A mobile cloud computing power leasing and scheduling method for virtual computers

Abstract

The present invention discloses a mobile phone cloud computing power leasing and scheduling method for virtual computers, which relates to the field of resource scheduling technology. The method includes: analyzing the performance change trend of the historical load data of the hardware of all cloud computing power circuits based on the historical load data of the hardware, and evaluating the dynamic load index of each cloud computing power circuit; obtaining historical cloud computing power user leasing data, analyzing the computing power expenditure requirements of the user leasing data, and constructing a user computing power leasing preference profile; performing regression analysis to predict the demand for pending leasing tasks per unit time node; and constructing a cloud computing power resource scheduling model based on the dynamic load indicators of each cloud computing power circuit and the demand for pending leasing tasks per unit time node, so as to maximize the utilization of the computing power resources of the cloud computing power circuit and generate dynamic cloud computing power leasing scheduling tasks. The present invention has the advantages of improving user experience and resource utilization.

Description

Technical Field

[0001] The present invention relates to the technical field of resource scheduling, and in particular to a method for leasing and scheduling mobile phone cloud computing power for virtual computers. Background Art

[0002] Mobile cloud computing power leasing and scheduling for virtual computers refers to the process of leasing mobile cloud computing power as a resource to users in need through virtualization technology in a cloud computing environment, and dynamically allocating and managing these computing resources based on the user's actual needs, the availability of cloud computing resources, and scheduling strategies.

[0003] The existing virtual computer mobile cloud computing power rental scheduling is relatively complete in allocating tasks to users when the line conditions are good. However, since user demands are usually concentrated, the actual resource scheduling cannot meet the user's task needs. When the cloud computing power line is congested, it is very easy for users to go offline to perform tasks. Summary of the Invention

[0004] In order to solve the above technical problems, a mobile phone cloud computing power leasing and scheduling method for virtual computers is provided. This technical solution solves the problem that when the line status of the above-mentioned existing virtual computer mobile phone cloud computing power leasing and scheduling is good, the scheduling and allocation of user task requirements are relatively complete, but because user needs are usually more concentrated, the actual resource scheduling cannot meet the user task requirements. When the cloud computing power line is congested, it is very easy to cause the user to be offline when performing tasks.

[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0006] A method for leasing and scheduling mobile cloud computing power for virtual computers, comprising:

[0007] Based on the historical load data of the hardware of all cloud computing power lines, analyze the performance change trend of the historical load data of the hardware and evaluate the dynamic load indicators of each cloud computing power line;

[0008] Obtain historical cloud computing power user rental data, analyze the computing power expenditure requirements of user rental data, and build a profile of user computing power rental preferences;

[0009] Conduct regression analysis to predict the demand for rental tasks to be executed at a unit time node;

[0010] Based on the dynamic load indicators of each cloud computing power line and the demand for pending leasing tasks per unit time node, a cloud computing power resource scheduling model is constructed to maximize the utilization of the computing power resources of the cloud computing power line and generate dynamic cloud computing power leasing scheduling tasks.

[0011] Preferably, based on the historical load data of the hardware of all cloud computing power lines, the performance change trend of the historical load data of the hardware is analyzed, and the dynamic load indicators of each cloud computing power line are evaluated, specifically including:

[0012] Based on the historical load data of the hardware of all cloud computing power lines, the data is processed smoothly according to the difference method and divided into unit time to form the historical load time series data of the hardware of all cloud computing power lines;

[0013] Based on the historical load time series data of the hardware of all cloud computing power lines, with unit time as the observation window and the historical load data of the hardware as the observation variable, the basic characteristic data of the historical load of the hardware of each cloud computing power line is marked; the basic characteristic data includes: average load, maximum load value, and minimum load value;

[0014] Using wavelet transform, we extract the load cycle characteristic data from the historical load time series data of the hardware of each cloud computing power line;

[0015] Combine the historical load basic characteristic data of the hardware of each cloud computing power line with the historical load cycle characteristic data of the hardware of each cloud computing power line to form the original data set of cloud computing power line load characteristics;

