Network resource allocation method and system based on network load

Abstract

The invention discloses a network resource allocation method and system based on a network load, and relates to the technical field of network resource allocation, and the method comprises the following steps: 1, determining a service demand satisfaction degree; step 2, calculating a network load balancing degree; 3, predicting future business demand data; and 4, defining constraint conditions, maximizing the service demand satisfaction degree and the network balancing load degree, and formulating a network resource allocation strategy. According to the method, the situation that only the overall average level is concerned and individual key indexes are ignored to be seriously substandard when the business requirements are evaluated in the prior art can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of network resource allocation, and particularly to a network resource allocation method and system based on network load. Background Art

[0002] In today's digital age, the types of network services are becoming increasingly complex, covering from core key services such as financial trading systems and medical data transmission, to less important services like enterprise daily office applications, and general services such as ordinary web browsing. Different services have different requirements for network resources and vary greatly in their sensitivity to network performance. The satisfaction of service requirements is related to the quality of service operation, which involves considerations of resource requests and actual allocations, as well as numerous quantitative indicators such as latency time and packet loss rate, which reflect the actual performance of the network when carrying services. The degree of network load balance is a key indicator for measuring whether the load distribution of each node in the network is reasonable and is closely related to load data such as the CPU usage rate, memory occupancy rate, and bandwidth utilization rate of the nodes. Accurate prediction of future service demand data can help the network plan resources in advance and improve resource utilization efficiency.

[0003] However, there are many deficiencies in the existing technology in terms of network resource allocation. When evaluating whether resource allocation meets service requirements, the existing technology may mainly focus on whether the total amount of resource allocation reaches the service request volume, such as only looking at whether sufficient bandwidth resources are allocated. If the allocated bandwidth reaches the bandwidth value requested by the service, it is considered that the resource allocation meets the requirements. However, in fact, there may be problems such as too long latency time and high packet loss rate. For example, in a video conferencing service, although the allocated bandwidth can ensure that the video stream is basically not stuck, too high latency will cause audio and video out-of-sync, seriously affecting the user experience, and this situation where key indicators are not met will be masked by the seemingly sufficient bandwidth allocation with better indicators. Summary of the Invention

[0004] The purpose of the present invention is to provide a network resource allocation method and system based on network load to solve the problems raised in the existing technology.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A network resource allocation method based on network load, the method includes the following steps:

[0006] Step 1, determine the service requirement satisfaction degree;

[0007] Step 2, calculate the degree of network load balance;

[0008] Step 3, predict future service demand data;

[0009] Step 4: Define the constraint conditions, maximize the satisfaction of business requirements and the degree of network balanced load, and formulate a network resource allocation strategy.

[0010] In Step 1, the specific steps to determine the satisfaction of business requirements are as follows:

[0011] Step 1.1: Define the types of different services and their corresponding priorities;

[0012] The service types include but are not limited to core critical services, less important services, and general services;

[0013] The priorities include but are not limited to the highest priority, high priority, medium priority, and low priority;

[0014] Step 1.2: Collect the resource request data and actual allocation situations of different service types;

[0015] The resource request data includes but is not limited to the required bandwidth, computing resources, and storage resources;

[0016] Step 1.3: Select quantification indicators and establish evaluation criteria according to different service types;

[0017] The quantification indicators include but are not limited to the latency time, packet loss rate, average processing time, and resource utilization rate;

[0018] Set different quantification indicator conditions for different service types based on different priorities;

[0019] Step 1.4: Compare the resource request data and actual allocation situations of each service type, and calculate the satisfaction of business requirements in combination with the corresponding quantification indicator conditions in the evaluation criteria;

[0020] For a certain actual allocation situation or quantification indicator of a certain service type:

[0021] When the actual allocation situation Re meets the resource request, set the satisfaction to 1;

[0022] When the actual allocation situation Re does not meet the resource request condition threshold Ru, set the satisfaction to 1 - |Re - Ru| / Ru; where |Re - Ru| represents the absolute value of the difference between the actual allocation situation Re and the resource request condition threshold Ru;

[0023] When the quantification indicator T meets the quantification indicator condition, set the satisfaction to 1;

[0024] When the quantification indicator T does not meet the quantification indicator condition threshold Th, set the satisfaction to 1 - |T - Th| / Th; where |T - Th| represents the absolute value of the difference between the quantification indicator T and the quantification indicator threshold Th;

[0025] Weight allocation method based on the short - board effect:

[0026] The satisfaction set is denoted as S: S = {s1, s2,..., sn}; where n is a positive integer representing the number of quantization metrics and requested resources; s1 to sn respectively represent the satisfaction degrees of the 1st to nth quantization metrics or requested resources; s1, s2,..., sn ∈ [0, 1].

[0027] The corresponding weight set W: W = {w1, w2,..., wn}; where w1 to wn respectively represent the weights corresponding to s1 to sn; w1, w2,..., wn ≥ 0.

[0028] Utilize the exponential function to amplify the influence of low-value terms: wi = e -si / (e -s1 + e -s2 + … + e -sn ); where i ∈ {1, 2, …, n}, si represents the satisfaction degree sequence of quantization metrics or requested resources, and wi represents the weight corresponding to si.

[0029] The total satisfaction degree of this service: St = s1 × w1 + s2 × w2 + … + sn × wn.

