Communication flow distribution method and system based on network load awareness

Abstract

The invention discloses a communication flow distribution method and system based on network load awareness, and relates to the technical field of network communication, and the method comprises the steps: collecting a multi-dimensional load index of each target node in a network, carrying out the weighted fusion processing through a sliding time window, and generating a comprehensive load score sequence of each target node; based on the historical accumulated data of the comprehensive load score of each target node, adopting a time sequence decomposition algorithm to extract short-term load change trend characteristics of each node; a traffic distribution decision model is established by combining the comprehensive load scoring sequence and the trend characteristics, and the optimal traffic distribution weight of each target node is obtained; according to the optimal weight and the type characteristics of the to-be-distributed traffic, realizing differential distribution of the traffic; and meanwhile, the distribution execution state is monitored, and the distribution process is automatically re-executed when abnormality is detected. According to the invention, full-process intelligent optimization from load sensing to differentiated distribution is realized, and the load balancing effect and the operation stability of the system are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of network communication, in particular to a network load-aware communication traffic distribution method and system. BACKGROUND

[0002] With the rapid growth of Internet application scale, network traffic distribution technology has become a key technology to guarantee system performance. Existing traffic distribution methods mainly include round-robin algorithm, weighted round-robin algorithm and load-based dynamic allocation algorithm, etc. Round-robin algorithm distributes requests to each node in turn according to a preset order, which is simple to implement but does not consider the performance difference of nodes; weighted round-robin algorithm adjusts the traffic distribution ratio by setting static weight values, which can balance the load of different performance nodes to a certain extent; load-based dynamic allocation algorithm dynamically adjusts the traffic distribution strategy according to the real-time load status of nodes. In terms of load evaluation, existing methods usually use a single or a few indicators such as CPU usage, memory usage and connection number to measure the node load level.

[0003] However, existing traffic distribution methods usually adopt a unified distribution strategy when handling different types of traffic, without fully considering the characteristic differences and priority requirements of different service traffic, which cannot meet the differentiated service requirements. At the same time, these methods also have limitations in the comprehensiveness of load evaluation and dynamic response capability. The above technical problems lead to the problems of low resource utilization and degraded service quality in existing traffic distribution systems under high concurrency and multi-service scenarios. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a network load-aware communication traffic distribution method and system to solve the technical problems of unbalanced load distribution, system response lag and degraded service quality.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a network load-aware communication traffic distribution method, which includes:

[0008] Collecting multi-dimensional load indicators of each target node in the network;

[0009] Performing weighted fusion processing on the multi-dimensional load indicators using a sliding time window to generate a comprehensive load score sequence of each target node;

[0010] Based on the historical cumulative data of the comprehensive load scores of each target node, extracting the short-term load trend characteristics of each target node using a time series decomposition algorithm;

[0011] A traffic distribution decision model is established based on the comprehensive load score sequence and the short-term load change trend characteristics, and optimal traffic distribution weights of each target node are obtained.

[0012] Based on the optimal traffic distribution weights, combined with the type characteristics of the to-be-distributed traffic, the traffic is differentially distributed to the corresponding target nodes.

[0013] The traffic distribution execution state of each target node is monitored, and when an abnormal distribution execution or node response is detected, the traffic distribution process is re-executed.

[0014] As a preferred scheme of the network load-aware communication traffic distribution method, the weighted fusion processing of the multi-dimensional load indicators by using the sliding time window includes:

[0015] The multi-dimensional load indicators are standardized;

[0016] Based on the change characteristics of the multi-dimensional load indicators, an adaptive weight calculation method is used to determine the weight coefficients of each load indicator;

[0017] The standardized multi-dimensional load indicators are weighted and fused by using a double-layer sliding time window including a short-time window and a long-time window, and a comprehensive load score sequence of each target node is generated;

[0018] According to the fluctuation characteristics of the comprehensive load score sequence, the length parameters of the double-layer sliding time window are dynamically adjusted.

[0019] As a preferred scheme of the network load-aware communication traffic distribution method, the extraction of the short-term load change trend characteristics of each target node by using the time series decomposition algorithm includes:

[0020] The historical cumulative data of the comprehensive load scores of each target node are obtained, and a historical load score sequence is generated;

[0021] Based on the load change period characteristics of the historical load score sequence, the trend smoothing window size of the STL decomposition is determined, the STL time series decomposition is performed, and the trend component and the residual component are obtained;

[0022] The trend component is linearly regressed and fitted, and the regression slope is calculated as the load growth rate;

[0023] Based on the residual component, the statistical characteristic parameters are calculated, and the load fluctuation amplitude is obtained by combining the autocorrelation analysis;

[0024] Based on the load growth rate and the load fluctuation amplitude, the short-term load change trend characteristics of each target node are generated.

[0025] As a preferred scheme of the network load-aware communication traffic distribution method, the construction process of the traffic distribution decision model comprises:

[0026] Based on the comprehensive load score sequence of each target node, the mean stability and fluctuation intensity of the comprehensive load score sequence are analyzed, and a load stability evaluation index is constructed.