[0016] Based on ARIMA autoregressive integrated moving average, a dynamic load forecasting model for cloud computing power lines is constructed;

[0017] Substitute the original data set of cloud computing power line load characteristics into the computing power line dynamic load prediction model, use the load characteristic data of each cloud computing power line at a unit time node as input, and use the load label data of each cloud computing power line at a future unit time node as output to obtain the dynamic load index of each cloud computing power line;

[0018] The cloud computing power line dynamic load prediction model is specifically as follows:

[0019]

[0020] Where Y kt is the dynamic load index of the t-th unit time node of the k-th cloud computing power circuit, X kt-i is the load characteristic data of the ti-th unit time lag node of the k-th cloud computing power circuit, α is a constant term, φ i is the i-th lagged autoregressive coefficient, θ j is the jth sliding average coefficient, ∈ t-j is the error term of the sliding average node at the tjth unit time, ∈ t is the error term of the t-th unit time, p is the total number of autoregressive orders, and q is the total number of sliding average orders.

[0021] Preferably, historical cloud computing power user rental data is obtained, computing power expenditure requirements of user rental data are analyzed, and a user computing power rental preference profile is constructed, specifically including:

[0022] Based on the historical cloud computing power user rental data, normalization is performed and the cloud computing power rental action events of the user unit time node are marked to construct a cloud computing power rental action event dataset of historical users;

[0023] PCA principal component analysis is used to perform dimensionality reduction on the cloud computing power rental action event dataset of historical users;

[0024] According to several dimensions associated with cloud computing power, an initial user computing power leasing preference profile is established; the several dimensions associated with cloud computing power include: computing power leasing time, computing power leasing configuration, computing power request scenario

[0025] Using the Euclidean distance formula, we calculate the average distance between the historical user's cloud computing power rental action event dataset and several dimensions in the initial user's computing power rental preference profile to perform initial clustering of the cloud computing power rental action event data.

[0026] For the initial clustering of cloud computing power leasing action event data, the distance between each cluster and other clusters is used to iteratively merge clusters according to the average connection method to obtain a profile of user computing power leasing preferences.

[0027] Preferably, the regression analysis is performed to predict the demand for rental tasks to be executed per unit time node, specifically including:

[0028] Based on SVR support vector regression, a user rental task prediction model is constructed;

[0029] Based on the preference values ​​of each associated dimension of the user's computing power leasing preference profile, linear mapping is performed on the associated dimension preference values ​​to form the associated dimension preference feature vector of the user's computing power leasing preference profile;

[0030] Based on the user rental task prediction model, the associated dimension preference feature vector of the user's computing power rental preference profile is used as feature data input to train the associated dimension preference feature vector to minimize the error of the rental task demand, determine the optimal hyperplane in the feature space, and predict the demand for rental tasks to be executed per unit time node as output;

[0031] The user rental task prediction model is specifically:

[0032]

[0033] Where G i′k′ is the k′th rental task demand to be executed for the i′th user computing power rental preference profile at the unit time node, δV is the positive space constraint Lagrange multiplier corresponding to the V-th dimension preference eigenvector, is the negative space constraint Lagrange multiplier corresponding to the V-th dimension preference eigenvector, P i′V is the V-th dimension preference feature vector of the computing power leasing preference portrait of the i′th user, κ() is the kernel function, and b is the bias.

[0034] Preferably, a cloud computing resource scheduling model is constructed based on the dynamic load indicators of each cloud computing circuit and the requirements of the to-be-executed leasing tasks per unit time node to maximize the utilization of the computing resources of the cloud computing circuit. The generation of dynamic cloud computing leasing scheduling tasks specifically includes:

[0035] Determine the cloud computing power expenditure corresponding to the requirements of all pending leasing tasks per time node;

[0036] Based on the random forest, a decision tree for the executable leasing tasks of each cloud computing circuit is constructed according to the cloud computing resource redundancy corresponding to the dynamic load index of each cloud computing circuit at a unit time node, and the tree is connected in series to form a cloud computing resource scheduling model.

