With the explosive growth of Artificial Internet of Things (AIoT) devices and the advent of 5G networks, efficient content distribution at the network edge has become critical to meet stringent low-latency and high-bandwidth demands. Traditional centralized caching schemes face challenges including limited edge cache capacity, heterogeneous user preferences, and increasing privacy concerns due to centralized data collection. To tackle these issues, this paper proposes a novel Recommendationenabled Edge Caching (ReEC) framework that integrates federated learning to collaboratively optimize caching policies without exposing raw user data, thus preserving privacy. We model the cache replacement process as a Markov Decision Process (MDP) and develop a Federated Two-Delay Deep Deterministic Policy Gradient (FTD3PG) algorithm. This algorithm can effectively solve the heterogeneous user request and privacy protection problems. Extensive experiments on the MovieLens dataset demonstrate that ReEC significantly improves cache hit rates, reduces content delivery latency, and enhances user Quality of Experience (QoE) compared to existing methods.

In IoT environments, devices are often constrained by low power consumption and limited resources, making efficient connection establishment with minimal overhead crucial for real-time communication between edge devices and cloud servers. The QUIC protocol shows significant potential, but the encryption and decryption overhead is considerable. Reducing this overhead and improving connection establishment efficiency are critical to enhancing QUIC performance, especially for IoT edge devices that need to handle high-concurrency communication while maintaining low power consumption. This paper proposes CFcoQUIC, a CPU/FPGA co-design architecture that accelerates the handshake process and reduces the initial connection latency by parallelizing multiple encryption/decryption flows in high-concurrency environments. Experimental results demonstrate that the time for RSA encryption and decryption on FPGA is at least 19.1 times faster than on CPU, and the time for AES encryption and decryption is at least 5 times faster on FPGA. These results highlight the effectiveness of the proposed architecture in reducing QUIC handshake overhead in IoT environments with low-power edge devices.

Deep Neural Networks(DNN) have achieved remarkable success in diverse applications. However, deploying and training large-scale DNNs in real-world heterogeneous device-edge environments remains challenging due to uneven resource distributions and communication bottlenecks. We propose HEMS, a hierarchical, resource-aware model splitting framework that jointly performs horizontal decoupling (between terminal and edge) and vertical decoupling (among edge nodes), enabling dynamic optimization of split points and task assignments under system constraints. A multi-agent reinforcement learning (MARL) algorithm based on MAPPO is introduced to learn adaptive partitioning and node collaboration policies, optimizing for latency, energy, and accuracy. It comprehensively considers computational and communication costs between terminals and edge nodes, aiming to minimize latency and energy consumption while maintaining training accuracy. Vertically, the method achieves efficient intra-layer parallel computations and promotes collaborative optimization among edge nodes. Experimental evaluations conducted in heterogeneous computational environments validate that this approach significantly enhances model training efficiency, effectively reduces overall system latency and energy consumption, and achieves optimized, high-efficiency model training.

With the rapid advancement of AI, deep neural network (DNN) have become core technologies. However, executing computation-intensive DNN tasks on mobile devices is challenging due to limited computational resources. Traditional cloud-assisted inference methods suffer from high network latency, restricting real-time performance. To overcome these issues, edge-end collaborative inference methods based on model partitioning have gained attention. Most research focuses on optimizing inference latency on the end device, with less emphasis on edge server efficiency, leading to underutilization of edge server resources in high-load scenarios. This paper presents a dynamic co-inference framework (DCF-DS) using critical soft actors and service level agreement (SLA)-aware task batching to improve DNN inference efficiency through collaboration between end devices and edge servers. A novel deep reinforcement learning method, Discrete Soft Actor Critic (DSAC), dynamically selects model partition points and data quantization levels based on a latency-accuracy reward, adapting to variable environments like wireless fluctuations and diverse CPU capabilities. Additionally, a task batching mechanism optimizes resource allocation and task scheduling on the edge server under SLA constraints. Experimental results demonstrate that DCF-DS outperforms existing methods on Nvidia Jetson TX2 NX and PC platforms, highlighting its advantages in collaborative inference between end devices and edge servers.

