WO2021012930A1

WO2021012930A1 - Voting node configuration method and system

Info

Publication number: WO2021012930A1
Application number: PCT/CN2020/100099
Authority: WO
Inventors: 帖军; 黄坤; 李子茂; 宋中山; 尹帆; 马尧; 罗均
Original assignee: 中南民族大学
Priority date: 2019-07-23
Filing date: 2020-07-03
Publication date: 2021-01-28
Also published as: CN110417591B; CN110417591A

Abstract

The present application discloses a voting node configuration method and system, the method comprising: a master node regularly issuing parameter acquisition tasks to at least two slave nodes in a distributed cluster; each of the at least two slave nodes acquiring a performance index parameter and feeding back same to the master node; the master node performing weight calculation on respective performance index parameters on the basis of a preset combined algorithm, so as to obtain index weights corresponding to the respective performance index parameters; calculating voting powers of the respective slave nodes according to the index weights and parameter values corresponding to the respective performance index parameters; and determining, according to a preset allowable number of crashes, the number of voting nodes corresponding to the distributed cluster, and selecting, according to the number of voting nodes and the voting powers, corresponding voting nodes from the at least two slave nodes.

Description

Voting node configuration method and system

Cross references to related applications

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office, application number 201910670424.1, and application name "Voting Node Configuration Method and System" on July 23, 2019, the entire content of which is incorporated into the application by reference.

Technical field

This application relates to the field of computer technology, and in particular to a method and system for configuring voting nodes.

Background technique

At present, the message broadcast mode in distributed clusters (such as ZooKeeper clusters) basically adopts the "over-half" strategy. This strategy is that ZooKeeper makes a trade-off between availability and consistency to ensure that even if less than half of the servers in the cluster are down, the cluster Can still provide external services. In a distributed cluster, when the master node (that is, the master server) frequently initiates transactional requests, since ZooKeeper processes transactional operations in order, the master node will only process the next request after the current transactional operation is completed. . Therefore, once the ZooKeeper cluster is larger, the more votes that need to be more than half of the number of votes, the longer it takes for more than half of the votes, and the lower the performance of the ZooKeeper set for transactional operations.

When there are multiple nodes (ie, servers) in a distributed cluster, there is no uniform configuration standard for how to configure voting nodes and non-voting nodes. If a subjective configuration scheme is adopted, the cluster synchronization may not achieve the best performance. In order to find the best configuration scheme, the operation and maintenance personnel must go through repeated experiments and quantitatively analyze the data to obtain the results. Although this solution is feasible, since network fluctuations are frequent, frequent manual configuration is not the best method.

The above content is only used to assist the understanding of the technical solution of this application, and does not mean that the above content is recognized as prior art.

Application content

The main purpose of this application is to provide a voting node configuration method and system, which aims to solve the technical problem that the prior art cannot effectively configure voting nodes in a distributed cluster.

In order to achieve the above objective, this application provides a voting node configuration method, which includes the following steps:

The master node issues parameter collection tasks to at least 2 slave nodes in the distributed cluster where it is located every preset time period;

Each of the at least two slave nodes collects performance index parameters of a preset dimension according to the parameter collection task, and feeds back the collected performance index parameters to the master node;

The master node calculates the weight of each performance index parameter based on a preset combination algorithm to obtain the index weight corresponding to each performance index parameter;

The master node respectively calculates the voting ability corresponding to each slave node according to the index weight and the parameter value corresponding to each performance index parameter; and

The master node determines the number of voting nodes corresponding to the distributed cluster according to a preset allowable number of downtimes, and selects corresponding votes from the at least two slave nodes according to the number of voting nodes and the voting capacity node.

In an embodiment, the step of the slave node collecting performance index parameters of a preset dimension according to the parameter collection task, and feeding back the collected performance index parameters to the master node includes the following steps:

Create a target file locally according to the parameter collection task, perform read and write operations on the target file within a preset time period, and calculate the corresponding disk read rate according to the total number of reads and writes counted;

Reading the processor computing capability parameters included in the parameter collection task, where the processor computing capability parameters include: a calculation time limit and a value to be calculated;

Perform a number of prime number calculation operations on the value to be calculated within the calculation time limit, and obtain the corresponding processor computing power according to the execution result; and

The disk read rate and the processor computing power are fed back to the master node as performance index parameters.

In one embodiment, the step of calculating the weight of each performance index parameter by the master node based on a preset combination algorithm to obtain the index weight corresponding to each performance index parameter includes the following steps:

Obtain the subjective weight value corresponding to each performance index parameter through the preset analytic hierarchy process;

Obtain the objective weight value corresponding to each performance index parameter through the preset weight assignment algorithm; and

Based on the subjective weight value and the objective weight value, the index weight corresponding to each performance index parameter is obtained through a preset Lagrangian optimal multiplier method.

