A multi-objective optimization method and system for virtual machine service quality

Abstract

The present invention discloses a multi-objective optimization method and system for virtual machine service quality, which belongs to the field of cloud computing. It includes: inputting the user's multi-objective indicator expectation value of virtual machine service quality into a learned multi-objective optimization model to obtain the characteristic configuration value of the virtual machine operation element; the multi-objective optimization model is obtained by the following methods: S1. extracting the virtual machine operation element feature vector and the corresponding multi-objective indicator vector of virtual machine service quality, and constructing a data set sample, wherein the sample is the multi-objective indicator vector of virtual machine service quality-virtual machine operation element feature vector; S2. using the data set sample to learn and evaluate the multi-objective optimization model; the multi-objective optimization model uses the virtual machine operation element feature vector as the decision variable, the virtual machine service quality as the target, and the load balancing strategy, disaster recovery strategy, and fault tolerance strategy of the cloud computing platform as the constraint. Through multi-objective optimization for virtual machine service quality, high-quality computing services of the cloud computing platform are guaranteed.

Description

Technical Field

[0001] The present invention belongs to the field of cloud computing, and more specifically, relates to a multi-objective optimization method and system for virtual machine service quality. Background Art

[0002] Patent CN102662763A proposes a virtual machine resource scheduling method based on service quality under a multi-core architecture. This method satisfies the characteristics of load balancing in a multi-core environment through reasonable task allocation, reduces the response time of I / O requests, and meets the needs of service quality control. Patent CN112882917A proposes a virtual machine service quality dynamic prediction method based on Bayesian network migration. When the underlying environment characteristics of the virtual machine change, the static Bayesian network model cannot accurately predict its service quality, thereby realizing the dynamic prediction of the virtual machine service quality.

[0003] In summary, the existing virtual machine service quality prediction and optimization are mostly tuned for a certain performance feature, which cannot meet the different service quality requirements of various users for virtual machines, nor can it achieve quantitative automatic configuration of virtual machine operation elements. Summary of the invention

[0004] In view of the defects of the prior art, the purpose of the present invention is to provide a multi-objective optimization method and system for virtual machine service quality, aiming to solve the problem that different service quality requirements of various users for virtual machines cannot be met, and quantitative automatic configuration of virtual machine operation elements cannot be achieved.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a multi-objective optimization method for virtual machine service quality, comprising:

[0006] Input the user's expected values ​​of the multi-objective indicators of the virtual machine service quality into the learned multi-objective optimization model to obtain the characteristic configuration values ​​of the virtual machine operation elements, thereby realizing the multi-objective tuning of the virtual machine service quality;

[0007] The multi-objective optimization model is obtained in the following way:

[0008] S1. Extract the virtual machine operation factor feature vector and the corresponding virtual machine service quality multi-objective indicator vector, the multi-objective indicator vector of the virtual machine service quality includes: virtual machine comprehensive computing power, virtual machine response real-time performance, virtual machine availability and virtual machine security, and construct a data set sample, the sample is the multi-objective indicator vector of virtual machine service quality - virtual machine operation factor feature vector;

[0009] The calculation formula of the comprehensive computing power of the virtual machine is as follows:

[0010]

[0011] in, is the comprehensive computing power index value of the virtual machine, is the intermediate result, is the weight, They are the average CPU performance score, memory performance score, network performance score, and disk performance score of all virtual machines within a service cycle. It is the corresponding performance score of all virtual machines with the same configuration physical machine within a service period;

[0012] S2. Use dataset samples to learn and evaluate multi-objective optimization models;

[0013] The multi-objective optimization model uses the virtual machine operation factor feature vector as the decision variable, the virtual machine service quality as the goal, and the cloud computing platform's load balancing strategy, disaster recovery strategy, and fault tolerance strategy as constraints;

[0014] The method further includes: performing the following preprocessing on the extracted data:

[0015] First, MinMax standardization is used to standardize the characteristic values ​​of virtual machine operation factors to between [-1, 1];

[0016] Secondly, the normalized virtual machine operation factor feature values ​​are processed by backward difference;

[0017] Thirdly, according to the computing power characteristics of various heterogeneous resources in the cloud computing environment, the heterogeneous computing resources participating in the multi-objective optimization model training are grouped, and the data is first distributed in groups. The data distributed between groups is proportional to the computing power of the group, and the equal distribution strategy is used within the group for secondary distribution.

