最新刊期
-
Resource-incentivized-based asynchronous task scheduling for cross-regional cloud-edge collaboration
摘要：To address the collaborative challenges caused by node heterogeneity, traffic bursts, and network latency uncertainty in cross-regional cloud-edge multi-data center systems (MDCS), this paper proposes an asynchronous task scheduling method based on a Resource Incentive Stackelberg Game (RISG). First, M/M/1 queuing theory is utilized to characterize the non-linear congestion effects of heterogeneous nodes under high concurrency, a comprehensive cost model integrating energy consumption, latency, transmission overhead, and reliability risk is constructed, and a closed-form solution for the optimal response of data centers is derived. Second, a gradient-based Asynchronous Coordinate Descent (ACD) algorithm is designed, which supports the global scheduler in performing non-blocking policy updates using stale information, overcoming the inefficiency of synchronous waiting, and the convergence of the algorithm as well as the existence and uniqueness of the equilibrium are proven. Simulation results show that the convergence speed of the ACD algorithm is improved by approximately 26.9% compared to the traditional asynchronous gradient projection algorithm; in heavy-load system scenarios, the average response latency is reduced by approximately 45.17% compared to greedy strategies; in environments with strong network jitter, the total social cost fluctuation error is controlled within 8%, verifying the high throughput and strong robustness of this method.关键词：multi-data center system;asynchronous task scheduling;leader–follower game5|1|0更新时间：2026-05-11 - 摘要：With the advancement of artificial intelligence of things (AIoT) technology and the enhancement of hardware performance, convolutional neural network (CNN) models were increasingly introduced for inference in collaborative cloud-edge-end scenarios to satisfy real-time requirements of intelligent AI applications. To address the efficient utilization of heterogeneous computing resources including CPUs, GPUs, NPUs, and FPGAs, an innovative task scheduling system was proposed to support high-concurrency CNN model inference. The DeepSets model was employed to extract features from tasks and environmental constraints, while the generalization capability of the Proximal Policy Optimization (PPO) algorithm was enhanced to achieve task scheduling decisions. Precise and efficient task allocation was ensured in complex dynamic environments through this approach. The proposed system was deployed on Kubernetes and Kosmos platforms, with comprehensive validation conducted across various heterogeneous processing units, including high-performance servers, NVIDIA Jetson Xavier NX, and NVIDIA Jetson Nano devices. Experimental results demonstrate that compared to existing scheduling algorithms, the proposed system improves performance by 275.93% and increases resource utilization by 38.86%, significantly enhancing the efficiency and reliability of cloud-edge computing frameworks and providing robust support for intelligent AI applications.关键词：cloud-edge-end collaboration;high concurrency;CNN;DeepSets;PPO generalization;heterogeneous resources;task scheduling1|0|0更新时间：2026-05-11
- 摘要：With the in-depth development of the global digital economy, the demand for cross-border data circulation has surged, while traditional models are confronted with critical challenges such as lack of trust, security risks and regulatory barriers. To address the demand for trusted, efficient and secure cross-border data flow, this paper proposes a tokenization-based cross-border trusted data space. By integrating core tokenization technologies with data space technologies, this space enables trusted sharing, secure control and compliant circulation of cross-border data, thereby establishing an integrated cross-border data governance ecosystem. Its technical system and system construction scheme can provide important reference for the development of cross-border data governance ecosystems.关键词：tokenization;cross-border trusted data space;Data circulation;cross-border data compliance1|0|0更新时间：2026-05-11
- 摘要：To address the excessive storage overhead of neuron-level fault-tolerance methods on resource-constrained edge devices, this paper proposes a neuron-level threshold compression method based on Orthogonal Variable Spreading Factor (OVSF) basis codes. In the proposed approach, activation thresholds are represented as linear combinations of a small number of orthogonal OVSF basis codes. Only sparse coefficients and their indices are stored, while full thresholds are reconstructed on-the-fly during inference through a lightweight online reconstruction mechanism. This design significantly reduces threshold storage and transmission overhead without introducing substantial computational cost. Experimental results on representative DNN models, including AlexNet and VGG16, show that when the threshold fidelity score (TFS) is maintained above 94%, the proposed method achieves fault-tolerance performance comparable to the uncompressed FitAct scheme. Under this condition, the threshold storage overhead can be reduced by 50%–90%, depending on the selected compression ratio. Furthermore, FPGA-based edge experiments verify the feasibility and efficiency of the proposed method under bandwidth-constrained deployment scenarios. These results indicate that the proposed approach provides a practical and resource-efficient solution for deploying low-overhead fault-tolerant neural networks on edge devices.关键词：Deep Neural Networks (DNN);Fault Tolerance;Orthogonal Variable Spreading Factor (OVSF);model compression;Hardware Transient Faults3|0|0更新时间：2026-05-11