In an embodiment, the step of obtaining the subjective weight value corresponding to each performance index parameter by the master node through a preset analytic hierarchy process includes the following steps:

Construct a corresponding judgment matrix according to each performance index parameter, and calculate the maximum eigenvalue and eigenvector of the judgment matrix;

Acquiring the order corresponding to the judgment matrix, and searching for the random consistency evaluation index value corresponding to the order in a preset random consistency index table;

Calculating the target consistency index value corresponding to the judgment matrix according to the maximum characteristic root and the order;

Judging whether the judgment matrix is valid according to the random consistency evaluation index value and the target consistency index value; and

When the judgment matrix is valid, the element value corresponding to each vector element in the eigenvector is read, and the subjective weight value corresponding to each performance index parameter is determined according to the read element value.

In an embodiment, the step of obtaining the objective weight value corresponding to each performance index parameter by the master node through a preset weight assignment algorithm includes the following steps:

Dimensionlessly transform each performance index parameter to obtain a dimensionless performance index, and obtain the correlation coefficient between every two dimensionless performance indexes;

Obtain the standard deviation corresponding to each non-dimensional performance index, calculate the information amount corresponding to each non-dimensional performance index through the first preset formula according to the standard deviation and the correlation coefficient, and add the information amount to obtain the total information ;as well as

The objective weight value corresponding to each performance index parameter is determined according to the amount of information corresponding to each dimensionless performance index and the total amount of information; wherein, the first preset formula is:

In the formula, G _i is the amount of information corresponding to the dimensionless performance index i, σ _i is the standard deviation corresponding to the dimensionless performance index i, and r _ij is the correlation coefficient between the dimensionless performance index i and j.

In an embodiment, the step of obtaining the index weight corresponding to each performance index parameter by the master node based on the subjective weight value and the objective weight value through a preset Lagrangian optimal multiplier method includes the following steps:

Based on the subjective weight value and the objective weight value, the index weight corresponding to each performance index parameter is calculated by a second preset formula; wherein, the second preset formula is:

In the formula, W _i is the index weight corresponding to the i-th performance index parameter, n is the number of performance index parameters, W _i ^A is the subjective weight value, and W _i ^C is the objective weight value.

In an embodiment, the step of the master node separately calculating the voting ability corresponding to each slave node according to the index weight and the parameter value corresponding to each performance index parameter includes the following steps:

According to the index weight and the parameter value corresponding to each performance index parameter, the voting ability corresponding to each slave node is calculated through a third preset formula; wherein, the third preset formula is:

B=(a ₁ , a ₂ …a _n )×[W ₁ ,W ₂ …W _n ]

Wherein, B is the ability to vote, a _n is the n performance metrics parameters, W _n is the n performance indicators corresponding to the heavy weight parameter index.

In an embodiment, the master node determines the number of voting nodes corresponding to the distributed cluster according to a preset allowable number of downtimes, and according to the number of voting nodes and the voting capability, the at least two slaves The steps of selecting the corresponding voting node from the nodes include the following steps:

Determine the number of voting nodes corresponding to the distributed cluster according to the preset allowable number of downtimes; and

Sort the voting capabilities in descending order, and select slave nodes of the number of voting nodes as voting nodes according to the sorting result.

In addition, in order to achieve the above object, this application also proposes a voting node configuration system, the system includes: a master node and at least two slave nodes, the master node and the at least two slave nodes are in a distributed cluster;

The master node is configured to deliver parameter collection tasks to at least two slave nodes in the distributed cluster every preset time period;

Each of the at least two slave nodes is configured to collect performance index parameters of a preset dimension according to the parameter collection task, and feed back the collected performance index parameters to the master node;

The master node is also used to calculate the weight of each performance index parameter based on a preset combination algorithm to obtain the index weight corresponding to each performance index parameter;

The master node is also used to calculate the voting power corresponding to each slave node according to the index weight and the parameter value corresponding to each performance index parameter; and

The master node is also used to determine the number of voting nodes corresponding to the distributed cluster according to a preset allowable number of downtimes, and to select from the at least two slave nodes according to the number of voting nodes and the voting capacity Select the corresponding voting node.

In an embodiment, the master node is further configured to calculate the weight of each performance index parameter based on a preset combination algorithm to obtain the index weight corresponding to each performance index parameter, including:

In this application, the master node sends parameter collection tasks to at least two slave nodes in the distributed cluster at a preset time period; the slave nodes collect the performance index parameters of the preset dimensions according to the parameter collection task, and the collected performance indicators The parameters are fed back to the master node; the master node calculates the weight of each performance index parameter based on the preset combination algorithm to obtain the index weight corresponding to each performance index parameter; calculates each slave separately according to the index weight and the parameter value corresponding to each performance index parameter The voting capacity corresponding to the node; then the number of voting nodes corresponding to the distributed cluster is determined according to the preset allowable number of downtimes, and the corresponding voting node is selected according to the number of voting nodes and voting capacity. Since the voting capabilities of at least 2 slave nodes are calculated according to the performance index parameters of the slave nodes and the performance index parameter index weights determined by the combination algorithm, and then the voting nodes are screened according to the voting ability, it can ensure that the selected slave nodes have higher Good work performance also realizes the automatic configuration of voting nodes.

