Centralized Traffic Engineering (TE) plays a critical role in network management, owing to its potential to achieve optimal or near-optimal network performance. However, the overhead from frequent network-wide traffic measurements required to observe the global network view significantly limits its practical use. To address this issue, we propose an end-to-end learning approach called TEST for routing optimization using sparse Traffic Matrices (TMs) obtained from sparse traffic measurements, which require measuring traffic on only a subset of network nodes. Specifically, to mitigate the impact of unknown traffic demands in unmeasured nodes on network performance, we construct a set of synthesized TMs with diverse traffic patterns adaptively to enhance the robustness of the generated routing policies. To address the missed information in sparsely measured TMs, we leverage historical sparse TM sequences to provide sufficient evidence for generating routing policies. To effectively capture the spatio-temporal relationships in the sparse TM sequence, we propose designing a routing model by integrating a Transformer model with a Graph Convolutional Network (GCN). Extensive experiments conducted on topologies with different scales demonstrate that the proposed TEST achieves promising TE performance under sparse traffic measurements. Additionally, discussions on scenarios involving network failures and traffic changes further highlight its robustness.

Quantum networks, leveraging the principles of quantum mechanics, enable unconditionally secure communication and are emerging as a critical component of next-generation infrastructure. However, the scarcity of quantum network resources results in contention among requests for shared links and qubits. Such contention prolongs processing times and degrades resource utilization efficiency, which in turn leads to imbalanced completion times between high- and low-priority requests. This paper addresses the request scheduling problem by proposing DRL-QN, a scheduling framework based on Reinforcement Learning (RL) that optimizes scheduling sequences, reduces the completion time of network requests, and improves fairness. Specifically, to help the agent better perceive request contention and the network environment, request characteristics and network resources are described as the state, enabling accurate scheduling decisions. Furthermore, to ensure a balanced learning objective, the agent’s training is guided by a multi-objective reward function that simultaneously promotes low completion time and fairness. Extensive simulations demonstrate that the proposed DRL-QN method significantly reduces network completion time by 16.22%–21.25% compared with existing schemes, while achieving fairness comparable to the fairness-optimal method and improving by approximately 6.79%–13.61% over other approaches.

Routing optimization, as a significant part of Traffic Engineering (TE), plays an important role in balancing network traffic and improving quality of service . With the application of Machine Learning (ML) in various fields, many neural network-based routing optimization solutions have been proposed. However, most existing ML-based methods need to retrain the model when confronted with a network unseen during training, which incurs significant time overhead and response delay. To improve the generalization ability of the routing model, in this paper, we innovatively propose a routing optimization method GROM which combines Deep Reinforcement Learning (DRL) and Graph Neural Networks (GNN), to directly generate routing policies under different and unseen network topologies without retraining. Specifically, for handling different network topologies , we transform the traffic-splitting ratio into element-level output of GNN model . To make the DRL agent easier to converge and well generalize to unseen topologies, we discretize the huge continuous traffic-splitting action space. Extensive simulation results on five real-world network topologies demonstrate that GROM can rapidly generate routing policies under different network topologies and has superior generalization ability.

Network traffic classifiers of mobile devices are widely learned with federated learning(FL) for privacy preservation. Noisy labels commonly occur in each device and deteriorate the accuracy of the learned network traffic classifier. Existing noise elimination approaches attempt to solve this by detecting and removing noisy labeled data before training. However, they may lead to poor performance of the learned classifier, as the remaining traffic data in each device is few after noise removal. Motivated by the observation that the data feature of the noisy labeled traffic data is clean and the underlying true distribution of the noisy labeled data is statistically close to the clean traffic data, we propose to utilize the noisy labeled data by normalizing it to be close to the clean traffic data distribution. Specifically, we first formulate a distributionally robust federated network traffic classifier learning problem (DR-NTC) to jointly take the normalized traffic data and clean data into training. Then we specify the normalization function under Wasserstein distance to transform the noisy labeled traffic data into a certified robust region around the clean data distribution, and we reformulate the DR-NTC problem into an equivalent DR-NTC-W problem. Finally, we design a robust federated network traffic classifier learning algorithm, RFNTC, to solve the DR-NTC-W problem. Theoretical analysis shows the robustness guarantee of RFNTC. We evaluate the algorithm by training classifiers on a real-world dataset. Our experimental results show that RFNTC significantly improves the accuracy of the learned classifier by up to 1.05 times.

