In scenarios such as distributed collaborative training of AI large models in cloud data center networks, different types of data flows have differentiated Quality of Service (QoS) requirements for the network. To meet these differentiated requirements, it is necessary to first identify the types of data flows. Currently, there are two major challenges: one is the classification of encrypted flows. Traditional methods based on ports and DPI (Deep Packet Inspection) will fail, and intelligent recognition methods using machine learning need to be adopted. The second is the challenge of online real-time classification. Most existing machine learning-based traffic classification methods consume a lot of resources and have a slow speed, are limited to offline processing, and are difficult to cope with online real-time processing. Therefore, this paper proposes a lightweight encrypted traffic classification method based on deep learning. This method designs a lightweight multi-scale feature fusion module (Lightweight Multi-scale Atrous Spatial Pyramid Pooling, LM-ASPP), which enhances multi-scale feature extraction capabilities by using dilated convolution and depthwise separable convolution techniques while reducing computational overhead. Based on ShuffleNetV2, a lightweight encrypted traffic classification model integrating LM-ASPP and ShuffleNetV2 is designed. Experimental results show that this model achieves high accuracy while significantly reducing resource consumption and inference latency, achieving a balance between performance and resource consumption.

Many applications, such as AI training and distributed storage, rely on cross-datacenter (DC) networks to provide services. For compatibility with existing RDMA hardware and to improve quality of service, network providers connect to datacenters over dedicated lines. However, the current RDMA congestion control has some problems in cross-DC environment. First, due to the large bandwidth delay product (BDP) of cross-DC flows, the switch frequently triggers PFC, which impairs the transmission of all flows. Meanwhile, due to the lag of congestion signals, congestion control algorithms may cause unfair bandwidth allocation between intra-DC flows and cross-DC flows. In addition, cross-DC traffic will experience severe queuing at the data center interconnect (DCI) switch, increasing queuing delay. To address these challenges, this paper proposes MLCC, a cross-datacenter congestion control algorithm based on the fast control loop. MLCC uses micro congestion control loops to achieve fine-grained network state awareness and accurate rate adaptation, and reduce queue length in the transmission path. Experimental results show that MLCC can quickly converge all flows to fairness, achieve high link utilization, and ensure low queue length on the switch. Large-scale simulations show that MLCC can reduce the average FCT of intra-datacenter and cross-datacenter traffic by up to 46% and 27%, respectively.

Network management and security are significant for the development of the Internet of Things (IoT), The traditional traffic identification and detection methods face growing challenges with the widespread deployment of IoT devices, the extensive use of encryption technologies, and the continuous increase in malicious network traffic. These challenges further complicate traffic monitoring and malicious behavior detection in network management. Consequently, research is increasingly focusing on deep learning techniques to tackle these issues. Although new traffic classification models based on deep learning methods such as CNNs and Transformers have been introduced, they still encounter issues like insufficient feature extraction and poor performance with similar traffic types. To address these challenges, this paper introduces a malicious traffic classification method based on 1D ConvNeXt and multi-scale patches. This novel approach applies 1D ConvNeXt to malicious traffic classification, integrating 1D depth wise sep-arable convolution and multi-scale patches to capture spatio-temporal traffic features. Our method achieves 99.56% and 99.33% accuracy on the USTC-TFC2016 and CICIoT2022 datasets, respectively, significantly reducing sample confusion among highly similar traffic types. The results show that our approach outperforms the other baseline models.

Real-time video (RTV) systems place high demands on ultra low-latency (i.e., less than 100 ms). However, our large-scale measurements reveal that a significant portion of users still experience high video frame latency due to bandwidth jitters. Existing solutions attempt to mitigate this issue by lowering the sender's future video frame encoding bitrate. Nevertheless, as shown in our controlled experiments, they fail to drain existing packets queued on the bottleneck node (i.e., the 5G base station and Wi-Fi access point), still suffering from high tail latency as bandwidth decreases. In this paper, we propose Co-RTV, a collaborative RTV system that achieves predictable latency control. Specifically, Co-RTV enables endpoint-network collaboration between the bottleneck node and the sender. The collaboration speeds up the release of packets queued at the bottleneck node and facilitates accurate latency control at the RTV sender through scalable QoE-driven flow control. Extensive experiments in emulated networks and on a 5G testbed demonstrate the superior performance of Co-RTV, with tail latency reductions of 69.1% and 70.5%, respectively.