Description of the drawings

FIG. 1 is a schematic flowchart of a first embodiment of a method for configuring a voting node of the application;

FIG. 2 is a schematic flowchart of a second embodiment of a method for configuring a voting node of the application;

FIG. 3 is a schematic flowchart of a third embodiment of a voting node configuration method of the application;

FIG. 4 is a schematic diagram of the voting ability evaluation hierarchical relationship model of the third embodiment of the application voting node configuration method;

Fig. 5 is a structural block diagram of the first embodiment of the voting node system device of this application.

The realization, functional characteristics, and advantages of the purpose of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings.

Detailed ways

It should be understood that the specific embodiments described herein are only used to explain the application, and not used to limit the application.

An embodiment of the present application provides a method for configuring a voting node. Referring to FIG. 1, FIG. 1 is a schematic flowchart of a first embodiment of a method for configuring a voting node in this application.

In this embodiment, the voting node configuration method includes the following steps:

Step S10: The master node delivers parameter collection tasks to at least two slave nodes in the distributed cluster where it is located every preset time period.

It should be noted that the execution subject of this embodiment may be the Leader server (ie, the master node) in a distributed application coordination service (Zookeeper) cluster. In a distributed cluster, the master node is the core of the entire ZooKeeper cluster working mechanism. . The slave node is the other nodes in the same distributed cluster except the master node. The parameter collection task may be a parameter collection request or instruction initiated by the master node.

In this embodiment, the preset time period may be a preset time interval for issuing parameter collection tasks. Taking into account the actual situation, the computing power and disk read speed of each slave node does not change much, and the network delay is relatively changed by network fluctuations. Therefore, if the preset time period is set too small, and the frequency of collecting network delay, computing power, and disk reading speed is too high, it will cause unnecessary performance consumption and increase the server load. Therefore, in this embodiment, the preset time period can be set to 2 hours. Of course, the specific value of the preset time period is not limited in this embodiment.

In a specific implementation, the master node delivers a parameter collection task to at least two slave nodes in the distributed cluster every preset time period, so that each slave node responds to the parameter collection task when it receives the parameter collection task.

Step S20: Each of the at least two slave nodes collects performance index parameters of a preset dimension according to the parameter collection task, and feeds back the collected performance index parameters to the master node.

It should be noted that the performance index parameters of the preset dimensions in this embodiment include network latency, CPU computing power, and disk read speed. Because the network delay needs to be calculated by the master node. Therefore, when the slave node receives the parameter collection task in this step, it mainly collects the performance index parameters of the two dimensions of CPU computing power and disk reading speed.

In one embodiment, considering that SysBench is an open source, cross-platform multi-threaded performance testing tool that can be used for CPU, disk I/0, memory, and database, in this embodiment, the slave node can call the SysBench command through Java language to obtain CPU computing power and disk read speed.

Specifically, in this embodiment, the slave node may locally create a target file according to the parameter collection task, perform read and write operations on the target file within a preset time period, and calculate the corresponding disk based on the total number of reads and writes. Reading rate; where the read and write operations may include: sequential writing, sequential rewriting, sequential reading, and random reading.

When collecting CPU computing power, the slave node in this embodiment can also read the processor computing power parameters included in the parameter collection task. The processor computing power parameters include: calculation time limit and values to be calculated; Perform several prime number calculation operations on the to-be-calculated value within the time limit, and obtain the corresponding processor computing power according to the execution result; then, the disk read rate and the processor computing power are fed back as performance index parameters to all The main node.

It should be understood that the prime number obtaining operation is to calculate the prime number of a certain number. The calculation time limit is the time range for the prime number calculation. For example, if the value to be calculated is a prime number of 100, the calculation time limit is 30 seconds.

Considering that the network delay needs to be calculated by the master node, the master node in this embodiment can integrate the same ping command in the issued parameter collection task, and then record the sending time t _{1 of} the ping command (that is, the parameter collection task Sending time) and the receiving time t ₄ of receiving the feedback result of the ping command, and then acquiring the record of each slave node and feeding it back to the master node, the receiving time t ₂ of receiving the ping command, and the feedback when the response to the ping command ends The return time t ₃ when the result is returned to the client. Finally, the master node can calculate the current network delay of each slave node according to the formula "network delay=(t ₂ -t ₁ )+(t ₄ -t ₃ )" Time.

In specific implementation, each slave node collects performance index parameters of a preset dimension according to the parameter collection task, and feeds back the collected performance index parameters to the master node, and the master node performs subsequent voting according to these performance index parameters Capacity calculation.

Step S30: The master node calculates the weight of each performance index parameter based on the preset combination algorithm to obtain the index weight corresponding to each performance index parameter.

It should be noted that the preset combination algorithm in this embodiment may be a weight calculation strategy obtained by combining the Analytic Hierarchy Process (AHP) and the weight assignment algorithm (CRITIC assignment method).

