two discovery methods, based on ICMP and routing protocol respectively. We model the mapping process of topology results from the two methods above into a graph mapping problem, which is the key point of the entire solution, and design novel mapping algorithms. We focus on the mapping coverage and accuracy and validate the mapping algorithms by large scale simulation. We also implement and test the proposed algorithms on the real network CERNET2. The experiments and simulation results verify the practicability and excellent performance of our solutions, with 100% discovery accuracy and over 99% discovery coverage while spending less time and producing lower overhead.

In this paper, many aspects related to characteristics and development of IPv6 network are investigated. Additionally, in order to gain a deep view of IPv6 network, we correlate our system with a user authentication system, so we explore some meaningful user behaviors. According to the analysis, we obtain a comprehensive knowledge of current operating situation of IPv6 network which, we believe, can provide an experimental basis for IPv6 network operators and researchers.

IPv6 protocol has been widely deployed in the world. As the IANA pool of IPv4 addresses has run out, IPv6 will become increasingly important. Although the IPv6 protocol stack presents considerable advantages compared with the IPv4 protocol stack, IP source address spoofing is still exploited in IPv6 to initiate malicious attacks. Some techniques are proposed and deployed to implement source address validation at fine granularity. In this paper, we investigate the efficiency of fine granularity IP source address validation, e.g. whether filtering technology is deployed to prevent hosts from using forged IP address. We develop a detection tool with controlled spoofing ability which can infer whether the function of filtering spoofing address packets is enabled. We run this tool in 12 famous universities in China and collect the testing data. We gather a total of 41373 probes from 324 clients, and each probe includes sending at least 5 packets with the same spoofing source address to the control server. Results reveal that, 77.02% of the spoofing probes are completely filtered, 0.29% of the spoofing probes are partly filtered and the rest spoofing probes are not filtered at all. Overall, this illustrates that techniques of source address validation have been widely deployed in campus networks. Our statistical results provide practical basis for the deployment and further development of source address validation protocols in IPv6 networks.