Traffic Engineering (TE) is an efficient technique to balance network flows and thus improves the performance of a hybrid Software Defined Network (SDN). Previous TE solutions mainly leverage heuristic algorithms to centrally optimize link weight setting or traffic splitting ratios under the static traffic demand. Note that as the network scale becomes larger and network management gains more complexity, it is notably that the centralized TE methods suffer from a high computation overhead and a long reaction time to optimize routing of flows when the network traffic demand dynamically fluctuates or network failures happen. To enable adaptive and efficient routing in distributed TE, we propose a Multi-agent Reinforcement Learning method CMRL that divides the routing optimization of a large network into multiple small-scale routing decision-making problems. To coordinate the multiple agents for achieving a global optimization goal in a hybrid SDN scenario, we construct a reasonable virtual environment to meet different routing constraints brought by legacy routers and SDN switches for training the routing agents. To train the routing agents for determining the local routing policies according to local network observations, we introduce the difference reward assignment mechanism for encouraging agents to cooperatively take optimal routing action. Extensive simulations conducted on the real traffic traces demonstrate the superiority of CMRL in improving TE performance, especially when traffic demands change or network failures happen.

Network traffic prediction plays a significant role in network management. Previous network traffic prediction methods mainly focus on the temporal relationship between network traffic, and used time series models to predict network traffic, ignoring the spatial information contained in traffic data. Therefore, the prediction accuracy is limited, especially in long-term prediction. To improve the prediction accuracy of the dynamic network traffic in the long term, we propose an Attention-based Spatial-Temporal Graph Network (ASTGN) model for network traffic prediction to better capture both the temporal and spatial relations between the network traffic. Specifically, in ASTGN, we exploit an encoder-decoder architecture, where the encoder encodes the input network traffic and the decoder outputs the predicted network traffic sequences, integrating the temporal and spatial information of the network traffic data through the Spatio-Temporal Embedding module. The experimental results demonstrate the superiority of our proposed method ASTGN in long-term prediction.

Network traffic classification is the foundation for many network security and network management applications. Recently, to preserve the privacy of the data which are generated in the mobile ends, federated learning (FL)-based classification methods are being proposed. Unfortunately, the performance of FL-based methods can seriously degrade when the client data have skewness. This is particularly true for mobile network traffic classification where the environments in the mobile ends are highly heterogeneous. In this article, we first conduct a measurement study on traffic classification accuracy through FL using real-world network traffic trace and we observe serious accuracy degradation due to heterogeneous environments. We propose a novel federated analytics (FA) approach, FEAT, to improve the accuracy. Note that FL emphasizes on model training, yet our FA performs local analytic tasks that can estimate traffic data skewness and select appropriate clients for FL model training. Our analytics tasks are performed locally and in a federated manner; thus, we preserve privacy as well. Our approach has strong theoretical properties where we exploit Hoeffding inequality to infer traffic data skewness and we leverage the Thompson Sampling for client selection. We evaluate our approach through extensive experiments using real-world traffic data sets QUIC and ISCX. The extensive experiments demonstrate that FEAT can improve traffic classification accuracy in heterogeneous environments.