- 摘要：In the communication industry, ensuring a high level of customer satisfaction is the key to maintaining competitive advantage. Customer satisfaction prediction has low accuracy due to the scarcity of samples or incomplete data. At the same time, the interpretability of model prediction results is poor, which cannot provide effective data support for differentiated care for dissatisfied customers. This article proposes a customer satisfaction prediction method that combines zero sample learning. The auxiliary features extracted by ETS and t-SNE provide necessary support for zero sample learning, enabling the model to maintain high prediction accuracy without a large amount of labeled data. At the same time, SHAP is used to clearly demonstrate the contribution of each feature to the prediction results, helping enterprises identify and locate key factors of customer dissatisfaction. Collecting customer satisfaction data from a certain province in Southeast China to validate the model, the results showed that the accuracy of customer satisfaction prediction in this paper improved by nearly 9%, and the success rate of customer care based on explanatory analysis of dissatisfied customers reached over 81% .关键词：customer satisfaction prediction;Interpretability analysis;ETS decomposition;T-SNE dimensionality reduction1|0|0更新时间：2026-05-11
- 摘要：With the evolution of large model architectures towards the sparse Mixture-of-Experts (MoE), the proportion of communication overhead in end-to-end latency was observed to rise significantly in both training and inference scenarios, and communication performance gradually became a critical factor constraining system performance. To address challenges such as heavy All-to-All communication pressure, sensitivity to bandwidth and latency, and surging operational complexity in large-scale MoE training and inference scenarios, a high-performance network infrastructure solution based on hardware-software co-design was proposed in this paper. First, at the architecture level, the Astral 3.0 network architecture was designed by utilizing Optical Shuffle technology to construct a flattened two-layer single-rail network. This architecture was adapted to the All-to-All traffic characteristics of MoE, significantly improving communication performance and reducing networking costs. Second, at the communication software level, targeted All-to-All communication kernel optimizations were performed based on the distinct traffic characteristics of various stages in training and inference. By utilizing GPU-centric task dispatch technology and expert-granularity load balancing technology, high-bandwidth kernels adapted for training and Prefill stages, as well as low-latency kernels adapted for the Decode stage, were implemented, which drastically reduced end-to-end latency. Finally, at the operations level, network system operational workflows were comprehensively optimized using AI Agents, and proactive fault warning along with intelligent interactive diagnosis were achieved, ensuring the continuity of long-term training and the high availability of online services. Experimental results demonstrated that the communication wall in MoE models was effectively broken by this solution, providing a unified, high-performance, and highly reliable system foundation for the large-scale training and online service of trillion-parameter models.关键词：Mixture of Experts (MoE);Astral 3.0 Network Architecture;All-to-All Communication;AIOps;Inference System1|0|0更新时间：2026-05-11
- 摘要：As video applications rapidly advance toward intelligence and wireless capabilities, the open and dynamic nature of wireless networks poses severe challenges for high-quality video transmission, making it difficult to meet the stringent demands of low latency and high reliability required by critical applications such as digital security and industrial surveillance. To this end, this paper proposes a wireless video cross-layer transmission and bitrate adaptation scheme based on joint network and service perception. First, a cross-layer transmission optimization mechanism was designed to integrate wireless network congestion state awareness with video encoding control, achieving coordinated scheduling of video transmission and network resources, significantly improving bandwidth utilization efficiency; Secondly, an end-to-end bitrate adaptation method is proposed, which achieves collaborative bitrate adjustment between the playback and device ends through full-link wireless state sensing and joint control of encoding/decoding parameters, ensuring smooth video playback. The experimental results demonstrate that this solution effectively enhances the real-time performance and stability of video transmission in dynamic wireless environments, making it well-suited for high-demand intelligent service scenarios under the current trend of transitioning from wired to emerging wireless transmission technologies like 5G and Wi-Fi 6.关键词：Wireless video transmission;Cross-layer transmission optimization;Bitrate adaptation1|0|0更新时间：2026-05-11