[0018] After the first allocation, group The amount of data sample allocated The calculation formula is as follows:

[0019]

[0020] After the second allocation, the group The amount of data samples allocated to each node The calculation formula is as follows:

[0021]

[0022] in, is the total amount of data samples, , For the The number of nodes in a group, For grouping The slowest node speed in the calculation, For grouping The number of nodes in For the Groups The overall training speed.

[0023] Preferably, the virtual machine operation element feature vector includes: physical resource features, virtual machine hardware configuration features, virtual machine software configuration features, virtual machine intelligent configuration features, virtual machine operation environment features, application features, security policies, reliability policies and other optimization policies.

[0024] Preferably, the multi-objective optimization model adopts a three-layer BP neural network, takes the multi-objective index vector of the virtual machine service quality as an input variable, and takes the virtual machine operation element feature vector as an output variable.

[0025] Preferably, the multi-objective optimization model is described in the following manner:

[0026]

[0027] in, The feature vector for the virtual machine running, For a virtual machine The assignable matrix of service quality indicators is is the total number of virtual machines, is the length of the service quality indicator vector, For the The first virtual machine The priority of the service quality indicators, For the The first virtual machine service quality indicators, is the set of constraints, is the virtual machine target vector, A set of feature vectors for virtual machine operation elements. A set of multi-objective indicator vectors for virtual machine service quality.

[0028] Preferably, the priority calculation formula is as follows:

[0029]

[0030]

[0031] Among them, the weight value The larger it is, the more attention is paid to this indicator. For the virtual machines, For the The first virtual machine The normalized value of a service quality indicator.

[0032] To achieve the above-mentioned object, in a second aspect, the present invention provides a multi-objective optimization system for virtual machine service quality, including a processor and a memory;

[0033] The processor is used to store computer-executable instructions;

[0034] The processor is configured to execute the computer-executable instructions so that the method described in the first aspect is performed.

[0035] In general, the above technical solution conceived by the present invention has the following beneficial effects compared with the prior art:

[0036] The present invention proposes a multi-objective optimization method and system for virtual machine service quality, which realizes the construction of a multi-objective optimization model for virtual machine service quality by analyzing and representing the high-dimensional complex operating elements of virtual machines and the indicators representing the multi-objective service quality of virtual machines through a three-layer BP neural network to approximate the complex nonlinear relationship between the characteristics of virtual machine operating elements and the multi-objective service quality indicators of virtual machines. Based on this, the automatic configuration of virtual machine operating elements driven by the multi-objective service quality of virtual machines is realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A vector composition diagram representing the virtual machine operation elements provided by the present invention.

[0038] Figure 2 This is the comprehensive computing power hierarchy model of the virtual machine provided by the present invention.

[0039] Figure 3 A flow chart of a multi-objective optimization method for virtual machine service quality provided by the present invention.

[0040] Figure 4 This is a two-level data partitioning strategy framework diagram provided by the present invention.

[0041] Figure 5 This is a schematic diagram of the three-layer BP neural network model provided by the present invention. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0043] 1. Construct a virtual machine operation factor vector based on high-dimensional complex features.