Inter-DataCenter Wide Area Network (Inter-DC WAN) that connects geographically distributed data centers is becoming one of the most critical network infrastructures. Due to limited bandwidth and inevitable link failures, it is highly challenging to guarantee network availability for services, especially those with stringent bandwidth demands, over inter-DC WAN. We present$\mathsf {TEDAT}$, a novel Traffic Engineering (TE) framework for Diverse Availability Targets (DAT), where a Service Level Agreement (SLA) is defined to ensure that each bandwidth demand must be satisfied with a stipulated probability, when subjected to the network capacity and possible failures of the inter-DC WAN.$\mathsf {TEDAT}$has two core components, i.e., traffic scheduling and failure recovery, which are crystalized through different mathematical models and theoretically analyzed. They are also extensively compared against state-of-the-art TE schemes, using a testbed as well as real trace driven simulations across different topologies, traffic matrices and failure scenarios. Our evaluations show that, compared with the optimal admission strategy,$\mathsf {TEDAT}$can speed up the online admission control by$30\times $at the expense of less than 4% false rejections. On the other hand, compared with the latest TE schemes like FFC and TEAVAR,$\mathsf {TEDAT}$can meet the bandwidth availability SLAs for 23%~60% more demands under normal loads, and when network failure causes SLA violations, it can retain 10%~20% more profit under a pricing and refunding model.

Nowadays, a hybrid Software-Defined Network (hybrid SDN), which combines the robustness of the distributed network and the flexibility of the centralized network, is a prevailing network architecture. The performance of Traffic Engineering (TE) in a hybrid SDN is largely influenced by the location of SDN switches. To derive the optimal location of SDN switches, the previous SDN switches deployment strategies mainly focused on manually designing heuristics to search for the SDN deployment sequence and only take a static Traffic Matrix (TM) into consideration. However, the manually-designed heuristics cannot capture the complex intrinsic relations among the location of SDN switches, network topology, and dynamic traffic demands. In addition, the SDN deployment strategy optimized under a single TM exhibits poor performance in a dynamic environment. Therefore, in this paper, we propose a Deep Reinforcement Learning (DRL)-based algorithm SEED to intelligently learn the SDN deployment strategy under multiple TMs. Specifically, to capture the dynamic traffic information, we first cluster the historical traffic demand matrices for obtaining the representative Traffic Matrices (TM) that can depict the dynamic traffic. Then, to intelligently learn the intrinsic relations between the topology, TMs, and the location of SDN switches, we design a DRL agent under multiple TMs by interacting with the environment in a trial and error manner. The extensive experiments on three real network topologies and traffic demands demonstrate that our proposed SDN switches deployment strategy can better adapt to the dynamic traffic and better improve the TE performance than the other deployment strategies.

In modern datacenters, huge energy consumption is a significant problem that remains to be solved. Previous works reduce the system energy consumption by switching the idle servers to a low-power state. However, the workload demands on servers change dynamically and mainly depend on the real-time workload status. To maintain the system energy efficiency, when providing servers with some servers reserved, the status of workloads should be carefully considered. Generally, the status of workloads is characterized by some key factors sampled from the workloads. However, under dynamic workload demands, the accuracy of these sampled values varies. To accurately assess the status of server workloads, in this paper, we propose a Dynamic Time Scale based server Provision (DTSP) method, which takes the variability of workloads into consideration when providing servers for workload demands. To obtain accurate factor values indicating real-time workload, DTSP samples several key workload factors, including the coefficient of variation of arrival intervals, the request arrival rates of current workload and previous workload, and the mean service time of current requests, with a dynamic compatible rate. With these sampled factors, DTSP can accurately estimate the demands of workloads on servers and provide appropriate numbers of servers for the dynamic workloads. Extensive experiments demonstrate that taking the workload variability into consideration, DTSP can significantly promote the energy efficiency of a datacenter.