Taking into account, the advantage of subjective weighting methods such as AHP is that the calculation process is simple, but the disadvantage is that the subjectivity is relatively large. The advantage of objective weighting methods such as CRITIC assignment method is that it has strong objectivity, but the disadvantage is that it relies too much on the statistical or mathematical quantitative methods of the sample, and ignores the qualitative analysis of evaluation indicators. Therefore, this embodiment combines the two algorithms, that is, the subjective weighting method and the objective weighting method are combined to calculate the index weight. Specifically, the subjective weight value corresponding to each performance indicator parameter can be obtained through the preset analytic hierarchy process; then the objective weight value corresponding to each performance indicator parameter is obtained through the preset weight assignment algorithm; and then based on the subjective weight value and the objective weight Value to obtain the index weight corresponding to each performance index parameter.

In the specific implementation, after the master node obtains the performance index parameters, it can calculate the weight of each performance index parameter based on the preset combination algorithm, so as to obtain the performance index parameters (network delay, CPU computing power and disk read Speed, etc.) corresponding to the index weight.

Step S40: The master node calculates the voting ability corresponding to each slave node according to the index weight and the parameter value corresponding to each performance index parameter.

It should be understood that the so-called voting ability is mainly used to characterize the data synchronization efficiency of nodes, and nodes with strong voting ability have relatively high corresponding data synchronization and processing efficiency.

In specific implementation, after the master node calculates the index weight corresponding to each performance index parameter, it can calculate the voting ability corresponding to each slave node according to the index weight and the parameter value corresponding to each performance index parameter.

Step S50: The master node determines the number of voting nodes corresponding to the distributed cluster according to the preset allowable number of downtimes, and selects corresponding votes from the slave nodes according to the number of voting nodes and the voting capacity node.

It should be understood that the allowable number of downtimes refers to the number of nodes that are allowed to fail in a distributed cluster. Normally, the number of nodes that can work normally in a distributed cluster needs to account for half of the total number of cluster nodes plus one, that is, (n/2)+1, where n is the total number of nodes. Therefore, the number of allowed downtimes in this embodiment cannot exceed (n/2)-1, the number of voting nodes is at least (n/2)+1, and the specific value is not limited.

In specific implementation, the master node determines the number of voting nodes corresponding to the distributed cluster according to the preset allowable number of downtimes; then, the voting capacity is sorted in descending order, and selected according to the sorting result. The slave nodes with the stated number of voting nodes are regarded as voting nodes.

In this embodiment, the master node delivers parameter collection tasks to at least two slave nodes in the distributed cluster every preset time period; each slave node of the at least two slave nodes collects parameters of preset dimensions according to the parameter collection task. Performance index parameters, and feedback the collected performance index parameters to the master node; the master node calculates the weight of each performance index parameter based on the preset combination algorithm to obtain the index weight corresponding to each performance index parameter; according to the index weight and each performance index The parameter value corresponding to the parameter calculates the voting capacity corresponding to each slave node; then the number of voting nodes corresponding to the distributed cluster is determined according to the preset allowable downtime number, and the corresponding voting node is selected according to the number of voting nodes and voting capacity. Since the voting capacity of each slave node is calculated according to the performance index parameters of each slave node and the performance index parameter index weight determined by the combination algorithm, and then voting nodes are screened according to the voting ability, it can ensure that the selected slave nodes have Better working performance, also realized the automatic configuration of voting nodes.

Referring to FIG. 2, FIG. 2 is a schematic flowchart of a second embodiment of a voting node configuration method of this application.

Based on the foregoing first embodiment, in this embodiment, the step S30 includes:

Step S301: The master node obtains the subjective weight value corresponding to each performance index parameter through the preset analytic hierarchy process.

It should be understood that the analytic hierarchy process usually divides the decision-making problem into the target layer, the criterion layer and the scheme layer, and classifies and establishes a hierarchical relationship model according to the nature of each factor. In the hierarchical relationship structure, the target level is the highest level, which represents the purpose of solving the problem, that is, the ultimate goal that the level analysis needs to achieve. The target layer in this embodiment is the voting capability of each ZooKeeper server (ie, slave node). The criterion layer is the middle layer, which represents the intermediate links needed to take certain measures and programs to achieve the ultimate goal. The criterion layer in this embodiment mainly includes system performance indicators and network performance indicators. The plan layer is the lowest layer, representing the objects of evaluation, ranking, and selection. After establishing the above-mentioned hierarchical relationship model, the master node will also construct a judgment matrix, and then perform weight calculations based on the constructed judgment matrix.

Specifically, the master node can construct a corresponding judgment matrix according to each performance index parameter, and calculate the maximum eigenvalue and eigenvector of the judgment matrix; then obtain the order corresponding to the judgment matrix, and preset the random consistency index table Find the random consistency evaluation index value corresponding to the order; then calculate the target consistency index value corresponding to the judgment matrix according to the maximum characteristic root and the order; then according to the random consistency evaluation index value And the target consistency index value judges whether the judgment matrix is valid; finally, when the judgment matrix is valid, the element value corresponding to each vector element in the feature vector is read, and each performance is determined according to the read element value The subjective weight value corresponding to the indicator parameter.

