A method and system for managing network traffic scheduling of a large bandwidth

Abstract

The application is suitable for a large-bandwidth network flow scheduling management method and system, the method comprising: acquiring flow restriction parameters and determining maximum network bandwidth; monitoring the demand of each network port, sampling the bandwidth demand of each network port according to a preset sampling demand monitoring frequency, and generating bandwidth demand coordinates; constructing a network port bandwidth demand prediction function corresponding to each network port, synthesizing a comprehensive demand prediction function based on the network port bandwidth demand prediction function; predicting the bandwidth within a preset time length according to the comprehensive demand prediction function, obtaining prediction bandwidth data, and allocating network bandwidth to each network port. The application determines the sampling data of each network port by high-frequency sampling, constructs a prediction function of each network port based on the sampling data, predicts the bandwidth change within a short time according to the prediction function, and schedules the bandwidth of the network port according to the prediction, so as to fully utilize the 95 billing rules and save network cost.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of flow scheduling, and particularly relates to a large-bandwidth network flow scheduling management method and system. BACKGROUND

[0002] Large-bandwidth network flow scheduling management refers to a management and optimization mechanism designed for high-capacity data transmission requirements in network communication. It ensures that different types of traffic are effectively processed and transmitted according to their priority and requirements through intelligent analysis and dynamic allocation of network resources. This process involves using advanced algorithms and technologies to monitor network status, predict traffic trends, and adjust routing and bandwidth allocation in real time, thereby ensuring efficient operation and service quality of the network. The purpose is to fully utilize bandwidth resources while reducing latency and avoiding congestion, providing users with stable and fast network experience.

[0003] The existing flow calculation method adopts the 95 billing principle, and the existing flow management method cannot fully utilize the bandwidth, each network port is independently metered, and it is impossible to reduce the settlement cost. SUMMARY

[0004] The purpose of the present application is to provide a large-bandwidth network flow scheduling management method, which aims to solve the problem that the existing flow calculation method adopts the 95 billing principle, and the existing flow management method cannot fully utilize the bandwidth, each network port is independently metered, and it is impossible to reduce the settlement cost.

[0005] The present application is implemented as follows: a large-bandwidth network flow scheduling management method, the method comprising:

[0006] Obtaining flow restriction parameters to determine the maximum network bandwidth;

[0007] Monitoring the demand of each network port, sampling the bandwidth demand of each network port according to a preset sampling demand monitoring frequency, generating the bandwidth demand coordinates corresponding to each network port, the horizontal coordinate of the bandwidth demand coordinates being the time value, and the vertical coordinate being the sampling bandwidth;

[0008] Constructing a network port bandwidth demand prediction function corresponding to each network port based on the bandwidth demand coordinates of each network port, and synthesizing a comprehensive demand prediction function based on the network port bandwidth demand prediction function;

[0009] Predicting the bandwidth within a preset time length according to the comprehensive demand prediction function to obtain prediction bandwidth data, and allocating network bandwidth to each network port based on the prediction bandwidth data.

[0010] Another purpose of the present application is to provide a large-bandwidth network flow scheduling management system, which is applied to the above-mentioned large-bandwidth network flow scheduling management method, and comprises:

[0011] Parameter acquisition module, used to obtain traffic limit parameters and determine the maximum network bandwidth;

[0012] The bandwidth demand coordinate generation module is used to monitor the demand of each network port, sample the bandwidth demand of each network port according to the preset sampling demand monitoring frequency, and generate the bandwidth demand coordinate corresponding to each network port. The horizontal axis of the bandwidth demand coordinate is the time value, and the vertical axis is the sampling bandwidth;

[0013] A prediction function construction module is used to construct a network port bandwidth demand prediction function corresponding to each network port based on the bandwidth demand coordinates of each network port, and synthesize a comprehensive demand prediction function based on the network port bandwidth demand prediction function;

[0014] The bandwidth allocation module is used to predict the bandwidth within a preset time length according to the comprehensive demand prediction function, obtain predicted bandwidth data, and allocate network bandwidth to each network port based on the predicted bandwidth data.

[0015] The present invention provides a large-bandwidth network traffic scheduling and management method. By performing high-frequency sampling on each network port to determine the sampling data of each network port, a prediction function is constructed for each network port based on the sampling data. The bandwidth change in a short period of time is predicted according to the prediction function. The bandwidth of the network port is scheduled according to the prediction, so as to fully utilize the 95 billing rule and save network costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A flow chart of a large bandwidth network traffic scheduling and management method provided by an embodiment of the present invention;

[0017] Figure 2 This is an architectural diagram of a high-bandwidth network traffic scheduling and management system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0019] like Figure 1 As shown, a flowchart of a large bandwidth network traffic scheduling management method provided by an embodiment of the present invention includes:

[0020] S100: Obtain traffic limit parameters and determine the maximum network bandwidth.