[0044] The virtual machine operation factor characteristics are analyzed from the perspective of the input factors of the virtual machine service quality, and the virtual machine operation factor vector is defined. Figure 1 The virtual machine operation element provided by the present invention represents a vector composition diagram. Figure 1 As shown in the figure, the virtual machine operation factors include: simple operation characteristics and complex operation characteristics. Simple operation characteristics can be directly monitored and obtained, while complex operation characteristics are more difficult to obtain directly. However, as time goes by, they also affect the quality of virtual machine services. For example, the number of physical machine CPU cores allocated to the virtual machine is a simple operation characteristic, while the security policy scheme of the virtual machine, such as the data integrity verification method, needs to be confirmed by the attack means test score.

[0045] like Figure 1 As shown, the simple operation characteristics include: physical resource characteristics (CPU type, CPU main frequency, number of CPU cores, memory capacity, hard disk capacity and network bandwidth); virtual machine hardware configuration characteristics (number of vCPUs, virtual machine memory capacity and virtual machine image capacity); virtual machine software configuration characteristics (operating system configuration, basic software configuration, tool software configuration); virtual machine intelligent configuration characteristics (intelligent framework configuration, intelligent algorithm configuration); virtual machine operating environment characteristics (the number of virtual machines running on the same physical machine); application characteristics (the type of load running on the virtual machine).

[0046] like Figure 1 As shown, the complex operation characteristics are: security strategy (access control strategy, data integrity verification, vulnerability scanning solution, data encryption algorithm); reliability strategy (virtual machine migration algorithm, virtual machine fault recovery strategy, virtual machine hot standby solution, virtual machine playback solution); other optimization strategies (real-time optimization strategy, robustness optimization strategy).

[0047] 2. Construct a multi-objective indicator vector for virtual machine service quality.

[0048] From the output level of virtual machine service quality, the performance indicators of the virtual machine are analyzed, and a multi-objective indicator vector for virtual machine service quality is defined, which mainly includes: virtual machine comprehensive computing power, virtual machine response real-time performance, virtual machine availability, and virtual machine security.

[0049] (1) The virtual machine's comprehensive computing power is, to a certain extent, the computing power level of the virtual machine, which is mainly calculated by the virtual machine's performance score and performance impairment score. The specific calculation process includes the following sub-steps:

[0050] (1.1) Data normalization.

[0051] The value range and unit in the virtual machine comprehensive computing power evaluation system are different. The normalization process uses the following formula to normalize the evaluation data:

[0052]

[0053] in, is the normalized value of the score of each attribute evaluation. , , They are the average, minimum and maximum values ​​obtained from multiple measurements.

[0054] (1.2) Construct a service capability hierarchy model.

[0055] Based on the hierarchical analysis method, a hierarchical model of virtual machine comprehensive computing power indicators is defined. Figure 2 The virtual machine comprehensive computing power level model provided by the present invention. Figure 2 As shown in the figure, the comprehensive computing power of the virtual machine measures the computing power of the virtual machine from two aspects: virtual machine performance and virtual machine performance degradation. The virtual machine performance is directly reflected in the ability to provide users with computing, storage, network and other resource services, while the virtual machine performance degradation evaluates the performance degradation of the virtual machine compared with the performance of the same configuration physical machine from the perspective of computing performance, network performance and disk performance. The specific calculation is shown in Table 1.

[0056] Table 1 Virtual machine comprehensive computing power evaluation system

[0057]

[0058] (1.3) Establish the comparison matrix D.

[0059] Construct a second-order comparison matrix D, and assign values ​​to the comparison matrix according to the importance of each evaluation item.

[0060] (1.4) Solve the weights.

[0061] First, the largest eigenvector of the matrix D is denoted as , and the corresponding eigenvector , and the inconsistency value CI of the comparison matrix is ​​obtained according to the following formula, and the matrix is ​​judged whether it has satisfactory consistency by checking the data. If the matrix obtains satisfactory consistency, then ; Otherwise, readjust the comparison matrix D until satisfactory consistency is obtained.

[0062]

[0063] (1.5) Solve the comprehensive computing power index of the virtual machine.