For example, the master node can use the 9-degree scale method to establish a judgment matrix for the three performance index parameters of network delay, computing power, and disk read speed. The judgement matrix expresses the relative importance of each criterion or scheme, which is the same as the analytic hierarchy process. First, regardless of the impact of the cluster server (scheme) on the evaluation process, evaluate the pairwise relationship between the above three individual indicators in turn, so that a 3×3 judgment matrix can be obtained, and then the maximum characteristic root and feature of the judgment matrix are calculated Vector, and then verify whether the judgment matrix is valid through consistency verification. If it is valid, read the element value corresponding to each vector element in the feature vector, and determine the subjective weight value corresponding to each performance index parameter according to the read element value.

Step S302: The master node obtains the objective weight value corresponding to each performance index parameter through a preset weight assignment algorithm.

It should be understood that the weight assignment algorithm CRITIC collects the performance index parameters in the ZooKeeper server cluster that affect the voting ability, and then performs dimensionless processing and singularity processing on the collected information, and then analyzes the variability and conflict between the data , And then further determine the weight of each performance index parameter. Singularity refers to the data of a certain index of a certain body far exceeding the same index of that type of individual. In this embodiment, the master node will eliminate the detected singular points.

Specifically, the master node can first perform singular point detection on the performance index parameters, remove the singular points in the data corresponding to the performance index parameters according to the detection results, and then non-dimensionalize each performance index parameter to obtain the dimensionless performance index, and obtain The correlation coefficient between every two dimensionless performance indicators; then obtain the standard deviation corresponding to each dimensionless performance indicator, and calculate the information corresponding to each dimensionless performance indicator through the first preset formula according to the standard deviation and the correlation coefficient The amount of information is added up to obtain the total amount of information, and finally the objective weight value corresponding to each performance index parameter is determined according to the amount of information corresponding to each dimensionless performance index and the total amount of information; wherein, the first prediction Let the formula be:

It should be understood that the correlation coefficient is a statistical indicator used to reflect the closeness of the evaluation indicators to each other, and is usually calculated by the product difference method.

Step S303: Based on the subjective weight value and the objective weight value, the master node obtains the index weight corresponding to each performance index parameter through a preset Lagrangian optimal multiplier method.

In specific implementation, after the master node obtains the subjective weight value corresponding to the performance index parameter of the server's voting ability through the AHP method, and the objective weight value corresponding to the performance indicator parameter of the server's voting capability through the CRITIC method, it can pass Lagrang The daily multiplier method is the second preset formula to calculate the index weight corresponding to each performance index parameter;

Wherein, the second preset formula is:

In this embodiment, the Lagrangian optimal multiplier method is used to calculate the index weight according to the subjective weight value and the objective weight value corresponding to the performance index parameters of the server's voting ability, which can minimize the loss of information and make the final calculated weight value Close to the actual value as much as possible to improve the accuracy of the weight value determination.

Referring to FIG. 3, FIG. 3 is a schematic flowchart of a third embodiment of a voting node configuration method of this application.

Based on the foregoing embodiments, in this embodiment, the step S40 may specifically include:

Step S401: The master node separately calculates the voting ability of each slave node according to the index weight and the parameter value corresponding to each performance index parameter through a third preset formula;

Wherein, the third preset formula is:

B=(a ₁ , a ₂ …a _n )×[W ₁ ,W ₂ …W _n ]

In specific implementation, after the master node obtains the parameter value and index weight corresponding to each performance parameter, it can calculate the voting ability corresponding to the slave node according to the above formula.

The present embodiment and the foregoing embodiments will be described below in conjunction with specific examples.

Referring to FIG. 4, FIG. 4 is a schematic diagram of the voting ability evaluation hierarchical relationship model of the third embodiment of the voting node configuration method of this application.

(1) The steps for the master node to obtain the subjective weight value corresponding to each performance index parameter through the preset analytic hierarchy process are as follows:

As shown in FIG. 4, this embodiment first establishes a hierarchical relationship model for the voting capabilities of ZooKeeper server nodes. For the performance index parameters (network latency, processor computing power, and disk read rate) in the criterion layer in Figure 4, this embodiment uses multiple experts with knowledge in related fields to independently construct a judgment matrix, and then uses the geometric average method The final judgment matrix constructed is shown in Table 1, which is the judgment matrix of performance index parameters.

Table 1 Performance index parameter judgment matrix

That is, the final judgment matrix obtained by the master node is as follows:

The master node calculates the maximum characteristic root of A according to the constructed judgment matrix A as: λ _max =3.013, and the characteristic vector W ^A and the target consistency index CI are respectively:

W ^A ＝[W ₁ ,W ₂ ,W ₃ ]=[0.8158,0.1027,0.0815]

Assuming that the master node checks the value of the random consistency index RI in the preset random consistency index table to be 0.58, the CR value of the consistency check result can be calculated:

Since the CR value is less than 0.1, it can be determined that the judgment matrix A meets the consistency, that is, the judgment matrix A is valid.