[0030] Calculate the satisfaction degree of each service, and calculate the average value of the satisfaction degrees of all services as the total satisfaction degree of this scheduling.

[0031] In step 2, the specific steps for calculating the network load balancing degree are as follows:

[0032] Step 2.1: Deploy monitoring tools at each node in the network to collect the load data of the nodes in real time.

[0033] The load data includes but is not limited to CPU usage rate, memory occupancy rate, and bandwidth utilization rate.

[0034] Step 2.2: Set the optimal load values for different types of load data.

[0035] Step 2.3: Construct the probability density function of the Beta distribution for the corresponding different types of load data according to different optimal load thresholds, and calculate the load balancing degree.

[0036] For a certain type of load data: its optimal load value is denoted as A, indicating that the load balancing degree of this type of load data is the highest at A; 0 < A < 1.

[0037] Construct the probability density function f(x, α, β) of the Beta distribution: Combine the optimal load value A of this type of load data and the mean μ or variance σ of the load data within a period of time 2 , and determine α and β; where x represents the actual value of the load data, and α and β are the parameters of the probability density function of the Beta distribution; α > 1 and β > 1.

[0038] Take the probability density function value of the Beta distribution corresponding to the actual load data of this type as the load balancing degree value of the load data of this type;

[0039] For a certain node, calculate the average value of the sum of the load balancing degree values of all types of load data it contains as the load balancing degree value of this node;

[0040] Step 2.4, calculate the total network load balancing degree;

[0041] Take the average value of the sum of the load balancing degree values of each node as the total network load balancing degree.

[0042] In step 3, the specific steps to predict future business demand data are as follows:

[0043] Step 3.1, collect historical business demand data;

[0044] Collect past business demand data from data sources such as the logs and databases of the business system, including specific metrics such as business traffic, transaction volume, and user access volume; record the timestamps corresponding to the data to ensure that the data has the characteristics of a time series;

[0045] Step 3.2, construct and train an LSTM time series analysis model;

[0046] Divide the data into a training set and a test set;

[0047] Construct the LSTM model structure: The staff determines the hyperparameters of the LSTM model according to the complexity of the data and the difficulty of the prediction task; the number of neurons in each layer can be adjusted through experiments to find the optimal model structure; at the same time, add an input layer and an output layer to convert the input data into a format suitable for LSTM processing and output the prediction results;

[0048] Input the training set data into the constructed LSTM model and train it using the optimization algorithm Adam and the mean squared error loss function; during the training process, continuously adjust the parameters of the model to minimize the value of the loss function, thereby improving the fitting ability of the model to historical data; the training process can be iterated multiple times until the performance of the model no longer improves significantly;

[0049] Step 3.3, use the trained model for prediction;

[0050] Input the test set data into the trained LSTM model, and the model will output the predicted values of future business demands according to the learned patterns and trends.

[0051] In step 4, define the constraint conditions and maximize the business demand satisfaction degree and the network balanced load degree. The specific implementation steps for formulating the network resource allocation strategy are as follows:

[0052] Step 4.1, define the optimization objective;

[0053] Assign different weights to the business demand satisfaction degree and the network balanced load degree and perform weighted summation to form a new optimization objective;

[0054] Step 4.2, define the constraint conditions;

[0055] Business resource allocation constraint:

[0056] Total resource constraint: The total amount of resources actually allocated to all services does not exceed the available total amount of this resource;

[0057] Network load balancing constraint:

[0058] Load adjustment strategy: When the load of a certain node is greater than the optimal load value, reduce the tasks allocated to it and migrate some tasks to the node with a lower load; if the load of a certain node is too low, increase the tasks allocated to it;

[0059] Step 4.3, use the constraint optimization algorithm to solve the specified network resource allocation strategy;

[0060] Use the multi-objective optimization algorithm and the penalty function method for iterative optimization, continuously adjust the demand allocation and load to maximize the optimization objective; formulate the network resource allocation strategy according to the maximized optimization objective.

[0061] A network resource allocation system based on network load, which includes a business demand satisfaction calculation module, a network load balancing evaluation module, a future demand prediction module, and a resource allocation strategy formulation module;

[0062] The business demand satisfaction calculation module is used to determine the business demand satisfaction degree;

[0063] The network load balancing evaluation module is used to calculate the network load balancing degree;

[0064] The future demand prediction module is used to predict future business demand data;

[0065] The resource allocation strategy formulation module is used to define the constraint conditions and maximize the business demand satisfaction degree and the network balanced load degree, and formulate the network resource allocation strategy;

[0066] The output end of the service demand satisfaction calculation module is connected to the input end of the resource allocation strategy formulation module; the output end of the network load balancing evaluation module is connected to the input end of the resource allocation strategy formulation module; the output end of the future demand prediction module is connected to the input end of the resource allocation strategy formulation module.

[0067] The service demand satisfaction calculation module includes a service classification and priority definition unit, a resource request and allocation data collection unit, and a satisfaction calculation and evaluation unit;

[0068] The service classification and priority definition unit is used to define service types and their priorities to form a classification standard; the resource request and allocation data collection unit is used to collect resource request data and actual allocation results of each service type; the satisfaction calculation and evaluation unit is used to calculate the single-service satisfaction according to resource allocation comparison and quantization indexes, and calculate the total satisfaction through the short-board effect weight allocation method;

[0069] The output end of the service classification and priority definition unit is connected to the input end of the resource request and allocation data collection unit; the output end of the resource request and allocation data collection unit is connected to the input end of the satisfaction calculation and evaluation unit; the output end of the satisfaction calculation and evaluation unit is connected to the input end of the resource allocation strategy formulation module.