- 摘要：As data center energy consumption and decarbonization pressures continue to intensify, achieving cost-effective Direct Green Power Connection while maintaining high reliability has become a critical priority. To quantify the marginal cost patterns along a macro-level transition in green electricity penetration (GEP) from low to high levels, a source–grid–load–storage coordinated optimization model integrating wind–solar resource profiles, time-of-use tariffs, and demand-charge mechanisms was developed. A multi-scenario sensitivity analysis approach was used, and the levelized cost of electricity (LCOE) for 31 provincial-level administrative regions under GEP levels from 0% to 100% was systematically evaluated. The results showed that: (1) cost trajectories are jointly anchored by the “resource–tariff” interplay, with tariff structures exerting a more pronounced influence in the low-GEP range; (2) a “storage cost wall” emerges at high GEP levels, where the rigid capital investment required to smooth long-cycle variability rises exponentially; and (3) green power potential is spatially misaligned—resource-rich regions such as Inner Mongolia exhibit an absolute cost advantage, whereas load centers such as Jiangsu deliver higher relative benefits. These findings underscored the indispensable role of grid flexibility in enabling high-GEP pathways and provide quantitative evidence to support the “East Data, West Computing” initiative and differentiated policy design.关键词：data center;direct green power connection;energy storage cost wall;source-grid-load-storage coordination0|0|0更新时间：2026-05-11
- 摘要：To address the challenges of inter-cell co-channel interference and rapid channel variations in complex 5G network environments, a randomized resource block group (RBG) allocation method for the physical downlink shared channel (PDSCH) is proposed, and its performance boundaries are theoretically established. By randomizing the frequency-domain resource positions across different cells, the proposed method effectively avoids resource overlapping and significantly suppresses inter-cell interference. Meanwhile, through uniform frequency-domain scheduling for user equipments (UEs), it ensures stable service quality for users at various locations, thereby alleviating experience disparities caused by uneven resource distribution. System-level simulation results demonstrate that the proposed method improves the average cell throughput by more than 4% and the cell-edge user throughput by more than 16% under medium and low load scenarios.关键词：physical downlink shared channel (PDSCH);resource block group (RBG);randomized allocation;throughput1|0|0更新时间：2026-05-11
- 摘要：To address the challenge that existing heterogeneous graph neural network models face in simultaneously handling missing node attributes and limited label availability in heterogeneous graphs, this paper proposes a novel heterogeneous graph neural network model that integrates attribute completion with hierarchical contrastive learning. The model first employs an encoder-decoder architecture to achieve end-to-end reconstruction of missing node attributes. Next, it encodes target nodes by capturing both low-order structural neighborhoods and high-order meta-path semantics, where slot-enhanced aggregation is introduced in the low-order representation to reduce type-related semantic ambiguity. To align multi-scale representations effectively, hierarchical contrastive objectives are established within the meta-path space and across low-order and high-order views, enabling collaborative representation learning. Moreover, a fine-grained semantic-attribute dual-aware strategy is applied for positive sample selection to enhance the quality of contrastive signals, leading to highly discriminative node embeddings. Extensive experiments on multiple public datasets, compared against various baseline models, demonstrate that the proposed method achieves average improvements of 0.35% to 1.99% in Macro-F1 score for node classification and 1.2% to 2.1% in Normalized Mutual Information (NMI) for node clustering, thereby validating its effectiveness.关键词：heterogeneous graph;graph neural network;attribute completion;graph contrastive learning;attention mechanism2|0|0更新时间：2026-05-11