Segment Routing (SR) is a source routing paradigm that routes a packet through an ordered list of instructions called segments. It is widely used in Traffic Engineering (TE) because of its simplicity and scalability. Although there are lots of research about TE with SR (SR-TE), fewer consider network failures. The reactive approaches may suffer from latency and update issues, and the proactive approaches don't perform very well because the objectives aren't carefully designed. Besides, although different types of failures are considered, the failure probabilities are ignored. In this paper, we take failure probabilities in to consideration, and propose a proactive 2-SR model 2SRPF to handle SR-TE problem with network failures, aiming at minimizing maximum link utilization (MLU). Considering that severe failures are more noteworthy, we use probability as a severity threshold, and minimize the expectation of the larger MLUs whose corresponding failure states have probabilities sum to a specific threshold value. We solve it with probabilistic risk management. Experiments show that 2SRPF performs well with one threshold setting for different topologies consistently, and gets close to optimal results when network fails.

Inter-DataCenter Wide Area Network (Inter-DC WAN) that connects geographically distributed data centers is becoming one of the most critical network infrastructures. Due to limited bandwidth and inevitable link failures, it is highly challenging to guarantee network availability for services, especially those with stringent bandwidth demands, over inter-DC WAN. We present BATE, a novel Traffic Engineering (TE) framework for bandwidth availability (BA) provision, which aims to ensure that each bandwidth demand must be satisfied with a stipulated probability, when subjected to the network capacity and possible failures of the inter-DC WAN. The three core components of BATE, i.e., admission control, traffic scheduling and failure recovery, are formulated through different mathematical models and theoretically analyzed. They are also extensively compared against state-of-the-art TE schemes, using a testbed as well as real trace driven simulations across different topologies, traffic matrices and failure scenarios. Our evaluations show that, compared with the optimal admission strategy, BATE can speed up the online admission control by 30x at the expense of less than 4% false rejections. On the other hand, compared with the latest TE schemes like FFC and TEAVAR, BATE can meet the bandwidth availability targets for 23%~60% more demands under normal loads, and when network failure causes BA targets violations.

Segment Routing (SR) is a source routing paradigm which is widely used in Traffic Engineering (TE). By using SR, a node steers a packet through an ordered list of instructions called segments. By some extensions of interior gateway protocol, SR can be applied to IP/MPLS or IPv6 network without signal protocol. SR over IPv6 (SRv6) is attracting wide attention because of its interoperation ability with IPv6. However, upgrading the existing IPv6 network directly to a full SRv6 one can be difficult, because large-scale equipment replacement or software upgrade may cause economic and technical problems. TE in partially deployed SR network is becoming a hot research topic. In this paper, we propose the TE algorithm Weight Adjustment-SRTE (WA-SRTE) in partially deployed SRv6 network, in which SRv6 capable nodes are dispersedly deployed. Our objective is to minimize the network's maximum link utilization. WA-SRTE converts the TE problem into a Deep Reinforcement Learning problem and optimizes the OSPF weight, SRv6 node deployment and traffic paths simultaneously. Besides, traffic variation is also considered and we use a representative Traffic Matrix (TM) to epitomize the traffic characteristics over a period of time. Experiments demonstrate that with 20% to 40% of the SRv6 nodes deployed, we can achieve TE performance as good as in a full SR network for the experiment topologies. The results with WA remarkably outperform the results without it. Our algorithm also gets near-optimal results with changing traffic.

Traffic matrix (TM) is a critical information for network operation and management, especially for traffic engineering (TE). Due to the technical and mercantile problems, real time measurement for TM is difficult in large scale networks. In this paper, we focus on predicting TM for dynamic traffic engineering. We propose several TM prediction methods based on Neural Networks (NN) and predict TM from three perspectives: predict the overall TM directly, predict each origin-destination (OD) flow separately and predict the overall TM combined with key element correction. In addition to the prediction accuracy, we evaluate different prediction methods through the performance of TE, as well as the prediction time. We test the proposed methods by real world datasets from Abilene, CERNET and GÉANT. The experiment results show that prediction methods based on Recurrent Neural Networks (RNN) can achieve better prediction accuracy than methods leveraging Convolutional Neural Networks (CNN) and Deep Belief Networks (DBN). Predicting each OD flow through RNN models can further improve the prediction accuracy, as well as the performance of TE under the OSPF network scenario, the SDN/OSPF hybrid network scenario and the multi-commodity flow problem scenario. However, it takes longer prediction time for predicting each OD flow sequence. In contrast, predicting the overall TM combined with key element correction can provide a trade-off between the TE results and the prediction overhead, which is more appropriate for dynamic TE in the above three scenarios.