[0070] The network load balancing evaluation module includes a load data real-time collection unit, an optimal load threshold setting unit, and a load balancing degree calculation unit;

[0071] The load data real-time collection unit is used to collect real-time load data of each node through a monitoring tool; the optimal load threshold setting unit is used to set an optimal load value for different types of load data; the load balancing degree calculation unit is used to calculate the single-node load balancing degree based on the Beta distribution and summarize the network load balancing degree through the node average value;

[0072] The output end of the load data real-time collection unit is connected to the input end of the optimal load threshold setting unit; the output end of the optimal load threshold setting unit is connected to the input end of the load balancing degree calculation unit; the output end of the load balancing degree calculation unit is connected to the input end of the resource allocation strategy formulation module.

[0073] The future demand prediction module includes a historical data collection unit, an LSTM model training unit, and a demand prediction generation unit;

[0074] The historical data collection unit is used to extract historical business requirement data and format it in a time series; the LSTM model training unit is used to build an LSTM model, train the model through an Adam optimizer and a mean squared error loss function, and adjust hyperparameters; the demand prediction generation unit is used to predict future business demands using the trained LSTM model;

[0075] The output end of the historical data collection unit is connected to the input end of the LSTM model training unit; the output end of the LSTM model training unit is connected to the input end of the demand prediction generation unit; the output end of the demand prediction generation unit is connected to the input end of the resource allocation strategy formulation module.

[0076] The resource allocation strategy formulation module includes an optimization objective definition unit, a constraint condition setting unit, and a strategy solution and generation unit;

[0077] The optimization objective definition unit is used to perform a weighted sum of the business demand satisfaction degree and the network load balancing degree to form an optimization objective function; the constraint condition setting unit is used to define the total resource constraint and the load adjustment strategy; the strategy solution and generation unit is used to solve the optimal resource allocation strategy using a multi-objective optimization algorithm and a penalty function method and output a specific allocation plan;

[0078] The output end of the optimization objective definition unit is connected to the input end of the constraint condition setting unit; the output end of the constraint condition setting unit is connected to the input end of the strategy solution and generation unit.

[0079] Compared with the prior art, the beneficial effects of the present invention are as follows: The weight allocation based on the short board effect in the present invention can more prominently highlight the key factors affecting the business, avoid the situation where some key indicators are masked by other better indicators due to non-satisfaction, and make the evaluation of the business demand satisfaction degree more accurate and reasonable; by constructing a Beta distribution probability density function, the present invention can more accurately measure the distribution of load data around the optimal load value, thereby more precisely evaluating the network load balancing degree and providing a more reliable basis for network resource allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 It is a schematic diagram of the steps of a network resource allocation method based on network load according to the present invention;

[0081] Figure 2 It is a schematic diagram of the process of a network resource allocation system based on network load according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0082] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0083] Embodiment: As Figure 1 - Figure 2 shown, the present invention provides a technical solution, a network resource allocation method based on network load. The method includes the following steps:

[0084] Step 1, determine the satisfaction degree of business requirements;

[0085] Step 2, calculate the network load balance degree;

[0086] Step 3, predict future business requirement data;

[0087] Step 4, define constraint conditions and maximize the satisfaction degree of business requirements and the network balanced load degree, and formulate a network resource allocation strategy.

[0088] In step 1, the specific steps for determining the satisfaction degree of business requirements are as follows:

[0089] Step 1.1, define the types of different services and their corresponding priorities;

[0090] The service types include but are not limited to core key services, sub-important services, and general services;

[0091] The priorities include but are not limited to the highest priority, high priority, medium priority, and low priority;

[0092] Step 1.2, collect resource request data and actual allocation situations of different service types;

[0093] The resource request data includes but is not limited to required bandwidth, computing resources, and storage resources;

[0094] Step 1.3, select quantization indicators and establish evaluation criteria according to different service types;

[0095] The quantization indicators include but are not limited to delay time, packet loss rate, average processing time, and resource utilization rate;

[0096] Set different quantization indicator conditions for different service types based on different priorities;

[0097] Step 1.4, compare the resource request data and actual allocation situations of each service type, and calculate the satisfaction degree of business requirements in combination with the corresponding quantization indicator conditions in the evaluation criteria;

[0098] For a certain actual allocation situation or quantitative index of a certain business type:

[0099] When the actual allocation situation Re meets the resource request, set the satisfaction degree to 1;

[0100] When the actual allocation situation Re does not meet the resource request condition threshold Ru, set the satisfaction degree to 1 - |Re - Ru| / Ru; where, |Re - Ru| represents the absolute value of the difference between the actual allocation situation Re and the resource request condition threshold Ru;

[0101] When the quantitative index T meets the quantitative index condition, set the satisfaction degree to 1;

[0102] When the quantitative index T does not meet the quantitative index condition threshold Th, set the satisfaction degree to 1 - |T - Th| / Th; where, |T - Th| represents the absolute value of the difference between the quantitative index T and the quantitative index threshold Th;

[0103] Weight allocation method based on the short - board effect:

[0104] The satisfaction degree set is represented as S: S = {s1, s2,..., sn}; where, n is a positive integer representing the number of quantitative indexes and requested resources; s1 to sn respectively represent the satisfaction degrees of the 1st to nth quantitative indexes or requested resources; s1, s2,..., sn ∈ [0, 1];

[0105] The corresponding weight set W: W = {w1, w2,..., wn}; where, w1 to wn respectively represent the weights corresponding to s1 to sn; w1, w2,..., wn ≥ 0;

[0106] Using the exponential function to amplify the influence of low - value items: wi = e -si / (e -s1 +e -s2 +…+e -sn ); where, i ∈ {1, 2,…, n}, si represents the satisfaction degree sequence of the quantitative index or requested resource, and wi represents the weight corresponding to si;

[0107] The total satisfaction degree of this business: St = s1×w1 + s2×w2 +…+ sn×wn;

[0108] Calculate the satisfaction degree of each business, and calculate the average value of all business satisfaction degrees as the total satisfaction degree of this scheduling.

[0109] In step 2, the specific steps to calculate the network load balancing degree are as follows:

[0110] Step 2.1, Deploy monitoring tools on each node in the network to collect the load data of the nodes in real - time;

[0111] The load data includes, but is not limited to, CPU usage rate, memory occupancy rate, and bandwidth utilization rate;

[0112] Step 2.2: Set the optimal load value for different types of load data;

[0113] Step 2.3: Construct the probability density function of the Beta distribution for the corresponding different types of load data according to different optimal load thresholds, and calculate the load balancing degree;

[0114] For a certain type of load data: its optimal load value is denoted as A, indicating that the load balancing degree of this type of load data is the highest at A; 0 < A < 1;

[0115] Construct the probability density function f(x,α,β) of the Beta distribution: Combine the optimal load value A of this type of load data and the mean μ or variance σ of the load data over a period of time 2 , to determine α and β; where x represents the actual value of the load data, and α and β are the parameters of the probability density function of the Beta distribution; α > 1 and β > 1;

[0116] Take the value of the probability density function of the Beta distribution corresponding to the actual load data of this type as the load balancing degree value of this type of load data;

[0117] For a certain node, calculate the average value of the sum of the load balancing degree values of all types of load data it contains as the load balancing degree value of this node;

[0118] Step 2.4: Calculate the total network load balancing degree;

[0119] Take the average value of the sum of the load balancing degree values of each node as the total network load balancing degree.

[0120] In step 3, the specific steps for predicting future business demand data are as follows:

[0121] Step 3.1: Collect historical business demand data;

[0122] Collect past business demand data from data sources such as the logs and databases of the business system, including specific metrics such as business traffic, transaction volume, and user access volume; record the corresponding timestamps of the data to ensure that the data has the characteristics of a time series;

[0123] Step 3.2: Construct and train an LSTM time series analysis model;

[0124] Divide the data into a training set and a test set;

[0125] Construct the LSTM model structure: The staff determines the hyperparameters of the LSTM model according to the complexity of the data and the difficulty of the prediction task; The number of neurons in each layer can be adjusted through experiments to find the optimal model structure; At the same time, an input layer and an output layer are added to convert the input data into a format suitable for LSTM processing and output the prediction results;

[0126] Input the training set data into the constructed LSTM model and train it using the optimization algorithm Adam and the mean squared error loss function; During the training process, continuously adjust the parameters of the model to minimize the value of the loss function, thereby improving the model's fitting ability to historical data; The training process can be iterated multiple times until the performance of the model no longer improves significantly;

[0127] Step 3.3: Use the trained model for prediction;

[0128] Input the test set data into the trained LSTM model, and the model will output the predicted value of future business needs according to the learned rules and trends.

[0129] In step 4, define the constraint conditions and maximize the business demand satisfaction degree and the network balanced load degree. The specific implementation steps for formulating the network resource allocation strategy are as follows;

[0130] Step 4.1: Define the optimization objective;

[0131] Assign different weights to the business demand satisfaction degree and the network balanced load degree for weighted summation to form a new optimization objective;

[0132] Step 4.2: Define the constraint conditions;

[0133] Business resource allocation constraint:

[0134] Total resource constraint: The total amount of resources actually allocated to all businesses does not exceed the available total amount of this resource;

[0135] Network load balancing constraint:

[0136] Load adjustment strategy: When the load of a certain node is greater than the optimal load value, reduce the tasks allocated to it and migrate some tasks to the node with a lower load; If the load of a certain node is too low, increase the tasks allocated to it;

[0137] Step 4.3: Use the constraint optimization algorithm to solve the specified network resource allocation strategy;

[0138] Use the multi-objective optimization algorithm and the penalty function method for iterative optimization, continuously adjust the demand allocation and load to maximize the optimization objective; Formulate the network resource allocation strategy according to the maximized optimization objective.

[0139] A network resource allocation system based on network load, which includes a service demand satisfaction calculation module, a network load balancing evaluation module, a future demand prediction module, and a resource allocation strategy formulation module;

[0140] The service demand satisfaction calculation module is used to determine the service demand satisfaction;

[0141] The network load balancing evaluation module is used to calculate the degree of network load balancing;

[0142] The future demand prediction module is used to predict future service demand data;

[0143] The resource allocation strategy formulation module is used to define constraint conditions and maximize the service demand satisfaction and the degree of network balanced load, and formulate a network resource allocation strategy;

[0144] The output end of the service demand satisfaction calculation module is connected to the input end of the resource allocation strategy formulation module; the output end of the network load balancing evaluation module is connected to the input end of the resource allocation strategy formulation module; the output end of the future demand prediction module is connected to the input end of the resource allocation strategy formulation module.