[0037] The cloud computing power overhead corresponding to all pending leasing task demands per unit time node is substituted into the cloud computing power resource scheduling model. The cloud computing power line corresponding to each pending leasing task demand is divided into lines with the largest information gain of cloud computing power resource redundancy corresponding to the dynamic load index of the cloud computing power line, and dynamic cloud computing power leasing scheduling tasks are generated.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] This paper proposes a cloud computing power leasing and scheduling solution for mobile phones using virtual computers. Based on historical hardware load data of cloud computing circuits and user leasing data, this solution uses performance trend analysis and user preference profiling, and uses time series regression analysis to predict future leasing task demand. This solution then constructs a cloud computing power resource scheduling model to achieve efficient utilization of computing resources and smooth task scheduling. The beneficial effects include improved user experience and resource utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a flow chart of a method for leasing and scheduling mobile cloud computing power for virtual computers;

[0041] Figure 2 A flow chart of the method for evaluating the dynamic load indicators of each cloud computing power circuit;

[0042] Figure 3 A flowchart for building a user computing power leasing preference profile;

[0043] Figure 4A flow chart of a method for predicting the demand for rental tasks to be executed at a unit time node;

[0044] Figure 5 A flowchart for generating a dynamic cloud computing power leasing scheduling task method. DETAILED DESCRIPTION

[0045] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are merely examples, and those skilled in the art may conceive of other obvious variations.

[0046] Reference Figure 1 As shown, a method for leasing and scheduling mobile cloud computing power for virtual computers includes:

[0047] Based on the historical load data of the hardware of all cloud computing power lines, analyze the performance change trend of the historical load data of the hardware and evaluate the dynamic load indicators of each cloud computing power line;

[0048] Obtain historical cloud computing power user rental data, analyze the computing power expenditure requirements of user rental data, and build a profile of user computing power rental preferences;

[0049] Conduct regression analysis to predict the demand for rental tasks to be executed at a unit time node;

[0050] Based on the dynamic load indicators of each cloud computing power line and the demand for pending leasing tasks per unit time node, a cloud computing power resource scheduling model is constructed to maximize the utilization of the computing power resources of the cloud computing power line and generate dynamic cloud computing power leasing scheduling tasks.

[0051] It should be noted that the mobile cloud computing power leasing of the virtual computer in this solution refers to the local mobile phone leasing the computing power resources of the cloud virtual computer to realize local remote play.

[0052] This solution is based on historical hardware load data and user rental data from cloud computing circuits. Through performance trend analysis and user preference profiling, it uses time series regression analysis to predict future rental task demand. This allows for the construction of a cloud computing resource scheduling model to achieve efficient utilization of computing resources and smooth task scheduling. This has the beneficial effect of improving user experience and resource utilization.

[0053] Reference Figure 2 As shown in the figure, based on the historical load data of the hardware of all cloud computing power lines, the performance change trend of the historical load data of the hardware is analyzed, and the dynamic load indicators of each cloud computing power line are evaluated, including:

[0054] Based on the historical load data of the hardware of all cloud computing power lines, the data is processed smoothly according to the difference method and divided into unit time to form the historical load time series data of the hardware of all cloud computing power lines;

[0055] Based on the historical load time series data of the hardware of all cloud computing power lines, with unit time as the observation window and the historical load data of the hardware as the observation variable, the basic characteristic data of the historical load of the hardware of each cloud computing power line is marked; the basic characteristic data includes: average load, maximum load value, and minimum load value;

[0056] Using wavelet transform, we extract the load cycle characteristic data from the historical load time series data of the hardware of each cloud computing power line;

[0057] Combine the historical load basic characteristic data of the hardware of each cloud computing power line with the historical load cycle characteristic data of the hardware of each cloud computing power line to form the original data set of cloud computing power line load characteristics;

[0058] Based on ARIMA autoregressive integrated moving average, a dynamic load forecasting model for cloud computing power lines is constructed;