[0021] In this step, traffic limit parameters are obtained. When the user and the operator establish a network subscription contract, a maximum network bandwidth is negotiated. The maximum network bandwidth can also be a bandwidth limit set by the user, that is, the maximum transmission speed between the user and the operator. The maximum transmission speed is the maximum network bandwidth. The network bandwidth between the user and the operator is allowed to fluctuate between 0 and the maximum network bandwidth. Finally, the 95th billing principle is used to determine the network traffic fee generated by the user. The 95th billing rule is: for example, there are 100 billing values. First, these 100 billing values ​​are arranged from small to large, and the 95th billing value is taken as the billing settlement value. The reason is that 5% of the billing values ​​are discarded because they are floating traffic, which may be attacked in the middle. Therefore, the billing is more humane and ignores 5% of the peak traffic.

[0022] S200 , monitor the demand of each network port, sample the bandwidth demand of each network port according to a preset sampling demand monitoring frequency, and generate bandwidth demand coordinates corresponding to each network port, where the horizontal axis of the bandwidth demand coordinate is the time value and the vertical axis is the sampling bandwidth.

[0023] In this step, demand monitoring is performed on each network port, and information on the transmission status of each network port is collected. The frequency of collection is greater than the operator's frequency of collecting network bandwidth usage. For example, a millisecond-level sampling frequency is used for collection, and traffic conditions are collected once at a preset time interval to determine the bandwidth demand corresponding to that moment, so as to obtain the network bandwidth demand of each network port at that moment. Bandwidth demand coordinates are constructed based on the collected data. The bandwidth demand coordinates are used to record the bandwidth demand at each moment. Each sampling of each network port will obtain a set of bandwidth requirements to form a set of bandwidth demand coordinates.

[0024] S300 , constructing a network port bandwidth demand prediction function corresponding to each network port based on the bandwidth demand coordinates of each network port, and synthesizing a comprehensive demand prediction function based on the network port bandwidth demand prediction functions.

[0025] In this step, a network port bandwidth demand prediction function corresponding to each network port is constructed based on the bandwidth demand coordinates of each network port. For each network port, a corresponding number of bandwidth demand coordinates corresponding to the network port are retrieved. Using a preset function fitting software, the retrieved bandwidth demand coordinates are uniformly imported into the function fitting software. The function fitting software is used to perform fitting to obtain a network port bandwidth demand prediction function corresponding to the network port. Based on the network port bandwidth demand prediction function, the bandwidth demand change of the network port in a short period of time can be predicted. By summing up the entire bandwidth demand, the bandwidth demand of the entire network can be obtained. By superimposing the network port bandwidth demand prediction functions, a comprehensive demand prediction function for predicting the bandwidth demand of the entire network can be obtained.

[0026] S400 , predicting bandwidth within a preset time length according to a comprehensive demand prediction function to obtain predicted bandwidth data, and allocating network bandwidth to each network port based on the predicted bandwidth data.

[0027] In this step, the bandwidth within a preset time length is predicted based on the comprehensive demand prediction function, and an independent variable sequence is constructed according to the required predicted time length. The independent variable sequence contains a large number of time independent variables, and the time interval between the time independent variables is lower than the time interval for sampling the bandwidth demand. It is imported into the comprehensive demand prediction function to determine the bandwidth demand of the entire network at each moment in the future. It is determined whether to schedule based on the predicted bandwidth demand. If it exceeds the overall maximum network bandwidth, scheduling is performed. At this time, the predicted bandwidth demand of each network port is determined, so as to adjust the bandwidth limit of each network port to determine the transmission bandwidth at each subsequent moment and realize the allocation of network bandwidth to the network port.

[0028] As a preferred embodiment of the present invention, the step of monitoring the demand of each network port, sampling the bandwidth demand of each network port according to a preset sampling demand monitoring frequency, and generating bandwidth demand coordinates corresponding to each network port specifically includes:

[0029] Each network port is accessed at a preset time interval to determine the identity information of each network port and register the network port.

[0030] In this step, each network port is accessed at a preset time interval. The time interval for accessing the network port should be higher than the operator's collection frequency of the network bandwidth, and at least twice the operator's collection frequency. The smaller the time interval, the higher the frequency of accessing each network port. Identity authentication is performed when accessing the network port. Specifically, a corresponding validity period is set for each verification. For example, after each verification is completed, no verification will be performed within ten minutes, and the network port that has passed the verification will be registered.

[0031] After completing the identity authentication, the network port is sampled at a preset frequency to obtain the bandwidth demand data corresponding to each moment.

[0032] In this step, the network port is sampled after identity authentication is completed, and sampling is performed according to the preset sampling interval. Each time sampling is performed, the current sampling time and the collected bandwidth demand data are recorded. Each network port generates a bandwidth demand based on the data currently to be transmitted. The bandwidth demand is the bandwidth expected to be used by the network port. The actual bandwidth of each network port is subsequently determined based on the allocation result, and the bandwidth demand data is obtained through sampling.

[0033] The bandwidth demand coordinates of the network port at the moment are constructed based on the bandwidth demand data and the corresponding collection time, and the abscissa and ordinate of the bandwidth demand coordinates are determined.

[0034] In this step, bandwidth coordinates are constructed based on the bandwidth demand data. The corresponding bandwidth demand coordinates are constructed according to the generation time of the bandwidth demand data, such as the bandwidth demand collected at time A1 is P1, the bandwidth demand collected at time A2 is P2, and the bandwidth demand collected at time An is Pn.