[0027] In combination with the load growth rate and load fluctuation amplitude in the short-term load change trend feature, a load prediction confidence index of each target node is constructed through trend weighting processing.

[0028] Hardware resource configuration information of each target node is obtained, and a node resource carrying capacity score is calculated based on the resource capacity upper limit and the current load level.

[0029] A multi-factor decision matrix is established, the load stability evaluation index, the load prediction confidence index and the node resource carrying capacity score are taken as decision factors, and a comprehensive performance evaluation value of each target node is obtained through a dynamic weighted fusion algorithm.

[0030] Based on the comprehensive performance evaluation value, in combination with the current system load distribution state, a double-mode weight distribution algorithm based on prediction confidence is used to determine the optimal traffic distribution weight of each target node.

[0031] As a preferred scheme of the network load-aware communication traffic distribution method, the traffic is differentiated and distributed to the corresponding target node, which comprises:

[0032] The features of the traffic to be distributed are extracted, and the traffic type identifier and the priority identifier are obtained.

[0033] The service adaptation degree of each target node to the current traffic is calculated based on the traffic type identifier, and a candidate node set that meets the service adaptation degree threshold is selected.

[0034] The optimal traffic distribution weight of the candidate node is adjusted according to the priority identifier of the traffic, and a final distribution weight is generated.

[0035] Based on the final distribution weight, the traffic is distributed to the corresponding target node by using a weighted round-robin algorithm, and distribution statistical information is recorded.

[0036] As a preferred scheme of the network load-aware communication traffic distribution method, the double-mode weight distribution algorithm based on prediction confidence is used to determine the optimal traffic distribution weight of each target node, which comprises:

[0037] The load prediction confidence index of each target node is obtained, and a double-mode fusion ratio is calculated according to the load prediction confidence index of each target node.

[0038] The prospective weight distribution coefficient is calculated based on the comprehensive performance evaluation value of each target node and the short-term load change trend feature.

[0039] The responsive weight distribution coefficient is calculated based on the comprehensive performance evaluation value of each target node and the current system load distribution state.

[0040] The prospective weight distribution coefficient and the responsive weight distribution coefficient are weighted and combined according to a double-mode fusion ratio to generate the optimal flow distribution weight of each target node.

[0041] As a preferred scheme of the network load-aware communication flow distribution method, the multi-dimensional load index includes CPU usage, memory occupancy, network bandwidth utilization, I / O response delay and connection number.

[0042] In a second aspect, the present application provides a network load-aware communication flow distribution system, comprising:

[0043] A data collection module is configured to collect multi-dimensional load indexes of target nodes in a network.

[0044] A load evaluation module is configured to perform weighted fusion processing on the multi-dimensional load indexes by using a sliding time window to generate a comprehensive load score sequence of each target node.

[0045] A trend analysis module is configured to extract a short-term load change trend feature of each target node by using a time series decomposition algorithm based on historical cumulative data of the comprehensive load score of each target node.

[0046] A decision modeling module is configured to establish a flow distribution decision model based on the comprehensive load score sequence and the short-term load change trend feature to obtain the optimal flow distribution weight of each target node.

[0047] A flow distribution module is configured to distribute the flow to corresponding target nodes in a differentiated manner based on the optimal flow distribution weight and in combination with the type feature of the flow to be distributed.

[0048] A monitoring feedback module is configured to monitor the flow distribution execution state of each target node, and re-execute the flow distribution process when detecting distribution execution abnormity or node response abnormity.

[0049] In a third aspect, the present application provides a computer device comprising a memory and a processor, and the memory stores a computer program, wherein the computer program is executed by the processor to implement any step of the network load-aware communication flow distribution method according to the first aspect of the present application.

[0050] In a fourth aspect, the present application provides a computer readable storage medium having stored thereon a computer program, wherein the computer program, when executed by a processor, implements any step of the network load-aware communication traffic distribution method according to the first aspect of the present application.

[0051] The present application has the following beneficial effects: the present application extracts trend components and residual components through an STL time series decomposition algorithm, accurately quantifies the load growth rate and fluctuation amplitude, provides a reliable load trend prediction basis for traffic distribution decisions, and improves the accuracy of load prediction and the forward-looking nature of traffic allocation; through the construction of a multi-dimensional comprehensive evaluation system and a dual-mode weight distribution mechanism, adaptive intelligent traffic allocation is achieved, and the system load balancing effect and robustness are improved; through traffic feature extraction and service adaptation degree evaluation, combined with a priority adjustment mechanism, a differentiated traffic distribution strategy is achieved, and the accuracy of traffic distribution and service matching degree are improved; in combination with the above technical solutions, the present application realizes intelligent optimization of the whole process from load perception collection to differentiated distribution, and overall improves the system load balancing effect and running stability. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0053] Figure 1 Flowchart of the network load-aware communication traffic distribution method.

[0054] Figure 2 Flowchart of short-term load change trend feature extraction of the network load-aware communication traffic distribution method.

[0055] Figure 3 Flowchart of traffic distribution decision model construction of the network load-aware communication traffic distribution method.