- 摘要：To address the issue of low recognition accuracy in multi-UAV flight scenarios, a multi-domain feature fusion recognition method based on video transmission signals was proposed. The video transmission signal, characterized by its continuity, high bandwidth, and strong hardware dependence, was considered a more suitable stable signal source. To overcome the limitations of a single feature, short-time Fourier transform (STFT) and cyclic spectral analysis (CSA) were integrated for multi-domain modeling. STFT was employed to extract hardware fingerprints, such as power amplifier nonlinearity and local oscillator frequency offset, from time-frequency dynamics. CSA was utilized to mine spectral correlation features, including I/Q mismatch and carrier leakage, by leveraging the cyclostationarity of the signal, thereby comprehensively characterizing subtle identity differences. To fully exploit the complex nonlinear correlations within the high-dimensional multi-domain radio frequency fingerprint representation formed by the fused features, an Effiv2KAN model (EfficientNetV2-based Network with KAN Classifier) was designed. In this model, EfficientNetV2 was adopted at the front end for multi-scale feature extraction from the fused features, and a KAN classifier was employed at the back end. By replacing fixed activation functions with learnable spline functions, the model flexibly fit the complex nonlinear relationships among features, enhancing the discriminative capability for subtle fingerprints. Experimental results demonstrated that the proposed model outperformed classical deep learning models under different signal-to-noise ratios and exhibited stronger noise-robustness for discriminating UAV models with highly similar signals.关键词：UAV RF Fingerprint;Video Transmission Signal;Time-Frequency and Cyclic Spectral Feature Fusion;Effiv2KAN Model0|0|0更新时间：2026-05-11
- 摘要：By decoupling the inference process of the Mixture of Experts (MoE) model into a computationally intensive prefill phase and a memory-intensive decode phase, and deploying them on distributed physical computing nodes respectively, the efficient operation of inference systems is achieved, and the model inference efficiency is improved. This paper elaborates on the inference process of the MoE model in detail. Based on the inference processes of the Attention layer and the MoE layer, it models the online inference system for both single inference tasks and batch inference tasks. Through calculating the computation and transmission latency of the prefill phase and the decode phase, the throughput is obtained, aiming to achieve throughput balance between the two phases. A resource configuration and strategy deployment mechanism based on the binary search algorithm is proposed to determine the computation resource ratio, the number of deployed instances, and the parallel strategy for each phase. Experimental verification is conducted on two mainstream computing nodes, with comparisons made against both non-PD-disaggregation baseline and state-of-the-art PD disaggregation inference optimization approaches. Experimental results show that compared with non-PD-disaggregation baseline approach, PD disaggregation inference achieves a throughput improvement of more than 3x. It still outperforms state-of-the-art approach in performance. Additionally, the mechanism proposed in this paper can simplify the manual configuration and tuning process, finding near-optimal PD disaggregation deployment decisions under different conditions of input/output lengths, concurrency numbers, and request frequencies. Compared with other feasible PD resource allocation decisions, the average throughput per card is increased by 30-50%.关键词：Mixture-of-Experts Model;Prefill and Decode Disaggregation Inference;Online Inference System;Model Deployment Optimization0|0|0更新时间：2026-05-11
- 摘要：Telecom complaint adjudication involves multiple stages. Traditional manual processing faces clear bottlenecks in efficiency and consistency, while existing automation studies mostly focus on isolated tasks. To address this issue, this paper proposes an intelligent adjudication framework based on large language models. Through prompt engineering, the framework comprises five modules: invalid complaint detection, complaint type identification, complaint content decomposition, human-in-the-loop evidence chain collection, and adjudication report generation. Experiments on approximately 7,000 real complaint cases from a provincial telecom operator show that the proposed method achieves 83.2% accuracy in invalid complaint detection, outperforming BERT and other baselines by at least 14.2%; 97% accuracy in service issue classification; and an F1 score of 73.9% with 82.1% key-point coverage in complaint content decomposition. Processing efficiency across stages improves by up to 96.7% compared with manual handling. Cross-month generalization tests further demonstrate stable performance. The proposed framework provides a complete intelligent solution for telecom complaint adjudication and offers practical value for digital transformation in the telecom industry.关键词：Telecom Complaint;large language models;prompt engineering;Human-Machine Collaboration2|0|0更新时间：2026-05-11