[0145] The service demand satisfaction calculation module includes a service classification and priority definition unit, a resource request and allocation data collection unit, and a satisfaction calculation and evaluation unit;

[0146] The service classification and priority definition unit is used to define the service type and its priority to form a classification standard; the resource request and allocation data collection unit is used to collect the resource request data and actual allocation results of each service type; the satisfaction calculation and evaluation unit is used to calculate the single-service satisfaction according to the resource allocation comparison and quantitative indicators, and calculate the total satisfaction through the short-board effect weight allocation method;

[0147] The output end of the service classification and priority definition unit is connected to the input end of the resource request and allocation data collection unit; the output end of the resource request and allocation data collection unit is connected to the input end of the satisfaction calculation and evaluation unit; the output end of the satisfaction calculation and evaluation unit is connected to the input end of the resource allocation strategy formulation module.

[0148] The network load balancing evaluation module includes a load data real-time collection unit, an optimal load threshold setting unit, and a load balancing degree calculation unit;

[0149] The real-time load data acquisition unit is used to collect the real-time load data of each node through a monitoring tool; the optimal load threshold setting unit is used to set the optimal load value for different types of load data; the load balancing degree calculation unit is used to calculate the load balancing degree of a single node based on the Beta distribution and summarize the network load balancing degree through the node average value;

[0150] The output end of the real-time load data acquisition unit is connected to the input end of the optimal load threshold setting unit; the output end of the optimal load threshold setting unit is connected to the input end of the load balancing degree calculation unit; the output end of the load balancing degree calculation unit is connected to the input end of the resource allocation strategy formulation module.

[0151] The future demand prediction module includes a historical data collection unit, an LSTM model training unit, and a demand prediction generation unit;

[0152] The historical data collection unit is used to extract historical business demand data and format it in a time series; the LSTM model training unit is used to build an LSTM model, train the model through the Adam optimizer and the mean squared error loss function, and adjust the hyperparameters; the demand prediction generation unit is used to predict future business demands using the trained LSTM model;

[0153] The output end of the historical data collection unit is connected to the input end of the LSTM model training unit; the output end of the LSTM model training unit is connected to the input end of the demand prediction generation unit; the output end of the demand prediction generation unit is connected to the input end of the resource allocation strategy formulation module.

[0154] The resource allocation strategy formulation module includes an optimization goal definition unit, a constraint condition setting unit, and a strategy solution and generation unit;

[0155] The optimization goal definition unit is used to perform a weighted sum of the business demand satisfaction degree and the network load balancing degree to form an optimization goal function; the constraint condition setting unit is used to define the total resource constraint and the load adjustment strategy; the strategy solution and generation unit is used to solve the optimal resource allocation strategy using a multi-objective optimization algorithm and a penalty function method and output a specific allocation plan;

[0156] The output end of the optimization goal definition unit is connected to the input end of the constraint condition setting unit; the output end of the constraint condition setting unit is connected to the input end of the strategy solution and generation unit.

[0157] In this embodiment, an e-commerce company is about to usher in an annual large-scale promotion event, and the business volume is expected to increase significantly during the event. The company's network system carries core key businesses (such as commodity transactions, payment processing), secondary important businesses (such as user information query, order status tracking) and general businesses (such as advertising display, in-site letter push). In order to ensure stable operation of the network during the event, reasonable network resource allocation is required.

[0158] Step 1: Determine the degree of satisfaction of business requirements;

[0159] Step 1.1: Define business types and priorities:

[0160] Core key businesses: commodity transactions and payment processing, with the highest priority.

[0161] Secondary important business: user information query, order status tracking, with high priority.

[0162] General business: advertising display, in-site message push, the priority is medium.

[0163] Step 1.2: Collect resource request data and actual allocations:

[0164] Core business: The required bandwidth is estimated to be 100Mbps, the computing resources are 50 CPU cores, and the storage resources are 100GB. The actual allocated bandwidth is 80Mbps, the CPU cores are 40, and the storage resources are 80GB.

[0165] Secondary business: Required bandwidth 30 Mbps, computing resources 20 CPU cores, storage resources 50 GB. Actual allocated bandwidth 25 Mbps, CPU cores 15, storage resources 40 GB.

[0166] General business: Required bandwidth 10Mbps, computing resources 5 CPU cores, storage resources 20GB. Actual allocated bandwidth 8Mbps, CPU cores 4, storage resources 15GB.

[0167] Step 1.3: Select quantitative indicators and establish evaluation criteria:

[0168] Delay time: The threshold for core critical services is 50ms, for secondary important services is 100ms, and for general services is 200ms.

[0169] Packet loss rate: The threshold for core critical services is 0.1%, for secondary important services is 0.5%, and for general services is 1%.

[0170] Average processing time: The threshold for core critical services is 100ms, for secondary important services is 200ms, and for general services is 500ms.

[0171] Resource utilization rate: The threshold for core key services is 80%, for less important services is 70%, and for general services is 60%.

[0172] Step 1.4: Calculate the service demand satisfaction degree:

[0173] Core key services:

[0174] Bandwidth satisfaction degree: The actual allocation is 80 Mbps, the request is 100 Mbps, and the satisfaction degree is 1 - |80 - 100| / 100 = 0.8.