[0056] Figure 4 Schematic diagram of the network load-aware communication traffic distribution system. DETAILED DESCRIPTION

[0057] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings.

[0058] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced without the specific details set forth in this description. In other instances, well-known methods, procedures, components, and circuits have not been described in detail so as not to unnecessarily obscure aspects of the present application.

[0059] It should also be noted that, as used in the specification and in the claims, the article "a", "an", or "the" is intended to mean that there are one or more of the features or elements. Unless otherwise indicated, the use of the singular forms "a" "an" and "the" include the plural reference and vice versa, when appropriate. The terms "comprising", "having", "including", and "containing" are to be construed open-ended terms (i.e., meaning "including, but not limited to") unless otherwise noted.

[0060] Reference will now be made to the drawings, in which Figures 1-4 For one embodiment of the present application, the embodiment provides a network load-aware based traffic distribution method, a flow chart as shown in Figure 1 includes the following steps:

[0061] S1: Collecting multi-dimensional load indicators of each target node in the network.

[0062] Among them, the multi-dimensional load indicators include CPU usage, memory occupancy, network bandwidth utilization, I / O response delay, and connection number.

[0063] It should be noted that the selection of multi-dimensional load indicators comprehensively considers the key dimensions of node performance, in which CPU usage and memory occupancy reflect the basic computing resource status of the node, network bandwidth utilization embodies data transmission capability, I / O response delay represents storage system performance, and connection number reflects the concurrent processing load of the node. These five indicators can comprehensively and accurately evaluate the real-time load level of the target node.

[0064] S2: Using a sliding time window to perform weighted fusion processing on the multi-dimensional load indicators to generate a comprehensive load score sequence of each target node.

[0065] Specifically, step S2 includes:

[0066] S2.1: Standardizing the multi-dimensional load indicators.

[0067] Among them, the standardization processing adopts a piecewise standardization method, linear standardization for percentage indicators, and robust standardization for numerical indicators to eliminate the dimensional differences between different indicators.

[0068] S2.2: Based on the change characteristics of the multi-dimensional load indicators, an adaptive weight calculation method is used to determine the weight coefficients of each load indicator.

[0069] In one specific embodiment, a change feature of each load indicator is calculated, the change feature being quantified by calculating a coefficient of variation of each load indicator; based on the change feature, an objective weight of each load indicator is calculated using an entropy weight method; based on a preset index importance level, a subjective weight of each load indicator is determined using an analytic hierarchy process; the objective weight and the subjective weight are weighted and fused to obtain a weight coefficient of each load indicator. The index importance level is set by analyzing the correlation degree between historical load data and system performance bottlenecks.

[0070] S2.3: The standardized multi-dimensional load indicators are weighted and fused using a double-layer sliding time window comprising a short-time window and a long-time window to generate a comprehensive load score sequence of each target node.

[0071] It should be noted that the short-time window is used to capture the real-time change feature of the load, the long-time window is used to reflect the historical trend feature of the load, and the comprehensive load score is calculated by weighted average of the short-time score and the long-time score.

[0072] Further, the length of the short-time window is preferably 5-15 time units, the length of the long-time window is preferably 30-90 time units, and the weight ratio of the short-time score and the long-time score is preferably 0.6:0.4 to 0.8:0.2. The time unit can be seconds, minutes or other preset sampling intervals. The double-layer sliding window mechanism can ensure the real-time performance of load evaluation while taking into account the stability of historical trends.

[0073] S2.4: According to the fluctuation feature of the comprehensive load score sequence, the length parameter of the double-layer sliding time window is dynamically adjusted.

[0074] The fluctuation feature includes score variance. Specifically, the dynamic adjustment process is as follows: when the score variance is lower than a stability threshold, it indicates that the load change is relatively stable, and the length of the time window is appropriately lengthened to improve the stability of the score; when the score variance is higher than a volatility threshold, it indicates that the load change is relatively severe, and the length of the time window is appropriately shortened to improve the response speed to the load change; when the score variance is between the two thresholds, the current window length is kept unchanged.

[0075] As a preferred embodiment, the stability threshold is determined according to the lower quartile of the historical score variance distribution, the volatility threshold is determined according to the upper quartile of the historical score variance distribution, and the adjustment range of the length of the time window is preferably 10%-30% of the original length. This dynamic adjustment mechanism enables the system to adaptively balance the real-time performance and stability of load evaluation, and improves the effect of load balancing.

[0076] S3: Based on the historical cumulative data of the comprehensive load scores of each target node, a time series decomposition algorithm is used to extract the short-term load change trend feature of each target node.

[0077] Specifically, the short-term load change trend feature extraction flowchart as shown in Figure 2 includes:

[0078] S3.1: Obtain the historical cumulative data of the comprehensive load score of each target node, and generate a historical load score sequence.

[0079] It should be noted that the historical cumulative data refers to the cumulative record of the comprehensive load score sequence of each target node within a period of time.

[0080] S3.2: Determine the trend smoothing window size of STL decomposition based on the load change period characteristics of the historical load score sequence, perform STL time series decomposition, and obtain the trend component and residual component.