[0059] Substitute the original data set of cloud computing power line load characteristics into the computing power line dynamic load prediction model, use the load characteristic data of each cloud computing power line at a unit time node as input, and use the load label data of each cloud computing power line at a future unit time node as output to obtain the dynamic load index of each cloud computing power line;

[0060] The cloud computing power line dynamic load prediction model is specifically as follows:

[0061]

[0062] Where Y kt is the dynamic load index of the t-th unit time node of the k-th cloud computing power circuit, X kt-i is the load characteristic data of the ti-th unit time lag node of the k-th cloud computing power circuit, α is a constant term, φ i is the i-th lagged autoregressive coefficient, θ j is the jth sliding average coefficient, ∈ t-j is the error term of the sliding average node at the tjth unit time, ∈ t is the error term of the t-th unit time, p is the total number of autoregressive orders, and q is the total number of sliding average orders.

[0063] Understandably, since the connection between the virtual computer's cloud computing power and the mobile phone requires a public network connection, and is prone to disconnection or delays when the connection line is highly loaded, this can lead to large fluctuations in the number of real-time online game users renting the line, but has little impact on local stand-alone games. Therefore, by analyzing the dynamic load indicators of each cloud computing power line, we can provide planning targets for subsequent mobile phone cloud computing power lines.

[0064] Reference Figure 3 As shown in the figure, obtaining historical cloud computing power user rental data, analyzing the computing power expenditure requirements of user rental data, and building a user computing power rental preference profile specifically includes:

[0065] Based on the historical cloud computing power user rental data, normalization is performed and the cloud computing power rental action events of the user unit time node are marked to construct a cloud computing power rental action event dataset of historical users;

[0066] PCA principal component analysis is used to perform dimensionality reduction on the cloud computing power rental action event dataset of historical users;

[0067] According to several dimensions associated with cloud computing power, an initial user computing power leasing preference profile is established; the several dimensions associated with cloud computing power include: computing power leasing time, computing power leasing configuration, computing power request scenario

[0068] Using the Euclidean distance formula, we calculate the average distance between the historical user's cloud computing power rental action event dataset and several dimensions in the initial user's computing power rental preference profile to perform initial clustering of the cloud computing power rental action event data.

[0069] For the initial clustering of cloud computing power leasing action event data, the distance between each cluster and other clusters is used to iteratively merge clusters according to the average connection method to obtain a profile of user computing power leasing preferences.

[0070] It is understandable that the purpose of clustering analysis based on historical cloud computing power user rental data and constructing a user computing power rental preference profile is to find out the user's behavioral habits to facilitate subsequent line screening and matching. For example: based on the analysis of historical cloud computing power user rental data, the user attributes obtained are as follows: 1. User groups with high computing power and high latency sensitivity, 2. User groups with low computing power and low latency sensitivity, 3. Users with medium computing power requirements but low latency requirements, and the corresponding labels are: competitive players, light online game players, and single-player strategy players.

[0071] Reference Figure 4 As shown, for regression analysis, the demand for rental tasks to be executed per unit time node is predicted to include:

[0072] Based on SVR support vector regression, a user rental task prediction model is constructed;

[0073] Based on the preference values ​​of each associated dimension of the user's computing power leasing preference profile, linear mapping is performed on the associated dimension preference values ​​to form the associated dimension preference feature vector of the user's computing power leasing preference profile;

[0074] Based on the user rental task prediction model, the associated dimension preference feature vector of the user's computing power rental preference profile is used as feature data input to train the associated dimension preference feature vector to minimize the error of the rental task demand, determine the optimal hyperplane in the feature space, and predict the demand for rental tasks to be executed per unit time node as output;

[0075] The user rental task prediction model is specifically:

[0076]

[0077] Where G i′k′ is the k′th rental task demand to be executed for the i′th user computing power rental preference profile at the unit time node, δ V is the positive space constraint Lagrange multiplier corresponding to the V-th dimension preference eigenvector, is the negative space constraint Lagrange multiplier corresponding to the V-th dimension preference eigenvector, P i′V is the V-th dimension preference feature vector of the computing power leasing preference portrait of the i′th user, κ() is the kernel function, and b is the bias.