[0064] Finally, the scores obtained from the normalized attribute evaluation and the weights obtained are , and obtain the comprehensive computing power index of the virtual machine, as shown in the following formula:

[0065]

[0066] in, is the comprehensive computing power index value of the virtual machine, is the intermediate result, is the weight, They are the average CPU performance score, memory performance score, network performance score, and disk performance score of all virtual machines within a service cycle. It is the corresponding performance score of all virtual machines with the same configuration on the physical machine within a service period.

[0067] A two-level analytic hierarchy process was used to determine the weights of the evaluation items that affect the comprehensive computing power. The comprehensive computing power of the virtual machine was redefined by comprehensively considering the virtual machine hardware configuration and the virtual machine performance degradation.

[0068] (2) Real-time response of virtual machines: It is defined as the time difference from when a user initiates a service request to when the service receives a response. The definition of virtual machine service efficiency is as follows:

[0069]

[0070] in, is the number of applications tested on the virtual machine, Run the following command on the virtual machine: The runtime of a test application. That is, the real-time response of the virtual machine is essentially the efficiency of the virtual machine service with the runtime.

[0071] (3) Virtual machine availability: This is determined by the various fault-tolerant configuration schemes provided by the cloud computing system. Currently, most people use mean time between failures (MTBF) or mean time to failure (MTTF) to measure availability, and mean time to repair (MTTR) to measure system maintainability. Availability It can be defined as follows:

[0072] .

[0073] (4) Virtual machine security: Use NSFOCUS tools to obtain a security test report, including a comprehensive security performance score (0-100 points) and a comprehensive system security grade and sub-grade (A, B, C, D).

[0074] 3. Construct a multi-objective indicator tuning and control model for virtual machine service quality.

[0075] Figure 3A flowchart of a multi-objective optimization method for virtual machine service quality provided by the present invention. Figure 3 As shown in the figure, firstly, the data set sample is constructed by using the "virtual machine service quality multi-objective indicator vector-virtual machine operation element feature vector", and the training set and test set are divided according to a certain ratio; finally, the model is evaluated through the test set to determine the multi-objective tuning control model for virtual machine service quality. Through the intelligent control model, the multi-objective tuning of virtual machine service quality is realized to ensure the high availability scheduling capability of cloud computing environment resources.

[0076] Data preprocessing is a key step in the BP neural network model (also called reverse neural network). It can not only improve the accuracy of the model, but also improve the calculation efficiency of the model through reasonable data partitioning strategy. This step mainly includes data standardization, two-level data allocation strategy, etc.

[0077] 1) Dataset standardization: Considering the numerical differences in the virtual machine running status data, MinMax standardization is adopted. The following formula is used to convert the original virtual machine running feature value Normalized to [-1, +1], represented as :

[0078]

[0079] The specific implementation steps are:

[0080] Step 1: Get the training set data, mark the maximum value of the data set as the standard maximum value, and the minimum value as the standard minimum value;

[0081] Step 2: For each running feature monitoring quantity of the virtual machine in the training set, standardize it based on the standardization formula;

[0082] Step 3: For each running feature monitoring quantity of the virtual machine in the test set, standardize it based on the standardization formula.

[0083] 2) Backward difference processing: Backward difference refers to the difference processing of adjacent data moments. Backward difference of the original monitoring data can effectively expand the size of the data set and provide richer input data for the model.

[0084] 3) Two-level data partitioning strategy: The two-level data partitioning strategy aims to improve the computing efficiency of model construction. It maximizes the performance characteristics of computing nodes, makes full use of heterogeneous computing resources in the cloud computing environment, and avoids uneven use of computing resources.

[0085] Figure 4 The two-level data partitioning strategy framework diagram provided by the present invention. Figure 4As shown in the figure, the two-level data partitioning strategy first groups resources according to their computing power characteristics, and performs the first data allocation in groups. The data allocated between groups is proportional to the computing power of the group; the second allocation is performed within the group. Since the computing power of the resources within the group is similar, the nodes within the group adopt an equal allocation strategy. After that, data synchronization and update are performed according to the two-level data partitioning strategy during the neural network calculation process to improve the model calculation efficiency.