Traffic Engineering is an efficient tool to balance the network flows and, thus improving the network performance with limited network resources. The goal of traffic engineering is to find an efficient and robust routing to balance the flows with changing traffic. Multiple traffic matrices are good representatives of the changing traffic. The emergence of Software Defined Networking (SDN) provides us a more flexible way to route the network flows with multiple traffic matrices. We expect to optimize the routing of the average case performance over multiple traffic matrices and, at the same time, bound the worst case performance for some unexpected traffic in hybrid SDN network. In this paper, we first formulate the problem of optimizing both the average case and worst case performance of the routing over multiple traffic matrices. Then, we prove the problem is NP-hard and propose a heuristic algorithm to solve it. Finally, we evaluate our algorithm with real traffic datasets. Through extensive experiments, we observe that the worst case performance of our routing can be dramatically improved by 32.15% with a little sacrifice of the average case performance by 2.01% and demonstrate the effectiveness of our algorithm in optimizing both the average case and worst case performance of routing.

With the rapidly growing of geo-distributed applications in the Internet, there is a huge amount of data generated in geo-distributed datacenters everyday. However, because of region privacy concerns and limitation of inter-DC WAN bandwidth, moving all the geo-distributed data to a single datacenter for centralized processing is not practical. Therefore, we intend to process the data where it generates by the big data applications and optimize the coflow routing in inter-DC WAN to improve the performance of the applications. Previous studies consider only routing or task placement optimization, which is inefficient. In this paper, we propose an algorithm PRO with an approximation ratio (1+\epsilon) that jointly optimize the placement of tasks and the routing of a coflow. Our proposed algorithm can efficiently reduce the coflow completion time.

With the deployment of big data applications, more and more data are stored in the online storage. Erasure coding storage system has been widely used by companies such as Google and Facebook, since it provides space-optimal data redundancy to protect against data loss. In erasure coding storage system, (n, k) MDS erasure code is used to divide file into n chunks. When a user want to access the file, any subset of k out of n chunks will be needed to reconstruct the file. In this case, how to select k out of n chunks and how to let the chunks transfer quickly become important problems. In this paper, we joint the two problems together to optimize. Our optimization goal is to minimize average file access time (FAT). To achieve this, we propose smallest load first heuristic to do source selection and design an online algorithm to reduce chunks transfer latency. Base on this, we design and implement D-Target, a centralized scheduler that tries to minimize average FAT in distributed erasure coding storage system. We then test D-Target's performance by trace-driven simulation. Results show that, for the trace of AT&T, D-Target performs 2.5×, 1.7×, 1.8×, 3.6× better than TCP, Aalo, Barrat and pFabric respectively.

Nowadays, there are two major challenges to detect traffic anomalies in a large scale network. One is how to handle huge amounts of traffic data when we detect traffic anomalies in a network, and the other is how to carry out fast and detailed detection and classification. To address these two challenges, we propose a Change based Effective Frequent flow Features approach (CEFF), which can quickly obtain the anomaly detection and classification results by scanning the flow data only once. We implement CEFF for both offline and online detection and classification in Spark, a popular big data processing platform. Besides, we evaluate CEFF using China Telecom NetFlow format data in experiments, and make comparisons between CEFF and Shannon entropy based method, which has been proved to be effective for traffic anomaly detection. The experiment results show that CEFF has excellent performance in traffic anomaly detection and classification.