[0081] The STL time series decomposition algorithm is adopted in the present application because the STL time series decomposition algorithm can decompose the historical load score sequence into a trend component and a residual component, wherein the trend component reflects the long-term change direction of the load, and the residual component reflects the random fluctuation characteristics of the load, thereby providing a data basis for subsequent load growth rate calculation and fluctuation amplitude analysis.

[0082] In one specific embodiment, step S3.2 includes: performing periodic analysis on the historical load score sequence, identifying the dominant period length and periodic variation mode of the historical load score sequence; calculating the trend smoothing window size of STL decomposition using an empirical formula according to the identified dominant period length, and setting the seasonal smoothing parameter; performing the STL time series decomposition algorithm to complete the time series decomposition through inner and outer loop iteration calculation; extracting the trend component and residual component from the STL decomposition result, and verifying the integrity of the decomposition result.

[0083] Through the above-mentioned STL time series decomposition processing, the extraction of the trend component provides smooth data input for linear regression fitting, eliminating the influence of periodic fluctuations and random noise in the original sequence on the slope calculation. At the same time, the acquisition of the residual component provides pure fluctuation data for statistical feature calculation, making the quantification of load fluctuation amplitude more accurate. In addition, the trend smoothing window size determined based on the dominant period length can adaptively adjust the decomposition precision, ensuring that the trend component retains the long-term change characteristics.

[0084] S3.3: Perform linear regression fitting on the trend component, and calculate the regression slope as the load growth rate.

[0085] Further, a trend regression dataset is constructed with time sequence number as independent variable and trend component value as dependent variable; a linear regression model of the trend component and the time sequence number is established by linear regression fitting of the trend regression dataset using least square method, including: initializing regression parameters, solving optimal parameters by minimizing residual sum of squares, and establishing a linear relationship; a slope coefficient of the regression line is extracted from the linear regression model, which is the load growth rate of the target node; a significance test is performed on the regression fitting result to ensure the effectiveness of the linear regression model.

[0086] S3.4: Calculate statistical feature parameters based on the residual component, and obtain the load fluctuation amplitude based on autocorrelation analysis.

[0087] Further, the standard deviation of the residual component is calculated as a basic fluctuation feature parameter; autocorrelation function analysis is performed on the residual component to calculate autocorrelation coefficients at multiple lag orders; a correlation index of the residual component is calculated based on the autocorrelation coefficients; the standard deviation of the residual component and the correlation index are combined by weighting to obtain the load fluctuation amplitude of the target node.

[0088] S3.5: Based on the load growth rate and the load fluctuation amplitude, generate short-term load change trend features of each target node.

[0089] Preferably, through the comprehensive processing flow of time series decomposition and statistical analysis, the historical load data of each target node is converted into two quantitative indicators of load growth rate and load fluctuation amplitude. The generated short-term load change trend features can reflect the development direction and stability level of node load at the same time, providing data support for load prediction credibility evaluation in the subsequent traffic distribution decision model, so that the calculation of traffic allocation weight can be dynamically adjusted based on load trend prediction.

[0090] S4: Based on the comprehensive load score sequence and the short-term load change trend features, a traffic distribution decision model is established to obtain the optimal traffic allocation weight of each target node.

[0091] Specifically, the traffic distribution decision model construction flowchart is as shown in Figure 3 , including:

[0092] S4.1: Based on the comprehensive load score sequence of each target node, analyze the mean stability and fluctuation intensity of the comprehensive load score sequence, and construct a load stability evaluation index.

[0093] In one embodiment, the construction of the load stability evaluation index includes: calculating the arithmetic mean and weighted mean of the sequence of comprehensive load scores, and evaluating the mean stability through the relative deviation of the two; calculating the coefficient of variation through sliding variance analysis to quantify the degree of load fluctuation; identifying load mutation points through difference analysis, and counting the mutation frequency and amplitude; calculating the average absolute deviation of the comprehensive load score and the moving average, and obtaining the dispersion degree by combining the interquartile range; and weighting and fusing the above indexes using the information entropy weight method to obtain the load stability evaluation index.

[0094] S4.2: Based on the load growth rate and load fluctuation amplitude in the short-term load change trend feature, a load prediction reliability index of each target node is constructed through trend weighting processing.

[0095] Further, the construction of the load prediction reliability index includes: calculating a trend stability coefficient according to the absolute value size and consistency of the change direction of the load growth rate; calculating a fluctuation controllability coefficient based on the relative size and fluctuation frequency of the load fluctuation amplitude; calculating a prediction historical performance coefficient through historical prediction accuracy statistics; using a time decay weighting method to give higher weight to recent prediction performance; and after normalization processing of the trend stability coefficient, the fluctuation controllability coefficient and the prediction historical performance coefficient, obtaining the load prediction reliability index of each target node through weighted average.

[0096] S4.3: Obtain the hardware resource configuration information of each target node, and calculate the node resource carrying capacity score based on the resource capacity upper limit and the current load level.

[0097] The hardware resource configuration information includes the number of CPU cores, the memory capacity, the network bandwidth upper limit, the storage I / O performance, and the maximum number of concurrent connections.