[0035] As a preferred embodiment of the present invention, the steps of constructing a network port bandwidth demand prediction function corresponding to each network port based on the bandwidth demand coordinates of each network port, and synthesizing a comprehensive demand prediction function based on the network port bandwidth demand prediction function specifically include:

[0036] The bandwidth requirement coordinates corresponding to each network port are retrieved, and a preset number of bandwidth requirement coordinates obtained by the latest collection are filtered to obtain a bandwidth requirement coordinate sequence.

[0037] In this step, the bandwidth demand coordinates corresponding to each network port are retrieved. The number of bandwidth demand coordinates retrieved is related to the frequency of sampling the network port. The higher the sampling frequency, the more bandwidth demand coordinates are retrieved. The bandwidth demand coordinates are arranged in chronological order to obtain a bandwidth demand coordinate sequence.

[0038] The bandwidth demand coordinate sequence corresponding to each network port is retrieved, and the corresponding bandwidth demand coordinates are extracted from it in chronological order. The bandwidth demand coordinates are imported into a preset function fitting tool to generate a network port bandwidth demand prediction function.

[0039] In this step, the bandwidth demand coordinate sequence corresponding to each network port is retrieved, and the bandwidth demand coordinates are imported into the function fitting tool one by one according to the time value of the bandwidth demand coordinates. The function fitting tool adopts the Gnuplot function fitting tool. By importing the bandwidth demand coordinates and setting the corresponding function type for fitting, multiple function types are used for fitting to obtain multiple fitting results. The set of fitting results with the highest fitting accuracy is selected to obtain the network port bandwidth demand prediction function of the network port at that moment, as shown in FIG. .

[0040] All network port bandwidth demand prediction functions are integrated and synthesized into a comprehensive demand prediction function.

[0041] In this step, all network port bandwidth demand prediction functions are integrated. That is, by summing up the network port bandwidth demand prediction functions, a comprehensive demand prediction function can be obtained. The comprehensive demand prediction function is expressed as:

[0042] ;

[0043] in, is the comprehensive demand forecast function, It is the network port bandwidth demand prediction function.

[0044] In this step, the steps of retrieving the bandwidth demand coordinate sequence corresponding to each network port, extracting the corresponding bandwidth demand coordinates therefrom in chronological order, importing the bandwidth demand coordinates into a preset function fitting tool, and generating a network port bandwidth demand prediction function specifically include:

[0045] Perform quantum state encoding processing on the bandwidth requirement coordinate sequence to obtain a quantum state sequence;

[0046] Calculate the gradients of adjacent data points in the bandwidth requirement coordinate sequence to obtain a frequency matrix; perform a quantum Fourier transform on the quantum state sequence to obtain a frequency domain quantum state sequence;

[0047] The phase jump of the frequency domain quantum state sequence is detected by the inner product of the frequency domain state, and the phase jump detection result is verified by the gradient value of the frequency matrix to obtain the mutation point position;

[0048] Taking the mutation point position as the segmentation boundary, the frequency domain quantum state sequence is separated into low-frequency trend terms and high-frequency fluctuation terms to obtain quantum characteristic components.

[0049] Establishing a mapping relationship between the characteristic frequency and amplitude of quantum characteristic components and the network state;

[0050] The historical quantum characteristic components are projected into three-dimensional space. Based on the mapping relationship, the DBSCAN algorithm is used for density pre-clustering to identify typical network state patterns and obtain preliminary clustering results. Each cluster center node in the preliminary clustering results represents a network state.

[0051] Calculate the mean vector and covariance matrix of each cluster center of the preliminary clustering results, assign initial weights according to the cluster sample size, and construct an initial Gaussian mixture model based on the mean vector, covariance matrix and initial weights;

[0052] Calculate the Mahalanobis distance between the current quantum characteristic component and each cluster center, and sort them in descending order to obtain a list of Mahalanobis distance calculation results;

[0053] According to the Mahalanobis distance calculation result list, several clusters are selected as active knowledge units, and the initial weights of the initial Gaussian mixture model are adjusted using the active knowledge units to obtain activated meta-knowledge units;

[0054] The feature evolution pattern is taken from each activated meta-knowledge unit to construct a basis function set, which includes a linear trend function, a periodic fluctuation function, and a step response function;

[0055] The basis function combination weights are assigned according to feature correlation to obtain the candidate prediction function set of each cluster center node.

[0056] In this step, by converting the traditional numerical bandwidth data into quantum state form, using the principle of quantum superposition to represent multiple state characteristics, quantum encoding has a natural filtering effect on random noise, and the signal-to-noise ratio of the original data can be effectively improved. The amplitude coding maps the continuous bandwidth value to the superposition state of the quantum bit, so that the microscopic fluctuation characteristics can be explicitly expressed. The frequency matrix is constructed based on the change rate of adjacent data points to capture the instantaneous change trend of network traffic. Quantum Fourier transform converts the time domain signal into the frequency domain, separates the long-term trend and short-term fluctuation, and captures the millisecond-level fluctuation and hour-level trend, solving the problem of insufficient resolution of traditional methods. The quantum state inner product calculation is used to identify the network state mutation point, and when the quantum state is orthogonal, it is determined that the traffic mode has changed. And the probability distribution of the network state is represented by the Gaussian mixture model, and the dynamic clustering center reflects the typical traffic mode.