[0086] First-level allocation: Calculate the data sample size for each group, assuming grouping The slowest node speed is , group The number of nodes in , then Groups The overall training speed is . For the The number of nodes in each group. Assume that the total amount of data samples is , then group The amount of data sample allocated Obtained by the following formula.

[0087]

[0088] Second level allocation: The data sample size of each node in the calculation group is similar, so the nodes in the group are divided equally to allocate data to the calculation nodes. The amount of data samples allocated to each node in the group is obtained by the following formula. The slowest node speed is , group The number of nodes in , then Groups The overall training speed is . For the The number of nodes in a group.

[0089]

[0090] Assume that the virtual machine service quality The target indicators are , for virtual machines The definition of the target indicator optimization is as follows:

[0091]

[0092] in, The feature vector for the virtual machine running, For a virtual machine The assignable matrix of service quality indicators is is the total number of virtual machines, is the length of the service quality indicator vector, For the The first virtual machine The priority of the service quality indicators, For the The first virtual machine service quality indicators, is the set of constraints, is the virtual machine target vector, A set of feature vectors for virtual machine operation elements. is a set of multi-objective indicator vectors for the quality of service of virtual machines. There are often conflicts between the indicators of virtual machines. Therefore, multi-objective problems can only make the objectives as optimal as possible under the given resource conditions. It is impossible to make each objective reach the best target state, that is, it is impossible to achieve the optimal value. Moreover, different virtual machines have different focuses on the requirements of target service quality. Some virtual machines focus on response time, while some virtual machines focus more on security.

[0093] Virtual Machines The multi-objective priority evaluation index of service quality is shown as follows: Serving metrics for virtual machines The normalization function of Quality of service indicators for virtual machines The weight of .

[0094]

[0095] The normalized function of is as follows:

[0096]

[0097] Weight value The larger the value, the more attention is paid to the indicator. In practical applications, the parameter value can be adjusted according to the priority needs. When setting the specific value, the analytic hierarchy process can be used to determine the value of the weight. The normalization formula converts the multi-objective optimization problem into a single-objective problem based on priority authority, so the problem is easy to solve.

[0098] First, set The comparison matrix D is the value of the comparison matrix. The value of the comparison matrix is ​​determined by the importance of each sub-indicator of service quality. Secondly, the eigenvalue of the matrix D is used to obtain the eigenvector , and the inconsistency value CI of the comparison matrix is ​​obtained according to the following formula, and the inconsistency index of the matrix D is obtained by checking the data and obtaining the inconsistency ratio. If the matrix obtains satisfactory consistency, then ; Otherwise, readjust the comparison matrix D until satisfactory consistency is obtained.

[0099]

[0100] Since there is a complex nonlinear relationship between multi-objective indicators and high-dimensional complex operation characteristics, and the formula belongs to High-order formulas are difficult to solve. The three-layer reverse neural network model is particularly suitable for dealing with complex, nonlinear problems through continuous learning and training. Because the BP neural network is better than statistical methods and traditional expert systems in dealing with complex nonlinear problems by analyzing and training to explore complex internal relationships. The BP neural network can express relationships that are difficult to describe with mathematical formulas through the BP neural network. Moreover, compared with other artificial neural network models, the BP neural network algorithm has a complete theoretical basis, a clear learning mechanism, strong fault tolerance, and self-learning and self-adaptation capabilities. Therefore, for complex nonlinear problems that are difficult to express with mathematical formulas, the BP neural network has obvious advantages. In addition, the field of neural networks has proved that the BP neural network can approximate any nonlinear function with arbitrary accuracy.

[0101] Figure 5 The schematic diagram of the three-layer BP neural network model provided by the present invention is shown in FIG. Figure 5 As shown in the figure, the multi-objective index vector of virtual machine service quality is used as the input variable, and the feature vector of virtual machine operation elements is used as the output variable. The constructed multi-objective tuning control model for virtual machine service quality is used to optimize the virtual machine operation elements and meet the diversified needs of users for service quality.