[0098] In addition, the calculation process of the node resource carrying capacity score is as follows: based on the multi-dimensional load indexes collected in S1 and the hardware resource configuration information, the resource utilization rate of each target node in each dimension is calculated; the resource utilization rate in each dimension is converted into the percentage of the remaining available capacity; and the remaining available capacity in each dimension is weighted and averaged to obtain the overall resource carrying capacity score of the node.

[0099] It should be noted that the load stability evaluation index reflects the historical performance and stability of the node load, and the node with higher stability is not prone to performance fluctuation when carrying new traffic, which is conducive to ensuring the consistency of service quality. The load prediction reliability index is based on the load trend characteristics of the node, and evaluates the accuracy of the prediction of the future load state of the node. The node with higher reliability is convenient for accurate traffic planning and capacity estimation. The resource carrying capacity score is based on the hardware configuration and current resource usage of the node, and directly reflects the physical carrying capacity of the node to receive new traffic. The three dimensions comprehensively evaluate the comprehensive service capacity of the node from the historical stability, future predictability and current carrying capacity, and realize precise and efficient traffic distribution through multi-dimensional comprehensive evaluation.

[0100] S4.4: Establish a multi-factor decision matrix, take the load stability evaluation index, the load prediction reliability index and the node resource carrying capacity score as the decision factors, and obtain the comprehensive performance evaluation value of each target node through a dynamic weighted fusion algorithm.

[0101] Further, the calculation of the comprehensive performance evaluation value includes: constructing a multi-factor decision matrix of the three decision factors of each target node, wherein the rows represent the nodes and the columns represent the decision factors; normalizing each decision factor in the multi-factor decision matrix to eliminate the influence of different index dimensions; dynamically adjusting the weight coefficients of the three decision factors according to the current system load distribution state and combining the traffic distribution strategy target, for example, increasing the weight of the resource carrying capacity score when the system load is high, increasing the weight of the load stability evaluation index when the load fluctuation is large, and increasing the weight of the load prediction reliability index when the prediction accuracy is low; using a weighted summation method to fuse the normalized decision factors of each node according to the dynamic weight to obtain the comprehensive performance evaluation value of each target node.

[0102] Among them, the traffic distribution strategy target includes load balancing, system performance optimization, service stability guarantee, etc. Under the load balancing strategy, the load stability is given priority, under the performance optimization strategy, the resource carrying capacity is given priority, and under the stability guarantee strategy, the prediction reliability is given priority.

[0103] S4.5: Based on the comprehensive performance evaluation value, combining the current system load distribution state, using a double-mode weight distribution algorithm based on prediction reliability to determine the optimal traffic distribution weight of each target node.

[0104] Specifically, step S4.5 includes:

[0105] S4.5.1: Obtain the load prediction reliability index of each target node, and calculate the double-mode fusion ratio according to the load prediction reliability index of each target node.

[0106] Specifically, a high credibility threshold and a low credibility threshold are set, target nodes are divided into a high credibility node set, a medium credibility node set and a low credibility node set according to the comparison results of the load prediction credibility indicators of the target nodes with the thresholds, a comprehensive performance evaluation value of each target node is obtained, the sum of the comprehensive performance evaluation values of all nodes in each node set is calculated respectively to obtain the weight coefficients of each node set, different forward-looking preference coefficients are set for each node set, wherein the forward-looking preference coefficient of the high credibility node set is the highest and the forward-looking preference coefficient of the low credibility node set is the lowest, the weight coefficients of each node set and the corresponding forward-looking preference coefficients are weighted to obtain a forward-looking mode fusion ratio, and a responsive mode fusion ratio is calculated based on the constraint condition that the sum of the double mode fusion ratios is 1. Through node grouping and weight weighting, the abstract "credibility" is converted into the specific "fusion ratio".

[0107] S4.5.2: Based on the comprehensive performance evaluation value and the short-term load change trend characteristics of each target node, a forward-looking weight distribution coefficient is calculated.

[0108] Further, the comprehensive performance evaluation value of each target node is obtained as a basic weight, the load growth rate and the load fluctuation amplitude in the short-term load change trend characteristics are extracted, a trend adjustment factor is calculated based on the load growth rate, a higher trend adjustment factor is given to a node with a lower load growth rate, and a lower trend adjustment factor is given to a node with a higher load growth rate, a stability adjustment factor is calculated based on the load fluctuation amplitude, a higher stability adjustment factor is given to a node with a smaller load fluctuation amplitude, the comprehensive performance evaluation value of each target node is multiplied by the corresponding trend adjustment factor and stability adjustment factor to obtain the forward-looking weight initial value of each node, and the forward-looking weight initial value of each node is normalized so that the sum of the forward-looking weight distribution coefficients of all nodes is 1.

[0109] S4.5.3: Based on the comprehensive performance evaluation value and the current system load distribution state of each target node, a responsive weight distribution coefficient is calculated.