[0057] In this step, the step of integrating all the network port bandwidth demand prediction functions into a comprehensive demand prediction function includes:

[0058] The gradient descent method optimization and evaluation operation are performed on the candidate prediction function set parameters in sequence to obtain the optimal prediction function, and the prediction results of each clustering center node are output;

[0059] The optimal prediction function of each clustering center node is used as a query, and the historical quantum feature components are used as keys for attention calculation to obtain a spatio-temporal correlation matrix;

[0060] The spatio-temporal correlation matrix is normalized as the weight of the prediction result of each clustering center node to obtain a node weight;

[0061] The prediction results of adjacent clustering center nodes are aggregated by using the node weight for weighted aggregation to obtain a comprehensive demand prediction function.

[0062] In this step, the spatio-temporal correlation weight is calculated by query-key matching to quantify the mutual influence degree of different nodes and different time points, so as to realize the discovery of hidden cross-device dependent relationship and reduce the prediction error. The prediction results of adjacent nodes are aggregated in a collaborative correction manner, and the network topology information is used to compensate for the error caused by transmission delay, so that when a single node data is abnormal, the data is automatically corrected through the associated node data, and the fault tolerance is improved.

[0063] As a preferred embodiment of the present application, the step of predicting the bandwidth in a preset time length according to the comprehensive demand prediction function to obtain predicted bandwidth data, and allocating network bandwidth to each network port based on the predicted bandwidth data includes:

[0064] The prediction time independent variable is constructed based on a preset time step, and the prediction time independent variable is introduced into the comprehensive demand prediction function to output corresponding prediction bandwidth data.

[0065] In this step, the prediction time independent variable is constructed based on a preset time step, and the interval between the prediction time independent variables is adjusted according to the required measurement accuracy. The higher the required accuracy, the smaller the interval between the prediction time independent variables. The prediction time independent variable is introduced into the comprehensive demand prediction function to generate corresponding prediction data and obtain prediction bandwidth data.

[0066] According to the prediction bandwidth data, it is determined whether to perform local scheduling. If the prediction bandwidth is greater than the bandwidth preset value, local scheduling is performed.

[0067] In this step, according to the prediction bandwidth data, it is determined whether to perform local scheduling. If the total bandwidth demand obtained by prediction exceeds the bandwidth preset value, scheduling is required, otherwise, the actual bandwidth demand of each network port is directly used to allocate bandwidth to each network port.

[0068] When local scheduling is performed, the bandwidth type of each network port is determined based on the network port bandwidth demand prediction function, the bandwidth upper limit of the network port is adjusted, and the bandwidth allocation of the network port is completed.

[0069] In this step, when local scheduling is performed, the bandwidth type of each network port is determined based on the network port bandwidth demand prediction function, the service type of the current transmission data of each network port is determined, the network ports are classified according to the service type, the network transmission priority of each network port is determined, and the network ports are sorted, such as first-level network ports, second-level network ports and third-level network ports. The first-level network ports are preferentially allocated larger bandwidth, the second-level network ports are next, and the third-level network ports are allocated last. The actual charging value (i.e. the value of the actual total bandwidth obtained by sampling) in the time interval between the last billing node and the current billing node is counted, and the current billing settlement value is determined according to the 95 billing rule. The values of the actual total bandwidth at multiple time points in the future preset time length are determined. For example, if the current billing settlement value has been determined as M0, the future preset time length is 1 minute, and a total of N time points are set.

[0070] Among them, the actual total bandwidth corresponding to 95% of the time is lower than M0, and the actual total bandwidth of the remaining 5% is a fixed value, which is greater than M0, such as 1.1M0. In other words, within a period of 1 minute, 100 control points are set, and each control point corresponds to a time value. The actual total bandwidth at different time values ​​is different. Among them, the actual total bandwidth corresponding to 95 control points is lower than M0 (a fixed value of 0.99M0 can be used), and the bandwidth of the remaining 5 control points is controlled at 1.1M0. The bandwidth upper limit of all network ports is determined according to the actual total bandwidth set at each moment, so as to allocate bandwidth. When allocating bandwidth, the bandwidth is allocated according to the level of each network port.

[0071] Specifically, the total bandwidth demand of network ports of different levels is counted. For example, the total demand of the first-level network port is M1, the total demand of the second-level network port is M2, and the total demand of the third-level network port is M3. The actual total bandwidth is M, M=M1+M2+M3. The actual total bandwidth M first meets the demand M1 of the first-level network port, then meets the demand M2 of the second-level network port, and finally meets the total demand M3 of the third-level network port. M2 and M3 are set with a minimum bandwidth limit. When the bandwidth allocated to the second-level network port and / or the third-level network port is lower than the corresponding minimum bandwidth limit, the corresponding minimum bandwidth is directly allocated to the third-level network port and the second-level network port, and the remaining part is allocated to the first-level network port. For example, there are two first-level network ports, three second-level network ports, and four third-level network ports. The minimum bandwidth limits of the second-level network port and the third-level network port are K1 and K2 respectively.