(2) The master node obtains the objective weight value corresponding to each performance index parameter through the preset weight assignment algorithm, including:

The master node adjusts the master node in turn in the distributed cluster, and collects 100 sets of performance parameter index information for measuring voting ability. As shown in Table 2, Table 2 is the performance index parameter data.

Table 2 Performance index parameter data

For the above-mentioned 100 sets of data, the data obtained after singular point detection and dimensionless processing are shown in Table 3. Table 3 is the performance index parameter data after dimensionless processing.

Table 3 Performance index parameter data after dimensionless treatment

The main node is based on the deviation of the two indicators from the respective averages, and the correlation coefficients of the two indicators are calculated by multiplying the two products, and the correlation coefficient matrix M is obtained as follows:

Then the master node according to the above first preset formula

And the fourth preset formula calculates the weight value of each performance index parameter in the voting ability as follows:

W ^C ＝[W ₁ ,W ₂ ,W ₃ ]=[0.8253,0.0388,0.0359]

Wherein, the fourth preset formula is:

(3) Based on the subjective weight value and the objective weight value, the master node obtains the index weight corresponding to each performance index parameter through the preset Lagrangian optimal multiplier method, including:

Based on the subjective weight value and the objective weight value, the master node calculates the index weight corresponding to each performance index parameter through the second preset formula:

W=[0.7638,0.1858,0.0504]

In this embodiment, the master node obtains the subjective weight value corresponding to the performance index parameter of the server's voting capability through the AHP method, and obtains the objective weight value corresponding to the performance index parameter of the server's voting capability through the CRITIC method, and then through the Lagrangian multiplier method. The second preset formula calculates the index weight corresponding to each performance index parameter. Because the two weight calculation algorithms are combined, it overcomes the weight value calculation defects of each algorithm and improves the accuracy of the performance index parameter weight value calculation .

Referring to Figure 5, Figure 5 is a structural block diagram of the first embodiment of the voting node configuration system of this application.

As shown in Figure 5, the voting node configuration system proposed in the embodiment of the present application includes: a master node 50 and at least two slave nodes (501, 502, 503, etc.), the master node 50 and the at least two slave nodes ( 501, 502, 503) are in a distributed cluster, and the slave node 501 is taken as an example for description below.

The master node 50 is configured to deliver parameter collection tasks to at least two slave nodes in the distributed cluster every preset time period;

The slave node 501 is configured to collect performance index parameters of a preset dimension according to the parameter collection task, and feed back the collected performance index parameters to the master node 50;

The master node 50 is also used to calculate the weight of each performance index parameter based on a preset combination algorithm to obtain the index weight corresponding to each performance index parameter;

The master node 50 is further configured to calculate the voting ability corresponding to each slave node according to the index weight and the parameter value corresponding to each performance index parameter; and

The master node 50 is further configured to determine the number of voting nodes corresponding to the distributed cluster according to a preset allowable number of downtimes, and to obtain the number of voting nodes from the at least two slave nodes according to the number of voting nodes and the voting capacity. Select the corresponding voting node.

In this embodiment, the master node delivers parameter collection tasks to at least two slave nodes in the distributed cluster where it is located every preset time period; each of the at least two slave nodes collects the performance of the preset dimensions according to the parameter collection task Index parameters, and feed back the collected performance index parameters to the main node; the main node calculates the weight of each performance index parameter based on the preset combination algorithm to obtain the index weight corresponding to each performance index parameter; according to the index weight and each performance index parameter Corresponding parameter values are calculated for the corresponding voting ability of each slave node; then the number of voting nodes corresponding to the distributed cluster is determined according to the preset allowable downtime number, and the corresponding voting node is selected according to the number of voting nodes and voting ability. Since the voting capacity of each slave node is calculated according to the performance index parameters of each slave node and the performance index parameter index weight determined by the combination algorithm, and then voting nodes are screened according to the voting ability, it can ensure that the selected slave nodes have Better working performance, also realized the automatic configuration of voting nodes.

Based on the first embodiment of the voting node configuration system of the present application, a second embodiment of the voting node configuration system of the present application is proposed.

In this embodiment, the slave node 501 is also used to locally create a target file according to the parameter collection task, perform read and write operations on the target file within a preset time period, and perform read and write operations on the target file according to the statistical total read and write operations. The corresponding disk reading rate is calculated by the number of times; the processor computing capability parameters included in the parameter collection task are read, and the processor computing capability parameters include: a calculation time limit and a value to be calculated; The value to be calculated performs a number of prime number calculation operations, and the corresponding processor computing power is obtained according to the execution result; the disk read rate and the processor computing power are fed back to the master node 50 as performance index parameters.