[0110] Further, the comprehensive performance evaluation value of each target node is acquired as a basic weight; the current system load distribution state is acquired, including the current load level of each node and the system overall load balancing degree; a load adjustment factor is calculated based on the current load level of each node, a higher load adjustment factor is given to a node with a lower current load level, and a lower load adjustment factor is given to a node with a higher current load level; a balancing adjustment factor is calculated based on the system overall load balancing degree, and when the system load distribution is unbalanced, an additional weight compensation is given to a node with a lighter load; the comprehensive performance evaluation value of each target node is multiplied by the corresponding load adjustment factor and balancing adjustment factor to obtain the responsive weight initial value of each node; and the responsive weight initial value of each node is normalized to make the sum of the responsive weight distribution coefficients of all nodes equal to 1.

[0111] S4.5.4: The proactive weight distribution coefficient and the responsive weight distribution coefficient are weighted and combined according to the double-mode fusion ratio to generate the optimal traffic distribution weight of each target node.

[0112] Specifically, the proactive weight distribution coefficient of each target node is multiplied by the proactive mode fusion ratio to obtain a proactive weight component; the responsive weight distribution coefficient of each target node is multiplied by the responsive mode fusion ratio to obtain a responsive weight component; and the corresponding proactive weight component and responsive weight component of each node are added to obtain the optimal traffic distribution weight of each target node.

[0113] Preferably, the present application dynamically balances the proactive decision (based on prediction) and the responsive decision (based on the current state) according to the system state and the node characteristics through the double-mode weight distribution algorithm based on prediction credibility, so as to realize an adaptive traffic distribution strategy. Through the prediction credibility hierarchical mechanism and the differentiated weight adjustment, the load prediction accuracy is improved; through the double-mode fusion mechanism, intelligent traffic distribution in a complex dynamic environment is realized, the system load balancing effect is improved, and good adaptability and robustness are achieved, thereby reducing the system overload risk.

[0114] S5: Based on the optimal traffic distribution weight, the traffic is differentiated and distributed to the corresponding target node in combination with the type characteristics of the to-be-distributed traffic.

[0115] Specifically, step S5 includes:

[0116] S5.1: The characteristics of the to-be-distributed traffic are extracted to obtain a traffic type identifier and a priority identifier.

[0117] It should be noted that the traffic feature extraction identifies the traffic type (such as HTTP, database access, file transfer, etc.) and obtains the traffic type identifier by analyzing the packet header information, protocol type, port number, load content and other characteristics; at the same time, the priority identifier is determined according to the business importance, delay sensitivity, bandwidth demand and other factors, which provides basic data for subsequent differentiated traffic distribution.

[0118] S5.2: Calculate the service adaptation degree of each target node to the current traffic based on the traffic type identifier, and filter out the candidate node set whose service adaptation degree meets the preset adaptation degree threshold.

[0119] Further, the calculation of the service adaptation degree includes: obtaining the service processing capability configuration information of each target node, including the supported service type, the processing performance parameters of each service type; according to the matching degree calculation of the traffic type identifier and the supported service type of the node, the service type adaptation degree of complete matching is 1, and the adaptation degree value is calculated according to the similarity of partial matching; the performance adaptation degree is calculated by combining the service processing performance parameters of the node and the resource demand of the current traffic; the matching degree and the performance adaptation degree are weighted and fused to obtain the service adaptation degree of each target node to the current traffic.

[0120] In addition, the preset adaptation degree threshold is dynamically determined according to the system load state and the quality of service requirement: when the system load is low, a higher adaptation degree threshold is set to ensure optimal matching; when the system load is high, the adaptation degree threshold is appropriately reduced to expand the range of candidate nodes to ensure that the traffic can be distributed in time; at the same time, according to the historical distribution effect statistics, the setting strategy of the adaptation degree threshold is adjusted.

[0121] Preferably, through the service adaptation degree evaluation mechanism, accurate matching of traffic and nodes is realized, avoiding the distribution of specific business traffic to unsuitable nodes, and improving the business processing efficiency; at the same time, the dynamic adaptation degree threshold strategy is adopted, which takes into account the system load balancing demand on the premise of ensuring the quality of business matching, enhances the flexibility of traffic distribution and the overall performance of the system.

[0122] S5.3: Adjust the optimal traffic allocation weight of the candidate node according to the priority identifier of the traffic, and generate the final allocation weight.

[0123] Further, weight adjustment coefficients corresponding to different priorities are set, the adjustment coefficient of high-priority traffic is greater than 1, and the adjustment coefficient of low-priority traffic is less than 1; the performance ranking of each node in the candidate node set is obtained, and the node with better performance is marked as a preferred node; for high-priority traffic, the weight adjustment coefficient of the preferred node is increased, and the weight adjustment coefficient of other nodes is reduced; for low-priority traffic, a relatively balanced weight adjustment strategy is adopted; the optimal traffic allocation weight of each candidate node is multiplied by the corresponding weight adjustment coefficient to obtain an adjusted weight value; the adjusted weight value is normalized to ensure that the sum of the final allocation weights of all candidate nodes is 1. Through priority-differentiated weight adjustment, critical services can be ensured to obtain high-quality resources in priority, precise matching between service importance and resource allocation is achieved, and overall service quality and resource utilization efficiency are improved.