RDMA over Converged Ethernet v2 (RoCEv2) is one of the most popular high-speed datacenter networking solutions. Watermark is the general term for various trigger and release thresholds of RoCEv2 flow control protocols, and its reasonable configuration is an important factor affecting RoCEv2 performance. In this paper, we propose ByteTuning, a centralized watermark tuning system for RoCEv2. First, three real cases of network performance degradation caused by non-optimal or improper watermark configuration are reported, and the network performance results of different watermark configurations in three typical scenarios are traversed, indicating the necessity of watermark tuning. Then, based on the RDMA Fluid model, the influence of watermark on the RoCEv2 performance is modeled and evaluated. Next, the design of the ByteTuning is introduced, which includes three mechanisms. They are (1) using simulated annealing algorithm to make the real-time watermark converge to the near-optimal configuration, (2) using network telemetry to optimize the feedback overhead, (3) compressing the search space to improve the tuning efficiency. Finally, We validate the performance of ByteTuning in multiple real datacenter networking environments, and the results show that ByteTuning outperforms existing solutions.

Network emulator is an equipment used in the field of computer networking to replicate and simulate real-world network conditions, especially poor-quality network conditions accompanied by various damages, in a controlled environment. It plays a crucial role in the development, testing, and validation of various new network-related technologies, protocols, and applications. Compared with simulation and test-bed methods, network emulation possesses the advantages of accuracy and cost-efficiency. However, legacy network emulation methods are implemented serially, which are typically restricted in efficiency and waste computing resources. In this paper, we propose a combined network emulator, CombNE. To implement this emulator, we consider P4 programmable switches as a desirable option. CombNE consists of three logical components. First, CombNE provides a policy specification scheme to intuitively describe operator’s intents. Secondly, the CombNE parallelizer intelligently identifies the dependencies between network damages, automatically determines parallelization and generates a combination strategy. Third, CombNE will generate optimized P4 files and flow table information based on the combination strategy and deploy them to P4 programmable switches. Finally, we evaluated the performance and resources of CombNE.

In this paper, we address the problem of recommendation-enabled caching for cloud-edge-end collaborative networks. We leverage collaborative filtering techniques to predict user content preferences and construct a content recommendation matrix. To minimize content delivery delay, we model the cache replacement process as a MDP and propose an enhanced FDDPG algorithm for optimization. Experimental results demonstrate that our approach significantly reduces average delivery latency and improves the cache hit rate compared to existing methods.

Cyber-Physical-Social Systems (CPSS) aim to enhance societal efficiency by integrating intelligent, interactive systems. To support CPSS, data is often gathered from the cloud, but non-real-time cloud-user connections affect Quality-of-Service (QoS). To solve this, we propose offloading storage to the edge, caching popular content near users. We select cluster heads using local clustering, social intimacy, and betweenness centrality. Then, we group users using the Pairwise Constrained K-Means-Monotonic (PCKMM) algorithm based on distance, relationships, and content preferences. Using unsupervised recurrent federated learning (URFL), we predict content popularity and adjust clusters dynamically. Content is cached with a popularity-driven greedy strategy, reducing access latency.

While current Time-Sensitive Network (TSN) research focuses on reducing end-to-end delay through scheduling optimization, it often overlooks routing impacts. Traditional methods like Shortest Path First (SPF) struggle with new flow schedulability and existing flow disruption. This paper presents a TSN routing algorithm based on network calculus, calculating worst-case delays to ensure deadline compliance and balance link loads. A new flow prioritization method further optimizes scheduling. Experiments show the proposed algorithm improves scheduling success, keeps average delay under 70 μs, reduces maximum link load by 18.5% compared to wt-ECMP, and achieves a tenfold runtime reduction.

With the extensive application of network traffic encryption technology, the accurate and efficient classification of encrypted traffic has become a critical need for network management. Deep learning has become the predominant method for traffic classification, primarily involving the transformation of network traffic into grayscale images and their subsequent classification using Convolutional Neural Networks (CNNs). However, traditional grayscale image generation methods are plagued with issues of redundant and lost information, and conventional channel attention mechanisms are still insufficient in capturing key traffic features, collectively hindering the enhancement of classification performance. To tackle these issues, this paper introduces a classification method based on Byte Transfer Probability and Coordinate Attention Mechanism in Residual Network (BTP-CAResNet). This method, on the foundation of the classic ResNet architecture, incorporates a new grayscale image generation method that utilizes Byte Transfer Probability, effectively overcoming the deficiencies of traditional approaches. Additionally, this paper integrates a Coordinate Attention Mechanism into the ResNet model, which effectively overcomes the limitations of traditional channel attention mechanisms and further improves the performance of traffic classification. Experimental validation on the ISCX VPN-nonVPN dataset demonstrates that, compared to previous CNN-based methods, the method proposed in this paper exhibits superior performance in key metrics such as accuracy, precision, recall, and F1 score. It provides a new perspective for traffic classification based on convolutional neural networks.