Reliable low-latency media streaming is increasingly critical for delivering seamless interactive and immersive services. To meet this need, the IETF and 3GPP have introduced the Low Latency, Low Loss, Scalable Throughput (L4S) architecture, which mitigates queuing delays in IP traffic and supports latency-sensitive applications. This paper focuses on real-time video streaming and proposes an enhanced framework, Enhanced L4S (EL4S), which integrates a link load factor feedback mechanism to more accurately reflect real-time network conditions. We design and implement a cross-layer end-side transmission control algorithm based on EL4S to improve real-time video performance over 5G networks. In this design, EL4S encodes link load information into packets at the 5G core and uses ACK-based feedback to inform the sender about current network states. At the sender, a dynamic bitrate adaptation algorithm adjusts the transmission rate in response to the reported link load factor, balancing throughput and delay while avoiding network congestion. This adaptive mechanism enables precise, frame-level control over video encoding rates and transmission behavior. Extensive experimental evaluations demonstrate that EL4S achieves high link utilization and low latency, significantly outperforming existing baseline algorithms. Overall, EL4S provides an efficient and deployable solution for achieving low-latency, high-quality real-time streaming in dynamic 5G networks.

The computing power network(CPN), as a current research hotspot, aims to provide users with reliable, efficient, and secure computing capabilities. However, research on the routing algorithm for computing power network is limited. Additionally, existing computing power networks are mainly applied in general domains, lacking improvement methods tailored for special scenarios such as emergency communications, which are prone to vulnerabilities and fluctuations. In this paper, we propose a routing algorithm for computing power network, namely Graph Attention Q-Network(GAQN), which integrates GAT and DQN to optimize routing strategies with the objective of maximizing overall system throughput. To evaluate the performance of GAQN, we conduct comparative experiments with multiple baseline models across various network topologies. Additionally, we perform experiments on topologies with different numbers of node failures. The results demonstrate that the proposed algorithm outperforms the baseline algorithms overall and performs well in disrupted network topologies, proving its generalization and robustness.

Access networks, even with advanced 5G technology, often face bottlenecks when supporting concurrent real-time Ultra High Definition (UHD) video streams with high bandwidth and low latency (e.g., under 10 ms of one-way delay) requirements. Traditionally, end systems employ a combination of flow and congestion control mechanisms to control the sending rate to avoid overwhelming the receiver and the network. However, such control efforts induce prolonged tail delays, thereby sharply reducing the number of UHD video streams meeting delivery deadlines, and sometimes even zero. These outcomes are largely due to the inaccurate network status estimation associated with the control mechanisms. To address this challenge, we propose CFC, a deadline-oriented flow control mechanism that employs cross-layer status estimation to maximize user satisfaction with deadlines. CFC accurately assesses cross-layer information, including flow status and 5G access network status at minimal expense, thus ensuring the deadlines through effective concurrent flow control. Our experiments, conducted in both simulation and testbed settings, demonstrate significant improvements in delay and load-balancing for both reliable and unreliable transmissions.

Cloud virtual reality (VR) has emerged as a promising technology, offering users a highly immersive and easily accessible experience. However, concurrent pulse VR flows can lead to significant congestion in 5G base stations, making network providers unable to guarantee the delay requirements for all users. Based on a comprehensive analysis of the poor delay per-formance of cloudVR flows within the existing 5G edge network, we propose a novel cross-layer congestion control mechanism that is assisted by access network status and flow characteristics. This mechanism is deployed within the 5G edge network and enables efficient global scheduling of concurrent flows. Experiment results show that our mechanism greatly optimizes the network delay in concurrent scenarios and guarantees the delay requirements of all users while avoiding network overload. Our work underscores the advantage of leveraging 5G edge nodes as a valuable resource to meet the anticipated demands of future services effectively.

VR applications that require extremely low latency and high image quality are widely used in online games and other 5G scenarios, becoming a key research field in recent years. However, the limited bandwidth in 5G edge networks fails to meet the peak rate requirements for multiple VR flows. MPTCP is suitable for 5G edge networks, supporting the simultaneous use of multiple networks on mobile devices. Nevertheless, accurately scheduling VR data blocks to different sub flows to satisfy their low latency requirements is challenging due to their micro-burst characteristic. In this paper, we propose a novel MPTCP scheduler for cloud VR applications in 5G edge networks, called the Cross-Layer Information-based One-Way Delay Predictive Scheduler (CPS). CPS accurately predicts one-way delay by incorporating cross-layer information from both the application and edge wireless sides, and adaptively schedules VR data blocks to the optimal subflow. Experimental results show that CPS outperforms existing strategies, supporting 125% more users for VR applications in the typical scenario. Additionally, CPS maintains completion times for 99% of cloud VR packets below 7 ms. CPS successfully meets the quality of experience needs of more users, providing a promising solution for large-scale deployment of cloud VR services in 5G edge networks.