[0072] When the bandwidth allocated to the Level 2 and / or Level 3 network ports is lower than the corresponding minimum bandwidth limit, each Level 2 network port is allocated K1 bandwidth, each Level 3 network port is allocated K2 bandwidth, and the remaining portion is allocated to the Level 1 network ports. When the needs of the Level 1 network ports are met, the surplus bandwidth is first allocated to the Level 2 network ports, and then to the Level 3 network ports. In this way, a bandwidth is set for each network port to prevent the final billing bandwidth from exceeding the maximum network bandwidth. At the same time, high-bandwidth areas can be utilized based on the 95 billing rule to reduce network usage costs.

[0073] As a preferred embodiment of the present invention, when performing local scheduling, the bandwidth type of each network port is determined based on the network port bandwidth demand prediction function, the bandwidth upper limit of the network port is adjusted, and the steps of completing the bandwidth allocation of the network port are specifically included:

[0074] Based on the network port bandwidth demand prediction function, the Monte Carlo method is used to generate N groups of future bandwidth demand curves, each of which simulates different network states.

[0075] Calculate the 95th percentile value of each set of future bandwidth demand curves to obtain the 95th percentile value of each set of future bandwidth demand curves;

[0076] Dynamic weights are determined based on historical data, future bandwidth demand curve trends, and bandwidth types;

[0077] Obtain the bandwidth to be allocated for each network port, maximizing the sum of the dynamically weighted bandwidth to be allocated while ensuring that the 95th percentile is lower than the set maximum bandwidth as the optimization goal. Construct an optimization objective function with the constraints that the allocated bandwidth of each network port must not be lower than the minimum guaranteed value and the total bandwidth must not exceed the physical upper limit.

[0078] Genetic algorithm is used to solve the optimization objective function and obtain the Pareto optimal solution, which is used as the preliminary bandwidth allocation scheme.

[0079] Calculate the mean and standard deviation of the historical 95th percentile values, set a dynamic safety threshold based on the mean and standard deviation of the historical 95th percentile values, and set an elastic adjustment range based on the dynamic safety threshold to obtain the real-time safe bandwidth range;

[0080] The system load pressure index is calculated based on the current total predicted bandwidth and the dynamic safety threshold;

[0081] According to the system load pressure index, the adjustment strategy is selected to obtain the anti-fragility adjustment coefficient;

[0082] The preliminary bandwidth allocation plan is adjusted using the anti-fragility adjustment coefficient and weightedly combined with the minimum guarantee value to obtain the bandwidth allocation plan. The bandwidth allocation plan is checked using the real-time safe bandwidth interval as the boundary to obtain the final bandwidth allocation plan.

[0083] In this step, a multi-scenario Monte Carlo simulation generates bandwidth demand curves for normal, congested, and attack scenarios. Probabilistic statistics are used to quantify the billing risks associated with different future states. Quantum superposition is employed to simultaneously consider the combined impact of multiple scenarios within the optimization objective. By weighting the scenarios by probability, bandwidth utilization is improved and 95% of billing overage risks can be identified in advance. Pareto optimality is employed to maximize bandwidth allocation for high-priority services while satisfying billing constraints.

[0084] Using a genetic algorithm to search for the optimal balance point in the solution space, this approach simultaneously optimizes service quality and cost, improving the assurance rate of critical services and achieving higher computational efficiency than traditional linear programming. The final bandwidth allocation is a weighted combination of the initial optimization results and the minimum guaranteed value. The anti-fragility adjustment coefficient reflects the system's security status in real time, dynamically balancing resource utilization and risk control. Even if the prediction model fails, basic service quality can still be guaranteed.

[0085] As a preferred embodiment of the present invention, the step of calculating the dynamic weight based on historical data, future bandwidth demand curve trends, and bandwidth types specifically includes:

[0086] Using a 5-minute sliding window, calculate the bandwidth usage integral value of each network port within the window period, and normalize the bandwidth usage integral values ​​of all network ports and map them to the range of 0-1 to obtain the normalized integral value result;

[0087] Apply exponential amplification to the normalized integral value to obtain a historical contribution coefficient matrix. Each element in the historical contribution coefficient matrix represents the historical bandwidth contribution strength of the corresponding network port.

[0088] Differentiate the future bandwidth demand curve of each network port and calculate the instantaneous rate of change;

[0089] Set a change rate threshold, compare the instantaneous change rate with the change rate threshold, and mark the change exceeding the threshold as a "sudden trend" and the change between the thresholds as a "stable trend";

[0090] Assign a trend gain coefficient to each network port according to the marking result to obtain a trend gain coefficient matrix;

[0091] Analyze the service types carried by each network port and record the cycle length and phase offset of periodic services in different service types;

[0092] Calculate the cycle alignment factor according to the cycle length and phase offset of the periodic service to obtain the cycle alignment matrix;

[0093] Given a basic weight, the historical contribution coefficient matrix, trend gain coefficient matrix, and cycle alignment matrix are weighted and fused according to the basic weight. During the weighted fusion process, the basic weight is dynamically adjusted according to the system load pressure index to obtain the fusion weight.