[0175] Calculate the resource satisfaction degree: The actual allocation is 40 CPU cores, the request is 50, and the satisfaction degree is 1 - |40 - 50| / 50 = 0.8.

[0176] Storage resource satisfaction degree: The actual allocation is 80 GB, the request is 100 GB, and the satisfaction degree is 1 - |80 - 100| / 100 = 0.8.

[0177] Latency time satisfaction degree: Assume the actual latency time is 60 ms, and the satisfaction degree is 1 - |60 - 50| / 50 = 0.8.

[0178] Packet loss rate satisfaction degree: Assume the actual packet loss rate is 0.2%, and the satisfaction degree is 1 - |0.2% - 0.1%| / 0.1% = 0.

[0179] Average processing time satisfaction degree: Assume the actual average processing time is 120 ms, and the satisfaction degree is 1 - |120 - 100| / 100 = 0.8.

[0180] Resource utilization rate satisfaction degree: Assume the actual resource utilization rate is 70%, and the satisfaction degree is 1 - |70% - 80%| / 80% = 0.875.

[0181] Satisfaction degree set S = {0.8, 0.8, 0.8, 0.8, 0, 0.8, 0.875}.

[0182] Weight set W:

[0183] w1 = e^(-0.8) / (e^(-0.8) + e^(-0.8) + e^(-0.8) + e^(-0.8) + e^(0) + e^(-0.8) + e^(-0.875)) ≈ 0.12.

[0184] Calculate other weights by analogy.

[0185] Total satisfaction degree St = 0.8 × 0.12 + 0.8 × 0.12 + 0.8 × 0.12 + 0.8 × 0.12 + 0 × 0.12 + 0.8 × 0.12 + 0.875 × 0.12 ≈ 0.67.

[0186] Calculate the satisfaction degrees of the second most important business and general business in the same way, which are 0.75 and 0.8 respectively.

[0187] The average value of the total satisfaction degree = (0.67 + 0.75 + 0.8) / 3 ≈ 0.74.

[0188] Step 2: Calculate the network load balancing degree;

[0189] Step 2.1: Deploy monitoring tools on 5 nodes in the network to collect load data in real time. For example, the CPU usage rate of node 1 is 70%, the memory occupancy rate is 60%, and the bandwidth utilization rate is 50%.

[0190] Step 2.2: Set the optimal load values:

[0191] The optimal load value A1 of the CPU usage rate = 0.6.

[0192] The optimal load value A2 of the memory occupancy rate = 0.5.

[0193] The optimal load value A3 of the bandwidth utilization rate = 0.4.

[0194] Step 2.3: Construct the probability density function of the Beta distribution:

[0195] For the CPU usage rate, the mean value μ = 0.65 and the variance σ2 = 0.01 within a period of time. Determine α = 3 and β = 2 through the formula.

[0196] For the CPU usage rate of node 1, the actual value x = 0.7, and its load balancing degree value is f(0.7, 3, 2).

[0197] Calculate the load balancing degree values of the memory occupancy rate and the bandwidth utilization rate in the same way, which are 0.75 and 0.85 respectively.

[0198] The load balancing degree value of node 1 = (0.8 + 0.75 + 0.85) / 3 = 0.8.

[0199] Calculate the load balancing degree values of other nodes, which are 0.78, 0.82, 0.75, and 0.8 respectively.

[0200] The total network load balancing degree = (0.8 + 0.78 + 0.82 + 0.75 + 0.8) / 5 = 0.79.

[0201] Step 3: Predict future business demand data;

[0202] Step 3.1: Collect business demand data in the past year from the business system logs and databases, such as the daily number of commodity transactions, user access volume, etc., and record the timestamps.

[0203] Step 3.2: Construct and train an LSTM time series analysis model: Divide the data with 70% as the training set and 30% as the test set. Construct the LSTM model structure: Assume that 3 LSTM layers are determined, and the number of neurons in each layer is 64, 32, and 16 respectively. Add an input layer and an output layer. Use the Adam optimization algorithm and the mean squared error loss function for training. After 50 rounds of iteration, the model performance no longer improves significantly.

[0204] Step 3.3: Input the test set data into the trained model to predict the daily business demand data for the next week.

[0205] Step 4: Define the constraint conditions and maximize the business demand satisfaction and network balanced load level, and formulate a network resource allocation strategy;

[0206] Step 4.1: Define the optimization objective: Assume that the weight of business demand satisfaction is 0.6 and the weight of network balanced load level is 0.4. Optimization objective = 0.6 × business demand satisfaction + 0.4 × network balanced load level.

[0207] Step 4.2: Define the constraint conditions: Business resource allocation constraint: Assume that the total company bandwidth is 200 Mbps, the total number of CPU cores is 100, and the total storage resource is 200 GB. The resources actually allocated to all businesses cannot exceed these totals.

[0208] Network load balancing constraint: When the node load is greater than the optimal load value, such as the CPU usage rate of node 1 is greater than 0.6, reduce its assigned tasks and migrate some tasks to nodes with lower load. When the node load is too low, increase the assigned tasks.

[0209] Step 4.3: Use the constraint optimization algorithm to solve: Use the multi-objective optimization algorithm and the penalty function method for iterative optimization. For example, after multiple iterative adjustments, it is found that increasing the core key business bandwidth allocation to 90 Mbps and adjusting the task allocation of some nodes can maximize the optimization objective.