[0078] As can be understood, by leveraging SVR support vector regression technology, combined with the associated dimension preference feature vectors of user computing power leasing preference profiles, and minimizing prediction errors through model training, the optimal hyperplane in the feature space is determined, allowing for accurate prediction of pending leasing task demand per unit time node. This beneficial effect is to improve prediction accuracy and provide decision support for the rational scheduling of leasing tasks and resource allocation.

[0079] Reference Figure 5 As shown in the figure, based on the dynamic load indicators of each cloud computing power line and the demand for pending leasing tasks per unit time node, a cloud computing power resource scheduling model is constructed to maximize the utilization of the computing power resources of the cloud computing power line. The generation of dynamic cloud computing power leasing scheduling tasks specifically includes:

[0080] Determine the cloud computing power expenditure corresponding to the requirements of all pending leasing tasks per time node;

[0081] Based on the random forest, a decision tree for the executable leasing tasks of each cloud computing circuit is constructed according to the cloud computing resource redundancy corresponding to the dynamic load index of each cloud computing circuit at a unit time node, and the tree is connected in series to form a cloud computing resource scheduling model.

[0082] The cloud computing power overhead corresponding to all pending leasing task demands per unit time node is substituted into the cloud computing power resource scheduling model. The cloud computing power line corresponding to each pending leasing task demand is divided into lines with the largest information gain of cloud computing power resource redundancy corresponding to the dynamic load index of the cloud computing power line, and dynamic cloud computing power leasing scheduling tasks are generated.

[0083] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions merely illustrate the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for leasing and scheduling mobile cloud computing power for virtual computers, characterized in that: include: Based on the historical load data of the hardware of all cloud computing power lines, analyze the performance change trend of the historical load data of the hardware and evaluate the dynamic load indicators of each cloud computing power line; Obtain historical cloud computing power user rental data, analyze the computing power expenditure requirements of user rental data, and build a profile of user computing power rental preferences; Conduct regression analysis to predict the demand for rental tasks to be executed at a unit time node; Based on the dynamic load indicators of each cloud computing circuit and the demand for pending leasing tasks per unit time node, a cloud computing resource scheduling model is constructed to maximize the utilization of the computing resources of the cloud computing circuit and generate dynamic cloud computing leasing scheduling tasks; Among them, based on the historical load data of the hardware of all cloud computing power lines, the performance change trend of the historical load data of the hardware is analyzed, and the dynamic load indicators of each cloud computing power line are evaluated. Specifically, the following are included: Based on the historical load data of the hardware of all cloud computing power lines, the data is processed smoothly according to the difference method and divided into unit time to form the historical load time series data of the hardware of all cloud computing power lines; Based on the historical load time series data of the hardware of all cloud computing power lines, with unit time as the observation window and the historical load data of the hardware as the observation variable, the basic characteristic data of the historical load of the hardware of each cloud computing power line is marked; the basic characteristic data includes: average load, maximum load value, and minimum load value; Using wavelet transform, we extract the load cycle characteristic data from the historical load time series data of the hardware of each cloud computing power line; Combine the historical load basic characteristic data of the hardware of each cloud computing power line with the historical load cycle characteristic data of the hardware of each cloud computing power line to form the original data set of cloud computing power line load characteristics; Based on ARIMA autoregressive integrated moving average, a dynamic load forecasting model for cloud computing power lines is constructed; Substitute the original data set of cloud computing power line load characteristics into the computing power line dynamic load prediction model, use the load characteristic data of each cloud computing power line at a unit time node as input, and use the load label data of each cloud computing power line at a future unit time node as output to obtain the dynamic load index of each cloud computing power line; The cloud computing power line dynamic load prediction model is specifically as follows: Where Y kt is the dynamic load index of the t-th unit time node of the k-th cloud computing power circuit, X kt-i is the load characteristic data of the ti-th unit time lag node of the k-th cloud computing power circuit, α is a constant term, φ i is the i-th lagged autoregressive coefficient, θ j is the jth sliding average coefficient, ∈ t-j is the error term of the sliding average node at the tjth unit time, ∈ t is the error term of the t-th unit time, p is the total number of autoregressive orders, and q is the total number of sliding average orders.