Rapid development of the Internet has enabled explosive growth of various network traffic. How to classify and identify different categories of network traffic among these huge network traffic for cyberspace security has always been a hot research topic. In our study, we found that the composition structure of data frames and grayscale maps in the original traffic is very similar. Combined with recent research of deep learning in image processing, this paper proposes a 2D-CNN model-based network traffic recognition algorithm, while transforming traffic to grayscale maps for recognition. To validate the effectiveness of our proposed model, we use the public network dataset ISCX-VPN-NonVPN-2016 and USTC-TF2016. Experimental results prove that the average accuracy is 98.7% in regular encrypted traffic identification and 97.6% for malicious traffic identification. Our method provides new solutions for network traffic identification.

Network measurement provides rich data for network monitoring, control, and management. In-band network telemetry (INT) is a new network measurement technology that uses normal data packet to collect network information hop-by-hop. However, the design and implementation of INT protocol cannot do anything about packet loss: (1) The end-to-end telemetry mechanism makes INT unable to detect packet loss; (2) Since data packets may be lost due to various reasons, INT telemetry information will inevitably be lost. In summary, INT system by itself is unreliable. Incomplete telemetry data will seriously affect the performance of upper-layer network telemetry applications. In this paper, we present our successful experience in INT packet loss monitoring. We design, implement, and open source a powerful packet loss monitoring system for INT, called LossSight. The functions of LossSight include the detection of packet loss events, the deduction of the time and location of the losses, the diagnose of the root cause of the losses, and the recovery of the lost INT information. Experiment results show that LossSight provides excellent performance and extremely low overhead, including detection accuracy and diagnostic precision close to 100%, and detection latency of just milliseconds. In particular, LossSight uses a generative adversarial network to recover lost telemetry information, with excellent accuracy and reliability. LossSight has been running stably in the supercomputing interconnection environment of the National Supercomputing Center in Jinan. We suggest that all INT applications that require reliable telemetry information should be implemented based on LossSight.

The cloud-edge integration service model combines the advantages of computing capabilities both from cloud and edge. Therefore, the data centers with cloud-edge integrated are an irreversible trend for the evolution of future data center. Software-Defined Network (SDN), emerging as a novel network model, separates content forwarding and control, and that makes resource management across data center network more efficient. This article focuses on the network data transmission and management for future data centers. First, it reviews measurement, analysis, and monitoring methods meant for new features of global SDN network. Then it focuses on unified management of SDN resources like traffic scheduling theory for cross-domain data centers based on cloud. Specifically, we proposed a novel fault response mechanism across the network with a more precise location and less response time. With dynamic changes of cloud computing and edge computing services combined, global QoS control and QoE optimization methods are proposed correspondingly. Finally, a set of SDN control platforms supporting the functions mentioned above are formulated. We hope that our work will shed some new light and provide new theoretical support for cloud-edge-combined cross-domain data center network architecture.

Computer Supported Cooperation Work (CSCW) has been playing an increasingly important role in many areas of human social life. Cooperative design refers to the technique of product design based on CSCW and parallel engineering. With widespread use of CSCW, network bandwidth is becoming a bottleneck that affects user experience and service quality. Currently, global attention has been paid to the fifth-generation communication system (5G). In order to address the network bottleneck of the cooperative design system using the 5G network advantages, this paper focuses on optimizing QoE of the cooperative design system and proposes a distributed cache system for cooperative design in the 5G environment. The cache network is divided into different domains based on the characteristics of the 5G structure. Coupling between cache and cooperative design is implemented after taking the properties of the cooperative design system into account. Simulation results demonstrate the ability of the proposed system to considerably improve QoE of the cooperative design system and reduce bandwidth utilization.

Software-Defined Networking (SDN) is an emerging network architecture. SDN is currently attracting significant attention from both academia and industry. A large number of studies have been carried out in academic circles. However, how to build small scale experimental Software-Defined Networking is the basis of various researches. This paper builds a physical device called Xspider for building real SDN experimental environment based on NetFPGA with OpenFlow support. Xspider has the characteristics of saving space, being easy to carry, and the ability to simulate multiple topologies. This kind of physical device can facilitate the application of SDN teaching and experiment, which is beneficial to promote the technological progress of SDN.