In one embodiment, the master node 50 is also used to obtain the subjective weight value corresponding to each performance index parameter through the preset analytic hierarchy process; obtain the objective weight value corresponding to each performance index parameter through the preset weight assignment algorithm; For the subjective weight value and the objective weight value, the index weight corresponding to each performance index parameter is obtained through a preset Lagrangian optimal multiplier method.

In an embodiment, the master node 50 is also used to construct a corresponding judgment matrix according to various performance index parameters, and calculate the maximum characteristic root and characteristic vector of the judgment matrix; obtain the order corresponding to the judgment matrix, Search for the random consistency evaluation index value corresponding to the order in the preset random consistency index table; calculate the target consistency index value corresponding to the judgment matrix according to the maximum characteristic root and the order; The random consistency evaluation index value and the target consistency index value determine whether the judgment matrix is valid; when the judgment matrix is valid, the element value corresponding to each vector element in the eigenvector is read, and according to the read The element value determines the subjective weight value corresponding to each performance index parameter.

In an embodiment, the master node 50 is also used to dimensionlessly transform each performance index parameter to obtain a dimensionless performance index, and obtain the correlation coefficient between every two dimensionless performance indicators; According to the standard deviation corresponding to the performance index, the information amount corresponding to each dimensionless performance index is calculated by the first preset formula according to the standard deviation and the correlation coefficient, and the information amount is added to obtain the total amount of information; The amount of information corresponding to the performance index and the total amount of information determine the objective weight value corresponding to each performance index parameter; wherein, the first preset formula is:

In an embodiment, the master node 50 is further configured to calculate the index weight corresponding to each performance index parameter based on the subjective weight value and the objective weight value through a second preset formula; wherein, the second The preset formula is:

In one embodiment, the master node 50 is further configured to calculate the voting power corresponding to each slave node through a third preset formula according to the index weight and the parameter value corresponding to each performance index parameter; where The third preset formula is:

B=(a ₁ , a ₂ …a _n )×[W ₁ ,W ₂ …W _n ]

In an embodiment, the master node 50 is further configured to determine the number of voting nodes corresponding to the distributed cluster according to a preset allowable number of downtimes; and sort the voting capabilities in descending order , And select the slave nodes of the number of voting nodes as voting nodes according to the sorting result.

For other embodiments or specific implementations of the voting node configuration system of the present application, reference may be made to the foregoing method embodiments, which will not be repeated here.

It should be noted that in this article, the terms "include", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements not only includes those elements, It also includes other elements not explicitly listed, or elements inherent to the process, method, article, or system. If there are no more restrictions, the element defined by the sentence "including a..." does not exclude the existence of other identical elements in the process, method, article or system that includes the element.

The serial numbers of the foregoing embodiments of the present application are for description only, and do not represent the superiority of the embodiments.

Through the description of the above embodiments, those skilled in the art can clearly understand that the method of the above embodiments can be implemented by means of software plus the necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is better.的实施方式。 Based on this understanding, the technical solution of this application essentially or the part that contributes to the existing technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium (such as read-only memory/random access The memory, magnetic disk, and optical disk) includes several instructions to make a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) execute the methods described in the various embodiments of the present application.

The above are only preferred embodiments of this application, and do not limit the scope of this application. Any equivalent structure or equivalent process transformation made using the content of the description and drawings of this application, or directly or indirectly used in other related technical fields , The same reason is included in the scope of patent protection of this application.

Claims

A voting node configuration method, wherein the method includes:

The master node issues parameter collection tasks to at least 2 slave nodes in the distributed cluster where it is located every preset time period;

Each of the at least two slave nodes collects performance index parameters of a preset dimension according to the parameter collection task, and feeds back the collected performance index parameters to the master node;

The master node calculates the weight of each performance index parameter based on a preset combination algorithm to obtain the index weight corresponding to each performance index parameter;

The master node respectively calculates the voting ability corresponding to each slave node according to the index weight and the parameter value corresponding to each performance index parameter; and

The master node determines the number of voting nodes corresponding to the distributed cluster according to a preset allowable number of downtimes, and selects corresponding votes from the at least two slave nodes according to the number of voting nodes and the voting capacity node.
The method according to claim 1, wherein each of the at least two slave nodes collects performance index parameters of a preset dimension according to the parameter collection task, and feeds back the collected performance index parameters to the The steps of the master node include the following steps:

Create a target file locally according to the parameter collection task, perform read and write operations on the target file within a preset time period, and calculate the corresponding disk read rate according to the total number of reads and writes counted;

Reading the processor computing capability parameters included in the parameter collection task, where the processor computing capability parameters include: a calculation time limit and a value to be calculated;

Perform a number of prime number calculation operations on the value to be calculated within the calculation time limit, and obtain the corresponding processor computing power according to the execution result; and

The disk read rate and the processor computing power are fed back to the master node as performance index parameters.
The method of claim 1, wherein the step of calculating the weight of each performance index parameter by the master node based on a preset combination algorithm to obtain the index weight corresponding to each performance index parameter comprises the following steps:

Obtain the subjective weight value corresponding to each performance index parameter through the preset analytic hierarchy process;

Obtain the objective weight value corresponding to each performance index parameter through the preset weight assignment algorithm; and

Based on the subjective weight value and the objective weight value, the index weight corresponding to each performance index parameter is obtained through a preset Lagrangian optimal multiplier method.
The method according to claim 3, wherein the step of obtaining the subjective weight value corresponding to each performance index parameter by the master node through a preset analytic hierarchy process comprises the following steps:

Construct a corresponding judgment matrix according to each performance index parameter, and calculate the maximum characteristic root and characteristic vector of the judgment matrix;

Acquiring the order corresponding to the judgment matrix, and searching for the random consistency evaluation index value corresponding to the order in a preset random consistency index table;

Calculating the target consistency index value corresponding to the judgment matrix according to the maximum characteristic root and the order;

Judging whether the judgment matrix is valid according to the random consistency evaluation index value and the target consistency index value; and

When the judgment matrix is valid, the element value corresponding to each vector element in the eigenvector is read, and the subjective weight value corresponding to each performance index parameter is determined according to the read element value.
The method according to claim 3, wherein the step of obtaining the objective weight value corresponding to each performance index parameter by the master node through a preset weight assignment algorithm comprises the following steps:

Dimensionlessly transform each performance index parameter to obtain a dimensionless performance index, and obtain the correlation coefficient between every two dimensionless performance indexes;

Obtain the standard deviation corresponding to each non-dimensional performance index, calculate the information amount corresponding to each non-dimensional performance index through the first preset formula according to the standard deviation and the correlation coefficient, and add the information amount to obtain the total information ;as well as

The objective weight value corresponding to each performance index parameter is determined according to the amount of information corresponding to each dimensionless performance index and the total amount of information; wherein, the first preset formula is:

In the formula, G i is the amount of information corresponding to the dimensionless performance index i, σ i is the standard deviation corresponding to the dimensionless performance index i, and r ij is the correlation coefficient between the dimensionless performance index i and j.
The method of claim 3, wherein the master node obtains the index weight corresponding to each performance index parameter through a preset Lagrangian optimal multiplier method based on the subjective weight value and the objective weight value The steps include the following steps:

Based on the subjective weight value and the objective weight value, the index weight corresponding to each performance index parameter is calculated by a second preset formula; wherein, the second preset formula is:

In the formula, W i is the index weight corresponding to the i-th performance index parameter, n is the number of performance index parameters, W i A is the subjective weight value, and W i C is the objective weight value.
The method according to claim 1, wherein the step of the master node separately calculating the voting ability corresponding to each slave node according to the index weight and the parameter value corresponding to each performance index parameter comprises the following steps:

According to the index weight and the parameter value corresponding to each performance index parameter, the voting ability corresponding to each slave node is calculated through a third preset formula; wherein, the third preset formula is:

B=(a 1 , a 2 …a n )×[W 1 ,W 2 …W n ]

Wherein, B is the ability to vote, a n is the n performance metrics parameters, W n is the n performance indicators corresponding to the heavy weight parameter index.
The method of claim 1, wherein the master node determines the number of voting nodes corresponding to the distributed cluster according to a preset allowable number of downtimes, and according to the number of voting nodes and the voting capacity The at least two steps of selecting corresponding voting nodes from nodes include the following steps:

Determine the number of voting nodes corresponding to the distributed cluster according to the preset allowable number of downtimes; and

Sort the voting capabilities in descending order, and select slave nodes of the number of voting nodes as voting nodes according to the sorting result.
A voting node configuration system, wherein the system includes: a master node and at least two slave nodes, and the master node and the at least two slave nodes are in a distributed cluster;

The master node is configured to deliver parameter collection tasks to the at least two slave nodes in the distributed cluster every preset time period;

Each of the at least two slave nodes is configured to collect performance index parameters of a preset dimension according to the parameter collection task, and feed back the collected performance index parameters to the master node;

The master node is also used to calculate the weight of each performance index parameter based on a preset combination algorithm to obtain the index weight corresponding to each performance index parameter;

The master node is also used to calculate the voting power corresponding to each slave node according to the index weight and the parameter value corresponding to each performance index parameter; and

The master node is also used to determine the number of voting nodes corresponding to the distributed cluster according to a preset allowable number of downtimes, and to select from the at least two slave nodes according to the number of voting nodes and the voting capacity Select the corresponding voting node.
The system according to claim 9, wherein the master node is further configured to calculate the weight of each performance index parameter based on a preset combination algorithm to obtain the index weight corresponding to each performance index parameter, comprising:

Obtain the subjective weight value corresponding to each performance index parameter through the preset analytic hierarchy process;

Obtain the objective weight value corresponding to each performance index parameter through the preset weight assignment algorithm; and

Based on the subjective weight value and the objective weight value, the index weight corresponding to each performance index parameter is obtained through a preset Lagrangian optimal multiplier method.