[0094] The fusion weights of all network ports are linearly scaled, and a weight lower limit is set for the first-level key business to obtain a dynamic weight matrix.

[0095] This step effectively improves prediction accuracy compared to traditional single-dimensional weighting by simultaneously considering long-term behavioral patterns, short-term mutations, business patterns, and real-time status. Furthermore, weighting parameters automatically adjust with network load: high loads prioritize real-time demand, while low loads prioritize historical patterns. A service assurance mechanism is implemented that uses a cycle alignment factor to reduce key frame transmission latency and improve bandwidth assurance for primary services during bursts.

[0096] like Figure 2 As shown, a large-bandwidth network traffic scheduling and management system provided by the present invention is applied to the above-mentioned large-bandwidth network traffic scheduling and management method, and the system includes:

[0097] The parameter acquisition module 100 is used to obtain flow limit parameters and determine the maximum network bandwidth.

[0098] In this system, the parameter acquisition module 100 obtains the traffic limit parameter. When the user and the operator establish a network subscription contract, the maximum network bandwidth is negotiated. The maximum network bandwidth can also be a bandwidth limit set by the user, that is, the maximum transmission speed between the user and the operator. The maximum transmission speed is the maximum network bandwidth. The network bandwidth between the user and the operator is allowed to fluctuate between 0 and the maximum network bandwidth. Finally, the 95th billing principle is adopted to determine the network traffic fee generated by the user. The 95th billing rule is: for example, there are 100 billing values. First, these 100 billing values ​​are arranged from small to large, and the 95th billing value is taken as the billing settlement value. The reason is that 5% of the billing values ​​are discarded because they are floating traffic, which may be attacked in the middle. Therefore, the billing is more humane and ignores 5% of the peak traffic.

[0099] The bandwidth demand coordinate generation module 200 is used to monitor the demand of each network port, sample the bandwidth demand of each network port according to the preset sampling demand monitoring frequency, and generate the bandwidth demand coordinate corresponding to each network port. The horizontal axis of the bandwidth demand coordinate is the time value, and the vertical axis is the sampling bandwidth.

[0100] In this system, the bandwidth demand coordinate generation module 200 monitors the demand of each network port and collects information on the transmission status of each network port. The collection frequency is greater than the operator's collection frequency of network bandwidth usage. For example, a millisecond sampling frequency is used for collection, and traffic conditions are collected once at a preset time interval to determine the bandwidth demand corresponding to that moment, so as to obtain the network bandwidth demand of each network port at that moment. The bandwidth demand coordinates are constructed based on the collected data. The bandwidth demand coordinates are used to record the bandwidth demand at each moment. Each sampling of each network port will obtain a set of bandwidth requirements to form a set of bandwidth demand coordinates.

[0101] The prediction function construction module 300 is used to construct a network port bandwidth demand prediction function corresponding to each network port based on the bandwidth demand coordinates of each network port, and synthesize a comprehensive demand prediction function based on the network port bandwidth demand prediction functions.

[0102] In this system, the prediction function construction module 300 constructs a network port bandwidth demand prediction function corresponding to each network port based on the bandwidth demand coordinates of each network port, retrieves a corresponding number of bandwidth demand coordinates corresponding to each network port, and uses preset function fitting software to uniformly import the retrieved bandwidth demand coordinates into the function fitting software. The function fitting software is used to fit the network port bandwidth demand prediction function corresponding to the network port. Based on the network port bandwidth demand prediction function, the bandwidth demand change of the network port in a short period of time can be predicted. By summing up the entire bandwidth demand, the bandwidth demand of the entire network can be obtained. By superimposing the network port bandwidth demand prediction functions, a comprehensive demand prediction function for predicting the bandwidth demand of the entire network can be obtained.

[0103] The bandwidth allocation module 400 is configured to predict the bandwidth within a preset time period according to the comprehensive demand prediction function, obtain predicted bandwidth data, and allocate network bandwidth to each network port based on the predicted bandwidth data.

[0104] In this system, the bandwidth allocation module 400 predicts the bandwidth within a preset time length based on the comprehensive demand prediction function, and constructs an independent variable sequence based on the required predicted time length. The independent variable sequence contains a large number of time independent variables, and the time interval between the time independent variables is lower than the time interval for sampling the bandwidth demand. It is imported into the comprehensive demand prediction function to determine the bandwidth demand of the entire network at each moment in the future. It is determined whether to schedule based on the predicted bandwidth demand. If it exceeds the overall maximum network bandwidth, scheduling is performed. At this time, the predicted bandwidth demand of each network port is determined, so as to adjust the bandwidth limit of each network port to determine the transmission bandwidth at each subsequent moment, thereby realizing the allocation of network bandwidth to the network port.