[0124] S5.4: Based on the final allocation weight, the traffic is distributed to the corresponding target node using a weighted round-robin algorithm, and distribution statistical information is recorded.

[0125] Further, the traffic distribution based on the final allocation weight includes: converting the final allocation weight of each candidate node into an integer weight value to ensure weight precision through proportional scaling; establishing a weighted round-robin queue, and allocating a corresponding number of scheduling positions in the queue according to the weight value of each node; selecting target nodes in sequence according to the queue order for traffic distribution, and when the weight of a node is consumed, entering the next cycle; real-time recording of statistical information such as the distribution traffic size, distribution times, response time, etc. of each node; updating the load state of the node according to the distribution statistical information to provide data support for the next traffic distribution decision.

[0126] S6: Monitor the traffic distribution execution state of each target node, and when an abnormal distribution execution or node response is detected, re-execute the traffic distribution process.

[0127] Specifically, the traffic reception state, response time and load change index of each target node are collected in real time; the collected state indexes are statistically analyzed to calculate an abnormality degree score; when the abnormality degree score continuously exceeds the abnormality judgment threshold for a preset number of times, it is determined to be a continuous abnormal state; according to the severity of the abnormality, a processing strategy is selected: for a mild abnormality, only the weight of the node is adjusted, and for a severe abnormality, the traffic distribution to the abnormal node is suspended; based on the abnormal type and historical recovery time, the recovery interval time is dynamically determined, and then the traffic distribution process is re-executed.

[0128] The abnormality judgment threshold is determined through statistical analysis of historical normal operation data, and the preset number of times is set to 3.

[0129] The embodiment also provides a network load-aware communication traffic distribution system, as shown in a schematic diagram Figure 4 The system includes:

[0130] a data collection module configured to collect multi-dimensional load indicators of each target node in the network;

[0131] a load evaluation module configured to perform weighted fusion processing on the multi-dimensional load indicators by using a sliding time window to generate a comprehensive load score sequence of each target node;

[0132] a trend analysis module configured to extract short-term load change trend characteristics of each target node by using a time series decomposition algorithm based on historical cumulative data of the comprehensive load scores of each target node;

[0133] a decision modeling module configured to establish a traffic distribution decision model based on the comprehensive load score sequence and the short-term load change trend characteristics to obtain optimal traffic distribution weights of each target node;

[0134] a traffic distribution module configured to distribute traffic to corresponding target nodes in a differentiated manner based on the optimal traffic distribution weights and in combination with type characteristics of the traffic to be distributed;

[0135] a monitoring feedback module configured to monitor an execution state of traffic distribution of each target node, and re-execute the traffic distribution process when detecting distribution execution abnormities or node response abnormities.

[0136] The embodiment also provides a computer device suitable for the network load-aware communication traffic distribution method, which comprises a memory and a processor; the memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions to implement the network load-aware communication traffic distribution method proposed in the above embodiment.

[0137] The computer device can be a terminal, which comprises a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is configured to perform wired or wireless communication with external terminals. The wireless communication can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0138] The embodiment also provides a storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method for distributing communication traffic based on network load awareness proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic storage, a flash memory, a magnetic disk, or an optical disk.

[0139] To sum up, the application extracts the trend component and the residual component through the STL time series decomposition algorithm, realizes the accurate quantification of the load growth rate and the fluctuation amplitude, provides a reliable load trend prediction basis for the traffic distribution decision, improves the accuracy of the load prediction and the foresight of the traffic allocation, realizes the adaptive intelligent traffic allocation through the construction of the multi-dimensional comprehensive evaluation system and the double-mode weight distribution mechanism, improves the system load balancing effect and the robustness, realizes the differentiated traffic distribution strategy through the traffic feature extraction and the service adaptation degree evaluation combined with the priority adjustment mechanism, improves the accuracy of the traffic distribution and the service matching degree, and realizes the whole-process intelligent optimization from the load awareness collection to the differentiated distribution, and improves the system load balancing effect and the operation stability.

[0140] It should be noted that the above embodiments are only used to illustrate the technical solutions of the application rather than limit the application, and although the application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the application, and all should be covered in the scope of the claims of the application.

Claims

1. A communication traffic distribution method based on network load perception, characterized by: include: Collect multi-dimensional load indicators of each target node in the network; A sliding time window is used to perform weighted fusion processing on the multi-dimensional load indicators to generate a comprehensive load score sequence for each target node; Based on the historical accumulated data of the comprehensive load score of each target node, the time series decomposition algorithm is used to extract the short-term load change trend characteristics of each target node; Establishing a traffic distribution decision model based on the comprehensive load score sequence and the short-term load change trend characteristics to obtain the optimal traffic distribution weight for each target node; Based on the optimal traffic distribution weight and in combination with the type characteristics of the traffic to be distributed, the traffic is distributed to the corresponding target nodes in a differentiated manner; Monitor the traffic distribution execution status of each target node, and re-execute the traffic distribution process when an abnormal distribution execution or node response is detected.