[0210] Formulate a network resource allocation strategy according to the maximized optimization objective: Increase the core key business bandwidth to 90 Mbps, calculate the resources to 45 CPU cores, and the storage resources to 90 GB.

[0211] Adjust the resource allocation of less important and general businesses to ensure that the total resources do not exceed the limit.

[0212] Migrate tasks for network nodes to make the node load closer to the optimal load value.

[0213] Through the above embodiments, the data processing process and specific implementation steps of the network resource allocation method based on network load in the actual e-commerce business scenario are demonstrated, which can effectively improve the reasonable utilization of network resources and the stability of the business.

[0214] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above-described exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, in any regard, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Thus, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be embraced within the present invention. Any reference signs in the claims should not be construed as limiting the claims involved.

Claims

1. A network resource allocation method based on network load, characterized in that: The method includes the following steps: Step 1: Determine the satisfaction degree of business requirements; Step 2: Calculate the network load balancing degree; Step 3: Predict future business requirement data; Step 4: Define constraint conditions and maximize the satisfaction degree of business requirements and the network balanced load degree, and formulate a network resource allocation strategy.

2. The network resource allocation method based on network load according to claim 1, characterized in that: In Step 1, the specific steps for determining the satisfaction degree of business requirements are: Step 1.1: Define the types of different services and their corresponding priorities; Step 1.2: Collect resource request data and actual allocation situations of different service types; Step 1.3: Select quantitative indicators and establish evaluation criteria according to different service types; Set different quantitative indicator conditions for different service types based on different priorities; Step 1.4: Compare the resource request data and actual allocation situations of each service type, and calculate the satisfaction degree of business requirements by combining the corresponding quantitative indicator conditions in the evaluation criteria; For a certain actual allocation situation or quantitative indicator of a certain service type: When the actual allocation situation Re meets the resource request, set the satisfaction degree to 1; When the actual allocation situation Re does not meet the resource request condition threshold Ru, set the satisfaction degree to 1 - |Re - Ru| / Ru; where Ru represents the lowest lower limit or the highest upper limit of the resource request; |Re - Ru| represents the absolute value of the difference between the actual allocation situation Re and the resource request condition threshold Ru; When the quantitative indicator T meets the quantitative indicator condition, set the satisfaction degree to 1; When the quantitative indicator T does not meet the quantitative indicator condition threshold Th, set the satisfaction degree to 1 - |T - Th| / Th; where Th represents the lowest lower limit or the highest upper limit of the quantitative indicator; |T - Th| represents the absolute value of the difference between the quantitative indicator T and the quantitative indicator threshold Th; Weight allocation method based on the short board effect: The satisfaction degree set is expressed as S: S = {s1, s2,..., sn}; where n is a positive integer representing the number of quantitative indicators and requested resources; s1 to sn respectively represent the satisfaction degrees of the 1st to nth quantitative indicators or requested resources; s1, s2,..., sn ∈ [0, 1]; The corresponding weight set W: W = {w1, w2,..., wn}; where w1 to wn respectively represent the weights corresponding to s1 to sn; w1, w2,..., wn ≥ 0; Amplify the influence of low-value terms using the exponential function: wi = e -si / (e -s1 + e -s2 + … + e -sn ); where i ∈ {1, 2, …, n}, si represents the quantization index or the sequence of request resource satisfaction degrees, wi represents the weight corresponding to si; e represents the natural constant; The total satisfaction degree of this service: St = s1×w1 + s2×w2 + … + sn×wn; Calculate the satisfaction degree of each service, and calculate the average value of all service satisfaction degrees as the total satisfaction degree of this scheduling.

3. The network resource allocation method based on network load according to claim 2, wherein: In Step 2, the specific steps for calculating the network load balancing degree are: Step 2.1: Deploy monitoring tools at each node in the network to collect the load data of the nodes in real time; Step 2.2: Set the optimal load value for different types of load data; Step 2.3: Construct the probability density function of the Beta distribution for the corresponding different types of load data according to different optimal load thresholds, and calculate the load balancing degree; For a certain type of load data: its optimal load value is expressed as A, indicating that the load balancing degree of this type of load data is the highest at A; 0＜A＜1； Construct the probability density function f(x, α, β) of the Beta distribution: Combine the optimal load value A of this type of load data and the mean μ or variance σ of the load data over a period of time 2 , and determine α and β; where x represents the actual value of the load data, and α and β are the parameters of the probability density function of the Beta distribution; α > 1 and β > 1; Take the probability density function value of the Beta distribution corresponding to the actual load data of this type as the load balancing degree value of the load data of this type; For a certain node, calculate the average value of the sum of the load balancing degree values of all types of load data it contains as the load balancing degree value of this node; Step 2.4, calculate the total network load balancing degree; Take the average value of the sum of the load balancing degree values of each node as the total network load balancing degree.

4. A network resource allocation method based on network load according to claim 3, characterized in that: In step 3, the specific steps for predicting future business demand data are as follows: Step 3.1, collect historical business demand data; Collect past business demand data; record the corresponding timestamps of the data to ensure that the data has the characteristics of a time series; Step 3.2, construct and train an LSTM time series analysis model; Divide the data into a training set and a test set; construct the LSTM model structure; add an input layer and an output layer; input the training set data into the constructed LSTM model and train it using the optimization algorithm Adam and the mean square error loss function; During the training process, continuously adjust the parameters of the model and perform multiple rounds of iteration to minimize the value of the loss function; Step 3.3, use the trained model for prediction; Input the test set data into the trained LSTM model, and the model outputs the predicted values of future business demands.