[0105] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for scheduling and managing large bandwidth network traffic, characterized in that: The method comprises the following steps: Obtain traffic limit parameters and determine the maximum network bandwidth; Monitor the demand of each network port, sample the bandwidth demand of each network port according to the preset sampling demand monitoring frequency, and generate the bandwidth demand coordinates corresponding to each network port. The horizontal axis of the bandwidth demand coordinates is the time value, and the vertical axis is the sampling bandwidth; Based on the bandwidth demand coordinates of each network port, a network port bandwidth demand prediction function corresponding to each network port is constructed, and a comprehensive demand prediction function is synthesized based on the network port bandwidth demand prediction function; The bandwidth within a preset time length is predicted according to the comprehensive demand prediction function to obtain predicted bandwidth data, and network bandwidth is allocated to each network port based on the predicted bandwidth data; The steps of constructing a network port bandwidth demand prediction function corresponding to each network port based on the bandwidth demand coordinates of each network port, and synthesizing a comprehensive demand prediction function based on the network port bandwidth demand prediction function specifically include: Retrieve the bandwidth requirement coordinates corresponding to each network port, filter the preset number of bandwidth requirement coordinates obtained by the latest collection, and obtain a bandwidth requirement coordinate sequence; Retrieve the bandwidth demand coordinate sequence corresponding to each network port, extract the corresponding bandwidth demand coordinates from it in chronological order, import the bandwidth demand coordinates into a preset function fitting tool, and generate a network port bandwidth demand prediction function; Integrate all network port bandwidth demand prediction functions and synthesize them into a comprehensive demand prediction function; The step of retrieving the bandwidth demand coordinate sequence corresponding to each network port, extracting the corresponding bandwidth demand coordinates therefrom in chronological order, importing the bandwidth demand coordinates into a preset function fitting tool, and generating a network port bandwidth demand prediction function specifically includes: Perform quantum state encoding on the bandwidth requirement coordinate sequence to obtain a quantum state sequence; calculate the gradient of adjacent data points of the bandwidth requirement coordinate sequence to obtain a frequency matrix; perform quantum Fourier transform on the quantum state sequence to obtain a frequency domain quantum state sequence; The phase jump of the frequency domain quantum state sequence is detected by the inner product of the frequency domain state, and the gradient value of the frequency matrix is ​​used to verify the phase jump detection result to obtain the mutation point position; using the mutation point position as the segmentation boundary, the frequency domain quantum state sequence is separated into the low-frequency trend term and the high-frequency fluctuation term to obtain the quantum characteristic component; Establish a mapping relationship between the characteristic frequency and amplitude of quantum characteristic components and the network state; project the historical quantum characteristic components into three-dimensional space, and use the DBSCAN algorithm to perform density pre-clustering based on the mapping relationship to identify typical network state patterns and obtain preliminary clustering results. Each cluster center node in the preliminary clustering results represents a network state; Calculate the mean vector and covariance matrix of each cluster center of the preliminary clustering results, assign initial weights according to the cluster sample size, and construct an initial Gaussian mixture model based on the mean vector, covariance matrix and initial weights; Calculate the Mahalanobis distance between the current quantum feature component and the centers of each cluster, and sort them in descending order to obtain a list of Mahalanobis distance calculation results; based on the list of Mahalanobis distance calculation results, select several clusters as active knowledge units, and use the active knowledge units to adjust the initial weights of the initial Gaussian mixture model to obtain activated meta-knowledge units; The feature evolution pattern is taken from each activated meta-knowledge unit to construct a basis function set, which includes linear trend function, periodic fluctuation function and step response function; the basis function combination weights are assigned according to feature correlation to obtain the candidate prediction function set of each cluster center node.

2. The high-bandwidth network traffic scheduling and management method according to claim 1, characterized in that: The step of monitoring the demand of each network port, sampling the bandwidth demand of each network port according to a preset sampling demand monitoring frequency, and generating bandwidth demand coordinates corresponding to each network port specifically includes: Access each network port at a preset time interval, determine the identity information of each network port, and register the network port; After completing the identity authentication, the network port is sampled at a preset frequency to obtain the bandwidth demand data corresponding to each moment. The bandwidth demand coordinates of the network port at the moment are constructed based on the bandwidth demand data and the corresponding collection time, and the abscissa and ordinate of the bandwidth demand coordinates are determined.

3. The high-bandwidth network traffic scheduling and management method according to claim 1, characterized in that: The step of integrating all network port bandwidth demand prediction functions to form a comprehensive demand prediction function specifically includes: Taking minimizing the historical fitting error and minimizing the distance between the current network state and each cluster center as the optimization goal, the parameters of the candidate prediction function set are optimized and evaluated in sequence using the gradient descent method to obtain the optimal prediction function and output the prediction result of each cluster center node; The optimal prediction function of each cluster center node is used as the query, and the historical quantum feature component is used as the key to perform attention calculation to obtain the spatiotemporal correlation matrix; Normalize the spatiotemporal correlation matrix as the weight of the prediction result of each cluster center node to obtain the node weight; The node weights are used to perform weighted aggregation of the prediction results of adjacent cluster center nodes to obtain a comprehensive demand prediction function.