- 摘要：To address the copyright protection requirements of high-precision images in smart education, a robust blind watermarking algorithm integrating an adaptive attention mechanism and the SimSiam self-supervised contrastive learning framework was proposed. The adaptive attention mechanism dynamically allocated embedding weights in high-frequency texture regions, thereby enhancing robustness against attacks such as JPEG compression and random cropping while maintaining high image quality. The SimSiam module strengthened feature consistency through contrastive learning, effectively suppressing attack-induced variations and improving the semantic stability of the watermark. Experimental results demonstrated that the proposed method achieved a 100% bit recovery rate (BRR) under JPEG compression (quality factor 50) and 50% random cropping, with a 15.4% improvement in PSNR compared with the traditional DCT-based method. This study provides a copyright protection solution for educational images that balances robustness and visual fidelity in smart education scenarios.关键词：Image Watermarking;smart education;Adaptive Attention;SimSiam;Robust Feature Embedding1|0|0更新时间：2026-05-11
- 摘要：As 5G network construction progresses into the stage of refined operation, the quality of 5G coverage in urban residential areas has become a critical indicator influencing user experience. To overcome the limitations of traditional evaluation methods—such as insufficient multi-dimensional correlation analysis, low problem localization accuracy, and poor adaptability to refined planning requirements—this paper proposes a 5G coverage evaluation model for urban residential areas based on geospatial big data fusion. The model integrates multi-source heterogeneous information, including building points of interest (POI), cell-level network management data, and grid-level user perception data, and constructs four collaborative sub-models to enable full-process evaluation ranging from scenario segmentation to investment priority quantification. The LambdaMART algorithm is introduced as the core ranking mechanism, effectively supporting precise value ranking under the fusion of multi-dimensional indicators. Experimental results demonstrate that the model achieves a weak coverage identification accuracy of 92.5%, and its decision quality is highly consistent with expert judgment (NDCG>0.9). The proposed model significantly improves the efficiency and accuracy of coverage evaluation, providing scientific decision support for the refined planning and optimization of 5G networks.关键词：5G;Urban Residential Areas;Coverage Assessment;Multi-Source Heterogeneous Data Integration;Geospatial Big Data;Value Prioritization;machine learning0|0|0更新时间：2026-05-11
- 摘要：To address the high complexity of carrier recovery algorithms in wireless communication systems and their poor real-time performance when executed on CPUs (Central Processing Units), a parallel high-speed carrier recovery algorithm based on GPU (Graphics Processing Unit) is proposed. In wireless communication systems, the combination of a coarse frequency offset estimation algorithm based on FFT and a fine frequency offset estimation algorithm based on a decision-directed phase-locked loop (PLL) constitutes a typical carrier recovery approach. However, considering that the feedback structure inherent in the PLL is not well suited for parallel execution on GPUs, a frequency sweeping algorithm is proposed to replace the decision-directed PLL algorithm, thereby leveraging the powerful parallel processing capability of GPUs to accelerate the carrier recovery process. On this basis, within the Compute Unified Device Architecture (CUDA) platform, the execution speed of the carrier recovery algorithm is further improved by optimizing the frequency sweeping algorithm and its associated parallel reduction algorithm implemented in CUDA. Simulation results demonstrate that the proposed GPU-based parallel carrier recovery algorithm outperforms the CPU-based serial carrier recovery algorithm in terms of both throughput and algorithmic performance. For different modulation schemes, the throughput can be improved by 20 to over 40 times, making it more suitable for wireless communication systems in high-speed scenarios.关键词：Carrier Recovery;parallel computing;Graphics Processing Unit;Compute Unified Device Architecture2|0|0更新时间：2026-05-11
- 摘要：With the evolution of wireless communication systems toward 5G-Advanced and 6G, network environments are becoming increasingly complex and dynamic, making data-driven approaches and artificial intelligence essential for intelligent wireless systems. However, wireless communication data exhibit continuous streams, strong temporal variations, and high scenario dependency, resulting in high costs for acquiring and maintaining high-quality labeled data, which has become a critical bottleneck for practical deployment. Focusing on the problem of data annotation in wireless communications, this paper systematically reviews the main data modalities, label forms, and ground-truth acquisition mechanisms, and analyzes the sources of annotation noise and data distribution drift. We further survey several representative learning paradigms, including active learning, weak supervision, semi-supervised learning, self-supervised learning, noise-robust learning, generative models, and large-model-based approaches, and discuss their advantages and limitations in reducing annotation cost and improving label quality across typical applications such as spectrum monitoring, interference identification, modulation recognition, and Wi-Fi sensing. Finally, future research directions are outlined toward 6G and AI-RAN, including foundation models, continual learning, and privacy-aware annotation, aiming to provide insights for building low-cost and reliable annotation pipelines for intelligent wireless systems.关键词：Wireless communications;data annotation;artificial intelligence;weak supervision learning;active learning;self-supervised learning0|0|0更新时间：2026-05-11