5. A network resource allocation method based on network load according to claim 4, characterized in that: In step 4, the specific implementation steps for defining the constraint conditions and maximizing the business demand satisfaction degree and the network balanced load degree and formulating the network resource allocation strategy are as follows; Step 4.1, define the optimization objective; Assign different weights to the business demand satisfaction degree and the network balanced load degree for weighted summation to form a new optimization objective; Step 4.2, define the constraint conditions; Business resource allocation constraint: Total resource constraint: The total amount of resources actually allocated to all services does not exceed the available total amount of this resource; Network load balancing constraint: Load adjustment strategy: When the load of a certain node is greater than the optimal load value, reduce the tasks assigned to it and migrate some tasks to nodes with lower loads; when the load of a certain node is less than the optimal load value, increase the assigned tasks; Step 4.3, use the constraint optimization algorithm to solve the specified network resource allocation strategy; Use the multi-objective optimization algorithm and the penalty function method for iterative optimization, continuously adjust the demand allocation and load to maximize the optimization objective; formulate the network resource allocation strategy according to the maximized optimization objective.

6. A network resource allocation system based on network load, which is applied to any one of the network resource allocation methods based on network load described in claims 1-5, and is characterized in that: The system includes a business demand satisfaction degree calculation module, a network load balancing evaluation module, a future demand prediction module, and a resource allocation strategy formulation module; The business demand satisfaction degree calculation module is used to determine the business demand satisfaction degree; The network load balancing evaluation module is used to calculate the network load balancing degree; The future demand prediction module is used to predict future business demand data; The resource allocation strategy formulation module is used to define the constraint conditions and maximize the business demand satisfaction degree and the network balanced load degree, and formulate the network resource allocation strategy; The output end of the service requirement satisfaction calculation module is connected to the input end of the resource allocation strategy formulation module; the output end of the network load balancing evaluation module is connected to the input end of the resource allocation strategy formulation module; the output end of the future requirement prediction module is connected to the input end of the resource allocation strategy formulation module.

7. A network resource allocation system based on network load according to claim 6, characterized in that: The service requirement satisfaction calculation module includes a service classification and priority definition unit, a resource request and allocation data collection unit, and a satisfaction calculation and evaluation unit; The service classification and priority definition unit is used to define service types and their priorities to form classification criteria; the resource request and allocation data collection unit is used to collect resource request data and actual allocation results of each service type; the satisfaction calculation and evaluation unit is used to calculate the satisfaction of a single service according to resource allocation comparison and quantization indexes, and calculate the total satisfaction through the short-board effect weight allocation method; The output end of the service classification and priority definition unit is connected to the input end of the resource request and allocation data collection unit; the output end of the resource request and allocation data collection unit is connected to the input end of the satisfaction calculation and evaluation unit; the output end of the satisfaction calculation and evaluation unit is connected to the input end of the resource allocation strategy formulation module.

8. A network resource allocation system based on network load according to claim 7, characterized in that: The network load balancing evaluation module includes a load data real-time collection unit, an optimal load threshold setting unit, and a load balancing degree calculation unit; The load data real-time collection unit is used to collect real-time load data of each node through a monitoring tool; the optimal load threshold setting unit is used to set optimal load values for different types of load data; the load balancing degree calculation unit is used to calculate the load balancing degree of a single node based on the Beta distribution and summarize the network load balancing degree through the node average value; The output end of the load data real-time collection unit is connected to the input end of the optimal load threshold setting unit; the output end of the optimal load threshold setting unit is connected to the input end of the load balancing degree calculation unit; the output end of the load balancing degree calculation unit is connected to the input end of the resource allocation strategy formulation module.

9. The network resource allocation system based on network load according to claim 8, wherein: The future requirement prediction module includes a historical data collection unit, an LSTM model training unit, and a requirement prediction generation unit; The historical data collection unit is used to extract historical service requirement data and format it in a time series; the LSTM model training unit is used to construct an LSTM model, train the model through the Adam optimizer and the mean squared error loss function, and adjust hyperparameters; the requirement prediction generation unit is used to predict future service requirements using the trained LSTM model; The output end of the historical data collection unit is connected to the input end of the LSTM model training unit; the output end of the LSTM model training unit is connected to the input end of the requirement prediction generation unit; the output end of the requirement prediction generation unit is connected to the input end of the resource allocation strategy formulation module.

10. The network resource allocation system based on network load according to claim 9, wherein: The resource allocation strategy formulation module includes an optimization goal definition unit, a constraint condition setting unit, and a strategy solution and generation unit; The optimization objective definition unit is used to perform a weighted sum of the business requirement satisfaction degree and the network load balancing degree to form an optimization objective function; the constraint condition setting unit is used to define the total resource constraint and the load adjustment strategy; the strategy solution and generation unit is used to solve the optimal resource allocation strategy using a multi-objective optimization algorithm and a penalty function method, and output a specific deployment plan. The output end of the optimization objective definition unit is connected to the input end of the constraint condition setting unit; the output end of the constraint condition setting unit is connected to the input end of the strategy solution and generation unit.

Images