2. A method for leasing and scheduling mobile cloud computing power for virtual computers according to claim 1, characterized in that: Obtain historical cloud computing power user rental data, analyze the computing power expenditure requirements of user rental data, and build a profile of user computing power rental preferences. Specifically, the following are included: Based on the historical cloud computing power user rental data, normalization is performed and the cloud computing power rental action events of the user unit time node are marked to construct a cloud computing power rental action event dataset of historical users; PCA principal component analysis is used to perform dimensionality reduction on the cloud computing power rental action event dataset of historical users; Establishing an initial user computing power leasing preference profile based on several dimensions associated with cloud computing power; the several dimensions associated with cloud computing power include: computing power leasing time, computing power leasing configuration, and computing power request scenario; Using the Euclidean distance formula, we calculate the average distance between the historical user's cloud computing power rental action event dataset and several dimensions in the initial user's computing power rental preference profile to perform initial clustering of the cloud computing power rental action event data. For the initial clustering of cloud computing power leasing action event data, the distance between each cluster and other clusters is used to iteratively merge clusters according to the average connection method to obtain a profile of user computing power leasing preferences.

3. A method for leasing and scheduling mobile cloud computing power for virtual computers according to claim 2, characterized in that: For regression analysis, the forecast of the rental task demand to be executed per unit time node specifically includes: Based on SVR support vector regression, a user rental task prediction model is constructed; Based on the preference values ​​of each associated dimension of the user's computing power leasing preference profile, linear mapping is performed on the associated dimension preference values ​​to form the associated dimension preference feature vector of the user's computing power leasing preference profile; According to the user rental task prediction model, the associated dimension preference feature vector of the user computing power rental preference portrait is used as the feature data input to train the associated dimension preference feature vector to minimize the error of the rental task demand, determine the optimal hyperplane in the feature space, and predict the demand for rental tasks to be executed at a unit time node as the output.

4. A method for leasing and scheduling mobile cloud computing power for virtual computers according to claim 3, characterized in that: The user rental task prediction model is specifically: Where G i′k′ is the i-th prediction unit time node ′ The k′th rental task requirement to be executed by the user’s computing power rental preference profile, δ V is the positive space constraint Lagrange multiplier corresponding to the V-th dimension preference eigenvector, is the negative space constraint Lagrange multiplier corresponding to the V-th dimension preference eigenvector, P i′V For the i ′ The V-th dimension preference feature vector of the user's computing power leasing preference portrait, κ() is the kernel function, and b is the bias.

5. A method for leasing and scheduling mobile cloud computing power for virtual computers according to claim 4, characterized in that: Based on the dynamic load indicators of each cloud computing circuit and the demand for pending leasing tasks per unit time node, a cloud computing resource scheduling model is constructed to maximize the utilization of the computing resources of the cloud computing circuit. The dynamic cloud computing leasing scheduling tasks generated specifically include: Determine the cloud computing power expenditure corresponding to the requirements of all pending leasing tasks per time node; Based on the random forest, a decision tree for executing leasing tasks on each cloud computing circuit is constructed according to the cloud computing resource redundancy corresponding to the dynamic load indicators of each cloud computing circuit at a unit time node, and the tree is connected in series to form a cloud computing resource scheduling model. The cloud computing power overhead corresponding to all pending leasing task demands per unit time node is substituted into the cloud computing power resource scheduling model. The cloud computing power line corresponding to each pending leasing task demand is divided into lines with the largest information gain of cloud computing power resource redundancy corresponding to the dynamic load index of the cloud computing power line, and dynamic cloud computing power leasing scheduling tasks are generated.

Images