2. The communication traffic distribution method based on network load perception according to claim 1, characterized in that: The weighted fusion processing of the multi-dimensional load indicators using a sliding time window includes: Standardize multi-dimensional load indicators; Based on the variation characteristics of the multi-dimensional load indicators, a weight coefficient of each load indicator is determined using an adaptive weight calculation method; A double-layer sliding time window consisting of a short-term window and a long-term window is used to perform weighted fusion on the standardized multi-dimensional load indicators to generate a comprehensive load score sequence for each target node. According to the fluctuation characteristics of the comprehensive load score sequence, the length parameters of the double-layer sliding time window are dynamically adjusted.

3. The communication traffic distribution method based on network load perception according to claim 1, characterized in that: The method of extracting the short-term load change trend characteristics of each target node using the time series decomposition algorithm includes: Obtain the historical accumulated data of the comprehensive load score of each target node and generate a historical load score sequence; Determine the trend smoothing window size of STL decomposition based on the load change period characteristics of the historical load score sequence, perform STL time series decomposition, and obtain a trend component and a residual component; Performing linear regression fitting on the trend component and calculating the regression slope as the load growth rate; Calculate statistical characteristic parameters based on the residual components and obtain load fluctuation amplitude by combining autocorrelation analysis; Based on the load growth rate and the load fluctuation amplitude, a short-term load change trend feature of each target node is generated.

4. The communication traffic distribution method based on network load perception according to claim 1, characterized in that: The process of constructing the traffic distribution decision model includes: Based on the comprehensive load score sequence of each target node, the mean stability and fluctuation intensity of the comprehensive load score sequence are analyzed to construct a load stability evaluation index; Combining the load growth rate and load fluctuation amplitude in the short-term load change trend characteristics, the load forecast credibility index of each target node is constructed through trend weighting processing; Obtain the hardware resource configuration information of each target node and calculate the node resource carrying capacity score based on the resource capacity upper limit and the current load level; Establish a multi-factor decision matrix, use the load stability evaluation index, the load prediction credibility index and the node resource carrying capacity score as decision factors, and obtain the comprehensive performance evaluation value of each target node through a dynamic weighted fusion algorithm; Based on the comprehensive performance evaluation value and in combination with the current system load distribution status, a dual-mode weight allocation algorithm based on prediction credibility is used to determine the optimal flow distribution weight of each target node.

5. The communication traffic distribution method based on network load perception according to claim 1, characterized in that: The differentiated distribution of traffic to corresponding target nodes includes: Extract features of the traffic to be distributed and obtain traffic type and priority identifiers; Calculate the service adaptability of each target node to the current traffic based on the traffic type identifier, and select a set of candidate nodes that meet the service adaptability requirements according to a preset adaptability threshold; Adjust the optimal traffic allocation weight of the candidate node according to the traffic priority identifier to generate the final allocation weight; Based on the final distribution weights, a weighted round-robin algorithm is used to distribute traffic to corresponding target nodes, and distribution statistics are recorded.

6. The communication traffic distribution method based on network load perception according to claim 4, characterized in that: The method of determining the optimal traffic distribution weight of each target node by using a dual-mode weight distribution algorithm based on prediction credibility includes: Obtaining the load prediction credibility index of each target node, and calculating the dual-mode fusion ratio based on the load prediction credibility index of each target node; Calculate the forward-looking weight allocation coefficient based on the comprehensive performance evaluation value of each target node and the short-term load change trend characteristics; Calculate the responsive weight distribution coefficient based on the comprehensive performance evaluation value of each target node and the current system load distribution status; The forward-looking weight distribution coefficient and the responsive weight distribution coefficient are weighted and combined according to the dual-mode fusion ratio to generate the optimal traffic distribution weight of each target node.

7. The communication traffic distribution method based on network load perception according to claim 1, characterized in that: The multi-dimensional load indicators include CPU usage, memory occupancy, network bandwidth utilization, I / O response delay and number of connections.

8. A communication traffic distribution system based on network load perception, based on the communication traffic distribution method based on network load perception according to any one of claims 1 to 7, characterized in that: include: Data collection module, used to collect multi-dimensional load indicators of each target node in the network; A load evaluation module is used to perform weighted fusion processing on the multi-dimensional load indicators using a sliding time window to generate a comprehensive load score sequence for each target node; The trend analysis module is used to extract the short-term load change trend characteristics of each target node based on the historical accumulated data of the comprehensive load score of each target node using a time series decomposition algorithm; A decision modeling module is used to establish a traffic distribution decision model based on the comprehensive load score sequence and the short-term load change trend characteristics to obtain the optimal traffic distribution weight of each target node; A traffic distribution module is configured to distribute traffic to corresponding target nodes based on the optimal traffic distribution weight and the type characteristics of the traffic to be distributed; The monitoring feedback module is used to monitor the traffic distribution execution status of each target node. When an abnormal distribution execution or node response is detected, the traffic distribution process is re-executed.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the communication traffic distribution method based on network load perception according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the communication traffic distribution method based on network load perception according to any one of claims 1 to 7 are implemented.

Images