Data Center is now becoming an important facility for many applications (e.g, web search and retail). As TCP can't meet applications' demands for latency and throughput, many tcp-based protocols (e.g, DCTCP, D2TCP, L2DCT) have been proposed. Among them, protocols such as D2TCP incorporate explicit deadline into congestion window adjustment procedure to guarantee flows' latency and protocols such as L2DCT consider flow size when computing congestion window adjustment factor to guarantee the throughput of short flows. These two methods work well at some scenery but they have some deficiencies on two aspects. Firstly, we find that they can only reduce the percentage of flows missing deadline or reduce flow completion time, but can not meet both the goals simultaneously. Secondly, most of these methods need the user to know flow information (e.g, deadline, flow size), which may be hard to know exact value beforehand. In this paper, we advocate to use flow duration time into congestion window adjustment procedure. Based on this, we propose FDRC-Flow Duration Time based Rate Control algorithm. We find that without knowing flow information beforehand, FDRC can achieve the goal of reducing the percentage of flows missing deadline and cutting average flow completion time simultaneously. We theoretically analyze FDRC's behavior and implement FDRC into ns-2 as well as linux kernel. Our experiments show that FDRC performs better than D2TCP and L2DCT at nearly all the scenarios. On average, it performs 30% better than the state-of-art deadline-aware congestion control protocol D2TCP and 10% better than the state-of-art flowsize-aware protocol L2DCT.

Traffic engineering is a method to balance the flows and optimize the routing in the network. Software defined networking is a new network architecture and we can gain great benefit by migrating the traditional IP network to the SDN-enabled network from the perspective of traffic engineering. However, due to the economical, organizational and technical challenges, migrating to the network with a full deployment of SDN routers is impractical in the short term. It is a desirable choice to deploy SDN incrementally. In this paper, we seek to search for an optimal migration sequence of the legacy routers to SDN-enabled routers so that we can decide where and how many routers to migrate firstly. Our main contribution is that we propose a heuristic algorithm, i.e., genetic algorithm, to seek a migration sequence of the routers that obtains the most of the benefit from the perspective of traffic engineering. We evaluate the algorithm by conducting simulation experiments, making comparison to the greedy migration algorithm and static migration algorithms that we propose. The experiments exhibit that the genetic algorithm, outperforms the other migration algorithms in searching for a migration sequence. When properly deployed, about a migration of 40% of routers reaps most of the benefit.

Today network traffic anomaly detection is very challenging in a big and constantly changing network, because there are millions of flows being transferred in a network at the same time, and the flow numbers change all the time. Although traditional information entropy has been proved to be an effective metric on network traffic anomaly detection, such a metric shows some limitations in large scale networks with constantly changing flow numbers, and it makes the traditional entropy inefficient for traffic anomaly detection. Another challenge is how to process large-scale traffic data in a scalable way. In this paper, we propose Adjustable Piecewise Entropy for traffic anomaly detection, and implement Adjustable Piecewise Shannon entropy in Hadoop platform with a cluster of five servers in Tsinghua University Campus Network. Furthermore, we analyze and validate Adjustable Piecewise Entropy in both mathematics and experiments. The experiment results show that Adjustable Piecewise Entropy has better performance for traffic anomaly detection.

Traffic engineering under OSPF routes along the shortest paths, which may cause network congestion. Software Defined Networking (SDN) is an emerging network architecture which exerts a separation between the control plane and the data plane. The SDN controller can centrally control the network state through modifying the flow tables maintained by routers. Network operators can flexibly split arbitrary flows to outgoing links through the deployment of the SDN. However, SDN has its own challenges of full deployment, which makes the full deployment of SDN difficult in the short term. In this paper, we explore the traffic engineering in a SDN/OSPF hybrid network. In our scenario, the OSPF weights and flow splitting ratio of the SDN nodes can both be changed. The controller can arbitrarily split the flows coming into the SDN nodes. The regular nodes still run OSPF. Our contribution is that we propose a novel algorithm called SOTE that can obtain a lower maximum link utilization. We reap a greater benefit compared with the results of the OSPF network and the SDN/OSPF hybrid network with fixed weight setting. We also find that when only 30% of the SDN nodes are deployed, we can obtain a near optimal performance.