- 摘要：To address the insufficient exploitation of directional information and the limited recognition accuracy in RSSI-based wireless sensing, an event directional detection algorithm for single-link RSSI signals was proposed based on multi-view time-frequency feature fusion and a whale optimization algorithm (WOA)-optimized support vector machine (SVM). By exploiting the propagation characteristics of Fresnel zones, the transmitter and receiver were orthogonally deployed to form strong and weak detection regions, thereby enhancing directional information in single-link RSSI signals. RSSI event waveforms were then segmented, and statistical features in both the time and frequency domains were extracted from the complete waveform and its leading and trailing segments to construct multi-view fused directional feature vectors. Subsequently, WOA was employed to optimize the penalty factor and kernel function parameters of the SVM, and a WOA-SVM directional recognition model was established for experimental validation. The results demonstrated that the proposed method outperforms BP, CNN, and unoptimized SVM models in terms of accuracy, precision, and F1-score, with an accuracy of 91.42% and an F1-score of 91.48%.关键词：RSSI;wireless sensing;fresnel zone model;motion direction;SVM;WOA0|0|0更新时间：2026-05-11
- 摘要：Cloud resource scheduling in large-scale data centers faces severe challenges characterized by high concurrency, dynamic workload volatility, and conflicting optimization objectives (e.g., tradeoffs between makespan, energy consumption, and resource utilization). Traditional heuristics often lack adaptability, while single-agent Deep Reinforcement Learning (DRL) approaches suffer from the "curse of dimensionality" when scaling to large state spaces. To address these limitations, this paper proposes HGT-MARL-CS-PSO, a novel framework that orchestrates scheduling through a hierarchical game-theoretic approach. First, we formulate the scheduling problem as a Two-Level Stackelberg Game: a high-level Manager Agent (M-Agent) acts as the Leader, defining global strategic goals based on macro-workload patterns, while low-level Executor Agents (E-Agents) act as Followers, engaging in a non-cooperative Nash game to balance local resource competition with global cooperation. Second, to mitigate the explosion of discrete action spaces, we introduce a Hybrid Parameterized Control Mechanism. Instead of generating task mappings directly, the RL agents output continuous weight parameter vectors to dynamically configure a subordinate Cuckoo Search-Particle Swarm Optimization (CS-PSO) scheduler for precise micro-execution. Theoretical analysis validates the convergence of the proposed framework towards a Stackelberg Equilibrium. Extensive experiments using real-world Google Borg Traces demonstrate that HGT-MARL-CS-PSO significantly outperforms state-of-the-art DRL baselines and meta-heuristics in terms of makespan reduction, energy efficiency, and resource utilization, proving its superior generalization and robustness in complex environments.关键词：cloud resource scheduling;Multi-Agent Reinforcement Learning (MARL);Stackelberg Game;Parameterized Control1|0|0更新时间：2026-05-11
- 摘要：To address the high resource scheduling cost and low resource utilization caused by the imbalance of computing resource supply and demand across regions and significant electricity price differences, the total energy consumption cost involved in task computation and network transmission is formulated as a mixed-integer linear programming (MILP) problem. Based on this model, a Cost-minimized Cross-Regional Traffic Scheduling Algorithm (CM-CRTSA) was proposed. In this algorithm, a bipartite derivative flow graph with super source and sink nodes was constructed to achieve optimal matching between service requests and data centers using the minimum-cost maximum-flow method. During path selection, both energy cost and spectrum status were considered to rank candidate paths. In the spectrum allocation stage, a contiguity metric was introduced to prioritize available spectrum blocks. Simulation results demonstrated that CM-CRTSA effectively reduced the total scheduling cost and request blocking ratio, and maintained stable resource scheduling performance under high load, providing an efficient solution for cross-regional computing-network coordinated scheduling in computing optical networks.关键词：computing power networks;optical networks;traffic scheduling;cost optimization8|0|0更新时间：2026-05-11
- Technical support is provided by Beijing Founder electronics co., LTD 京ICP备09064830号-19
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
- Total Visits：1045027 Visits Today：1793
0