The pulse pattern generated by real-time cloud VR applications poses a new challenge to bandwidth-limited wireless access networks. Concurrent pulse flows can lead to significant congestion in access points, making network providers unable to guarantee the quality of experience (QoE) for all users. To maximize the number of satisfied users, we propose a cross-layer assisted traffic scheduling mechanism, CTS, to manage flows into the edge network. We evaluated CTS using a trace-driven simulator, and the experimental results prove that CTS achieves the theoretical maximum of satisfied users and maintains a more balanced load for the network.

The millimeter wave technology which provides the throughput of multi-gigabit per second is one of the key technologies for 5G/B5G communications. However, an optimal interaction between the transport layer protocols and the highly fluctuating millimeter wave networks is extremely challenging and lacks in-depth exploration in actual networks. In this paper, we examine and discuss the performance of several TCP congestion control algorithms in the real 60 GHz millimeter wave environment, and inspect the improvement of TCP performance over mmWave hybrid networks by TCP proxies in single-flow and multi-flows scenarios. Our results reveal severe adaptation problems associated with these congestion control algorithms over millimeter wave networks and the effectiveness of TCP proxies for the utilization of millimeter wave hybrid networks.

Traffic classification has always been a hot research topic, which can be used in network performance optimization, security management and some other scenarios. There have been a lot of high-performance classifiers in network classification domain, but nearly all these classifiers only focus on the overall accuracy. There exists tremendous traffic volume gaps among various network applications, which causes extreme imbalanced datasets when applying some artificial intelligence (AI) approaches to classify the traffic into categories or specific applications. The most intractable problem caused by training on imbalanced dataset is that even though the classifier misclassifies categories in rather small sample scale, the overall classification accuracy can be still quite high. This issue is intolerant if the minority category is vital but in small scale. To solve this problem mentioned above, we propose a self-defined loss function UniLoss, which greatly improves the classification accuracy of minority categories and maintains the performance of majorities meanwhile. VoIP traffic is representative for its imbalanced distribution, and thus VoIP traffic is chosen as the test instance. In addition, we design four deep neural networks and construct four test cases with different dramatic imbalanced category sample distributions, on which the results have verified the effectiveness of UniLoss.

In-network caching is a key feature of information-centric networking (ICN), in which routers take charge of caching passing contents. Such cache design enables efficient content distribution, but the benefit comes at a non-trivial cost, given adding workload to routers. With the emergence of edge computing, more edge devices with large storages are available, which is a good opportunity for new caching design. To this end, we propose a novel Edge-oriented Collaborative Caching (ECC) in ICN. In ECC, edge devices (such as edge server, micro datacenter, etc.) cache contents while routers only maintain cache indexes which are used to redirect subsequent requests towards the cached content. This enables ECC to work in a lightweight and collaborative fashion. We propose an optimization method to properly create cache indexes, considering both content popularity and cache benefit. Moreover, we also present a failure recovery mechanism to ensure system robustness. Simulation results show when deploying the same total cache capacity, ECC outperforms typical ICN caching schemes in terms of response latency, server load and bandwidth consumption of links.

Network traffic classification is an important research topic for computer network, such as QoS detection and admission monitoring. Traditional classification methods, such as port-based and DPI(deep packet inspect)-based, are out-of-date due to the computational expensiveness and inaccuracy. In this paper, we propose a novel traffic classification approach based on packet transport layer payload by ensemble learning. We use three kinds of base neural networks to form a strong classifier. Each model is trained separately and the final prediction result is decided by weight voting. The raw traffic data are reshaped into the format of sequence and matrix as the input, which avoids the TCP stream feature selection and extraction process. Our approach is applicable to both TCP and UDP, which means that it doesn't require a distinction between transport layer protocols. The experiment results show that our approach can reach the high accuracy of 96.38%, and is better than the state-of-the-art methods based on the same dataset. Besides, our proposed model can select packet samples randomly avoiding tracing the whole stream and the model works well even there's packet loss and disorder.

In-networking caching, as one of the critical features of the Information-Centric Networking (ICN), has been widely investigated. On-path caching is a light-weight implementation of in-networking caching. Existing studies on on-path caching strategy are mainly confined to the local information and neglect the interaction between multi-node cache decisions, which incurs cache redundancy and lead to performance degradation. To this end, we analyze the relationship between multi-node decisions and propose a lightweight on-path cooperative caching strategy, called GAC. In GAC, caching decisions are divided into two phases. In the first round, the upstream node makes caching decision based on the downstream caching information. In the second round, according to the feedback from upstream nodes, each node is given an opportunity to optimize the caching decision it made before. We evaluate the performance of our scheme through extensive simulations regarding a wide range of performance metrics. The experimental results indicate GAC can achieve significant performance improvement compared with representative approaches in terms of server load reduction ratio, average hop reduction ratio and average cache hit ratio. Meanwhile, GAC can significantly reduce the number of cache evictions.