[0102] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A multi-objective optimization method for virtual machine service quality, characterized in that: include: Input the user's expected values ​​of the multi-objective indicators of the virtual machine service quality into the learned multi-objective optimization model to obtain the characteristic configuration values ​​of the virtual machine operation elements, thereby realizing the multi-objective tuning of the virtual machine service quality; The multi-objective optimization model is obtained in the following way: S1. Extract the virtual machine operation factor feature vector and the corresponding virtual machine service quality multi-objective indicator vector, the multi-objective indicator vector of the virtual machine service quality includes: virtual machine comprehensive computing power, virtual machine response real-time performance, virtual machine availability and virtual machine security, and construct a data set sample, the sample is the multi-objective indicator vector of virtual machine service quality - virtual machine operation factor feature vector; The calculation formula of the comprehensive computing power of the virtual machine is as follows: in, is the comprehensive computing power index value of the virtual machine, is the intermediate result, is the weight, They are the average CPU performance score, memory performance score, network performance score, and disk performance score of all virtual machines within a service cycle. It is the corresponding performance score of all virtual machines with the same configuration physical machine within a service period; S2. Use dataset samples to learn and evaluate multi-objective optimization models; The multi-objective optimization model uses the virtual machine operation factor feature vector as the decision variable, the virtual machine service quality as the goal, and the cloud computing platform's load balancing strategy, disaster recovery strategy, and fault tolerance strategy as constraints; The method further includes: performing the following preprocessing on the extracted data: First, MinMax standardization is used to standardize the characteristic values ​​of virtual machine operation factors to between [-1, 1]; Secondly, the normalized virtual machine operation factor feature values ​​are processed by backward difference; Thirdly, according to the computing power characteristics of various heterogeneous resources in the cloud computing environment, the heterogeneous computing resources participating in the multi-objective optimization model training are grouped, and the data is first distributed in groups. The data distributed between groups is proportional to the computing power of the group, and the equal distribution strategy is used within the group for secondary distribution. After the first allocation, group The amount of data sample allocated The calculation formula is as follows: After the second allocation, the group The amount of data samples allocated to each node The calculation formula is as follows: in, is the total amount of data samples, , For the The number of nodes in a group, For grouping The slowest node speed in the calculation, For grouping The number of nodes in For the Groups The overall training speed.

2. The method according to claim 1, characterized in that The virtual machine operation element feature vector includes: physical resource features, virtual machine hardware configuration features, virtual machine software configuration features, virtual machine intelligent configuration features, virtual machine operation environment features, application features, security policies, reliability policies and other optimization policies.

3. The method according to claim 1, characterized in that The multi-objective optimization model adopts a three-layer BP neural network, takes the multi-objective index vector of the virtual machine service quality as an input variable, and takes the virtual machine operation element feature vector as an output variable.

4. The method according to claim 3, characterized in that The multi-objective optimization model is described in the following way: in, The feature vector for the virtual machine running, For a virtual machine The assignable matrix of service quality indicators is is the total number of virtual machines, is the length of the service quality indicator vector, For the The first virtual machine The priority of the service quality indicators, For the The first virtual machine service quality indicators, is the set of constraints, is the virtual machine target vector, A set of feature vectors for virtual machine operation elements. A set of multi-objective indicator vectors for virtual machine service quality.

5. The method according to claim 4, characterized in that The priority calculation formula is as follows: Among them, the weight value The larger it is, the more attention is paid to this indicator. For the virtual machines, For the The first virtual machine The normalized value of a service quality indicator.

6. A multi-objective optimization system for virtual machine service quality, characterized in that: including a processor and a memory; The processor is used to store computer-executable instructions; The processor is configured to execute the computer-executable instructions so that the method according to any one of claims 1 to 5 is performed.

Images