4. The high-bandwidth network traffic scheduling and management method according to claim 3, characterized in that: The step of predicting the bandwidth within a preset time length according to the comprehensive demand prediction function to obtain predicted bandwidth data, and allocating network bandwidth to each network port based on the predicted bandwidth data specifically includes: Construct a prediction time independent variable based on a preset time step, import the prediction time independent variable into the comprehensive demand prediction function, and output the corresponding prediction bandwidth data; Determine whether to perform local scheduling based on the predicted bandwidth data. If the predicted bandwidth is greater than the preset bandwidth value, perform local scheduling. When performing local scheduling, the bandwidth type of each network port is determined based on the network port bandwidth demand prediction function, the bandwidth upper limit of the network port is adjusted, and the bandwidth allocation of the network port is completed.

5. The high-bandwidth network traffic scheduling and management method according to claim 4, characterized in that: When performing local scheduling, the bandwidth type of each network port is determined based on the network port bandwidth demand prediction function, the bandwidth upper limit of the network port is adjusted, and the steps of completing the bandwidth allocation of the network port are specifically included: Based on the network port bandwidth demand prediction function, the Monte Carlo method is used to generate N groups of future bandwidth demand curves, each of which simulates different network states. Calculate the 95th percentile value of each set of future bandwidth demand curves to obtain the 95th percentile value of each set of future bandwidth demand curves; Dynamic weights are determined based on historical data, future bandwidth demand curve trends, and bandwidth types; Obtain the bandwidth to be allocated for each network port, maximizing the sum of the dynamically weighted bandwidth to be allocated while ensuring that the 95th percentile is lower than the set maximum bandwidth as the optimization goal. Construct an optimization objective function with the constraints that the allocated bandwidth of each network port must not be lower than the minimum guaranteed value and the total bandwidth must not exceed the physical upper limit. Genetic algorithm is used to solve the optimization objective function and obtain the Pareto optimal solution, which is used as the preliminary bandwidth allocation scheme. Calculate the mean and standard deviation of the historical 95th percentile values, set a dynamic safety threshold based on the mean and standard deviation of the historical 95th percentile values, and set an elastic adjustment range based on the dynamic safety threshold to obtain the real-time safe bandwidth range; The system load pressure index is calculated based on the current total predicted bandwidth and the dynamic safety threshold; According to the system load pressure index, the adjustment strategy is selected to obtain the anti-fragility adjustment coefficient; The preliminary bandwidth allocation plan is adjusted using the anti-fragility adjustment coefficient and weightedly combined with the minimum guarantee value to obtain the bandwidth allocation plan. The bandwidth allocation plan is checked using the real-time safe bandwidth interval as the boundary to obtain the final bandwidth allocation plan.

6. The high-bandwidth network traffic scheduling and management method according to claim 5, characterized in that: The step of calculating the dynamic weight based on historical data, future bandwidth demand curve trends, and bandwidth types specifically includes: Using a 5-minute sliding window, calculate the bandwidth usage integral value of each network port within the window period, and normalize the bandwidth usage integral values ​​of all network ports and map them to the range of 0-1 to obtain the normalized integral value result; Apply exponential amplification to the normalized integral value to obtain a historical contribution coefficient matrix. Each element in the historical contribution coefficient matrix represents the historical bandwidth contribution strength of the corresponding network port. Differentiate the future bandwidth demand curve of each network port and calculate the instantaneous rate of change; Set a rate of change threshold, compare the instantaneous rate of change with the rate of change threshold, and mark the change exceeding the threshold as "sudden trend" and the change between the thresholds as "stable trend"; Assign a trend gain coefficient to each network port according to the marking result to obtain a trend gain coefficient matrix; Analyze the service types carried by each network port and record the cycle length and phase offset of periodic services in different service types; Calculate the cycle alignment factor according to the cycle length and phase offset of the periodic service to obtain the cycle alignment matrix; Given a basic weight, the historical contribution coefficient matrix, trend gain coefficient matrix, and cycle alignment matrix are weighted and fused according to the basic weight. During the weighted fusion process, the basic weight is dynamically adjusted according to the system load pressure index to obtain the fusion weight. The fusion weights of all network ports are linearly scaled, and a weight lower limit is set for the first-level key business to obtain a dynamic weight matrix.

7. The high-bandwidth network traffic scheduling and management method according to claim 6, characterized in that: The comprehensive demand forecast function is expressed as: ； in, is the comprehensive demand forecast function, It is the network port bandwidth demand prediction function.

8. A high-bandwidth network traffic scheduling and management system, the system being applied to a high-bandwidth network traffic scheduling and management method according to any one of claims 1 to 7, characterized in that: The system comprises: Parameter acquisition module, used to obtain traffic limit parameters and determine the maximum network bandwidth; The bandwidth demand coordinate generation module is used to monitor the demand of each network port, sample the bandwidth demand of each network port according to the preset sampling demand monitoring frequency, and generate the bandwidth demand coordinate corresponding to each network port. The horizontal axis of the bandwidth demand coordinate is the time value, and the vertical axis is the sampling bandwidth; A prediction function construction module is used to construct a network port bandwidth demand prediction function corresponding to each network port based on the bandwidth demand coordinates of each network port, and synthesize a comprehensive demand prediction function based on the network port bandwidth demand prediction function; The bandwidth allocation module is used to predict the bandwidth within a preset time length according to the comprehensive demand prediction function, obtain predicted bandwidth data, and allocate network bandwidth to each network port based on the predicted bandwidth data.

Images