Ethernet plays an important role in the layer 2 network. Unfortunately, the tremendous Address Resolution Protocol (ARP) broadcast traffic among massive hosts limits the scale of Ethernet. Recently, Software-Defined Network (SDN) has been proposed to suppress broadcast traffic by centralized control. However, existing approaches based on SDN suffer from an adaptability limitation as they cannot independently obtain ARP table entries. In this paper, we propose SDARP, a Software-Defined Address Resolution Proxy, to suppress broadcast traffic by centrally processing all ARP packets. To overcome the adaptability limitation, SDARP centrally obtains and maintains ARP table entries by independently resolving the header of ARP messages. SDARP is a SDN application. We prototype SDARP based on the open-source SDN controller RYU, and conduct experiments on the Mininet-based virtual testbed. The emulation results demonstrate that SDARP is transparent to hosts, effectively reduces ARP traffic to 7.1%, eliminates the broadcast storm and reduces the response time of the Internet Control Message Protocol to 35.9%.

Named Data Networking (NDN) is currently a hot research topic in the field of network architecture, and its transport control mechanism is one of the key technologies needed to be studied. Since the transport in NDN network has the characteristic of multi-source, the implicit congestion detection mechanism of the traditional TCP protocol is no longer suitable for the NDN network. In this paper, we propose a novel congestion control protocol for NDN network based on explicit feedback - ECP (Explicit Control Protocol), which detects the condition of network congestion proactively, and sends explicit feedback to the receiver. According to the feedback, the receiver can adjust the sending rate of Interests in order to control the sending rate of Datas from the sender, thus to realize the congestion control of the network. The simulation results based on NdnSIM show that the ECP protocol performs higher transfer efficiency and stability compared to the current NDN transport protocol using TCP implicit detection mechanism.

Path load balancing is used for distributing workload across an array of paths to increase network reliability and optimize link utilization. However, it is not easy to realize the load balancing globally in traditional networks as the whole status of the network is difficult to obtain. To address this problem, we propose the Fuzzy Synthetic Evaluation Mechanism (FSEM), a path load balancing solution based on Software Defined Networking (SDN). In this mechanism, the network traffic is allocated to the paths operated by Open Flow switches, where the flow-handling rules are installed by the central SDN controller. The paths can be dynamically adjusted with the aid of FSEM according to the global view of the network. Experimental results verify that the proposed solution can effectively balance the traffic and avoid unexpected breakdown caused by link failure. The overall network performance is also improved as well.

The transmission control protocol (TCP) has contributed to the tremendous success of the Internet but it also faces many challenges which are becoming more and more significant as the network grows. Although numerous congestion control algorithms have been proposed to improve the performance of TCP in heterogeneous networks, designing a congestion control algorithm that could achieve high utilisation, ensure fairness and maintain stability remains a great challenge. A novel congestion control algorithm named fair TCP (FTCP) has been proposed to mitigate these challenges. FTCP mitigates these challenges through the following strategies: First, increase the round trip time (RTT)‐fairness by altering TCP's initial congestion control window (cwnd) and adjusting the cwnd's growth rate to make FTCP flows with different RTTs achieve the same throughput. Secondly, balance the transmission efficiency and TCP‐friendliness by dynamically adjusting the aggressiveness of FTCP according to the congestion level of the link. Preliminary experimental evaluations verify that FTCP has obvious advantages in transmission efficiency, RTT‐fairness and TCP‐friendliness comparing to the state‐of‐the‐art congestion control algorithms.

This paper analyzes the TCP incast problem in data centers by focusing on the relationships between the TCP throughput and the congestion control window size of TCP. The root causes of the TCP incast problem are explored and the essence of the current methods to smooth the TCP incast is well explained. To verify our analysis, extensive simulations are conducted. The simulation results are in conformity with the estimated results of our analysis, which verifies the accuracy of our analysis. The analysis as well as the simulation results are helpful for the improvement of the TCP incast problem.

The routing scalability and IP address exhaustion are two significant issues the current Internet faces. The "locator/identifier (Loc/ID) split" has become a well recognized design principle for future Internet architectures that make Internet routing more scalable. In this paper, a novel Loc/ID split routing and addressing architecture called IPv4+6 is proposed. It not only solves the routing scalability problem but also expands the IP address space. It is easy to deploy, which only needs to make simple changes on DNS and gateway router.

Most of existing transfer protocols performs poorly for transferring bulk data over fast long distance network. A key reason is the congestion control algorithm. This paper designs a novel congestion control algorithm, namely congestion detection and rate adaptation (CDRA) algorithm, which adjusts sending rate according to the congestion level of terminal in order to maximize transfer performance. For implementation, a modified protocol based on the UDP-based protocol Tsunami, called robust Tsunami (RTsunami), was developed. The experimental results show that the CDRA algorithm is efficient and the RTsunami protocol outperforms Tsunami.