A performance boundary bottleneck simulation method and system for cloud computing systems

Abstract

The present invention discloses a method and system for simulating and deducing performance boundary bottlenecks of a cloud computing system. The method comprises collecting performance impact index data of a target cloud computing system; extracting a convolution kernel and introducing a convolutional neural network model; selecting a set of performance impact index data x n , input the convolutional neural network model to calculate the performance boundary bottleneck a for the nth time n If a n If convergence is achieved, the gradient of the sum of the previous n performance boundary bottleneck deduction results is derived and the final boundary bottleneck deduction result is calculated. Otherwise, the iteration continues. The present invention can calculate the performance boundary bottlenecks of various cloud computing systems, including physical servers with different hardware combinations, server clusters with different numbers of hardware, server clusters with different numbers of hardware using different application software, and different teams using different application software in server clusters with different numbers of hardware.

Description

Technical Field

[0001] The present invention relates to a performance testing technology for a computer system, and in particular to a method and system for simulating and deducing performance boundary bottlenecks of a cloud computing system. Background Art

[0002] Bottleneck testing of computer systems primarily involves testing performance limits, such as hardware performance, CPU performance, memory size, network cable configuration, and disk I / O capacity. Although these specifications are specified by hardware manufacturers at the time of manufacture, they cannot be simply calculated to determine the actual upper limit of computer system application performance. It's impossible to exhaustively test different hardware configuration combinations due to the sheer number of hardware vendors and the varying physical hardware within the same vendor. It's also impossible to determine the bottlenecks of cloud virtual machines supported by physical cloud servers with varying hardware configurations. Traditional cloud servers can't compare the maximum performance boundaries achievable on different hardware. It's also impossible to compare the computing resource consumption required for the same operation on software service systems from different vendors using the same hardware. It's also impossible to compare the time and performance of the same load operation, nor can it compare the maximum performance boundaries achieved. System performance bottlenecks cannot be individually tested in various software usage scenarios, such as the number of virtual machines a cloud server supports, the number of virtual machines capable of simultaneously opening Excel, the number of virtual machines capable of simultaneously working with Word documents, the number of virtual machines capable of simultaneously working with PowerPoint, and other software support levels. In various team application scenarios, it's difficult to pinpoint specific system performance bottlenecks. For example, how many C++ R&D engineers, testers, UI designers, salespeople, and HR personnel can a physical machine support simultaneously using virtual machines? This is because each job type has different performance requirements: some require high CPU, some high memory, some high GPU, while others simply use documents or browse the web. Testing the system boundaries of a single physical server is relatively simple, but measuring the performance boundaries of a computing cluster comprised of ten, twenty, or even three hundred physical servers is impossible. Firstly, the hardware cost is prohibitive, making it difficult to realistically construct such a server cluster. Secondly, it's difficult to test these computer clusters in every scenario and with every software use case. The sheer number of combinations and the sheer volume of testing make it impossible to manually traverse all possible scenarios. Traditional cloud server clusters struggle to define the upper limit of resources available for running virtual machines. It's also difficult to define the computing resources required for specific operations within a virtual machine, or the network resources required for remote operations. Furthermore, it's difficult to effectively assess the maximum performance of cloud server hardware and the performance degradation caused by aging. Performance testing is difficult to assess, as everyone's judgment of performance is subjective. Standardized test data is difficult to establish, and every company claims their hardware and software combination performs better. Summary of the Invention

[0003] The technical problem to be solved by the present invention is as follows: In response to the above-mentioned problems in the prior art, a method and system for simulating and deducing the performance boundary bottlenecks of cloud computing systems are provided. The present invention can calculate the performance boundary bottlenecks of various cloud computing systems, including the performance bottlenecks of physical servers with different hardware combinations, the performance bottlenecks of server clusters with different numbers under different hardware, the performance bottlenecks of different application software used by different teams in server clusters with different numbers under different hardware, and the performance bottlenecks of different application software used by different teams in server clusters with different numbers under different hardware.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0005] A performance boundary bottleneck simulation method for a cloud computing system, comprising:

[0006] 1) Collect N types of performance impact indicator data of the target cloud computing system and build a data warehouse;

[0007] 2) Traverse and select the nth type of performance impact indicator data x in the data warehouse n , the nth type of performance impact indicator data x n After normalization and conversion to an image, convolution pooling is performed to extract the convolution kernel.

[0008] 3) Substitute the convolution kernel into the preset convolutional neural network model, and iteratively calculate the performance boundary bottleneck a for the nth time through the convolutional neural network model n ;

[0009] 4) Determine the nth performance boundary bottleneck a n Convergence or whether n is less than N is established. If so, the gradient of the sum of the performance boundary bottleneck deduction results of the previous n times is derived and the final boundary bottleneck deduction result is calculated and output. Otherwise, jump to step 2) to continue iteration.

[0010] Optionally, the N types of performance impact indicator data in step 1) include three types: server hardware performance information, server software performance information, and cloud desktop software performance information, and each type of performance impact indicator data includes at least one type of performance impact indicator data.

[0011] Optionally, the server hardware performance information includes the hardware configuration of the server and the number of servers, and the hardware configuration includes part or all of the CPU, memory, disk, motherboard, and network card; the server software performance information includes part or all of the CPU usage information per second, memory usage information per second, disk I / O information per second, thread list details per second, network data information per second, CPU and memory usage details of each thread, and boot-up duration on the server; the cloud desktop software performance information includes part or all of the CPU usage information per second, memory usage information per second, disk I / O information per second, thread list details per second, network data information per second, CPU and memory usage details of each thread, and boot-up duration on the cloud desktop system.

[0012] Optionally, in step 3), the n-th performance boundary bottleneck a is iteratively calculated through the convolutional neural network model n The function expression is:

[0013] a n =f(w1*x1+w2*x2+…+w n *x n )

[0014] In the above formula, a n is the nth performance boundary bottleneck, f is the calculation formula of a single performance bottleneck, x1~x n They are the performance impact indicator data of categories 1 to n, w1~w n They are the convolution kernels corresponding to the 1st to nth types of performance impact indicator data.

[0015] Optionally, the function expression of the single performance bottleneck calculation formula f is: f = x n *n1 / e, where x n is the nth type of performance impact indicator data, and the nth type of performance impact indicator data x n The value of is normalized to between 0 and 1, n1 represents the number of nodes that simultaneously collect N types of performance impact indicator data of the target cloud computing system, and e represents the total performance of the nth type of performance impact indicator data.

[0016] Optionally, the nth performance boundary bottleneck a in step 4) n Convergence means: first calculate the performance boundary bottleneck a for the nth time n and the n-1th performance boundary bottleneck a n-1 The difference between them is taken as the performance jitter difference A; then the performance boundary bottleneck a of the nth time is calculated according to the following formula n Corresponding Softmargin loss function:

[0017]

[0018] In the above formula, n is the number of iterations, max is the maximum value function, a n is the nth performance boundary bottleneck, A is the performance jitter difference, x n is the nth type of performance impact indicator data, a is the regularization parameter, and b is the intercept parameter; if the value of the calculated Softmargin loss function is less than the preset threshold, the nth performance boundary bottleneck a is determined. n Convergence, otherwise determine the nth performance boundary bottleneck a n Not converged yet.

[0019] Optionally, the nth performance boundary bottleneck a in step 4) n Convergence means: calculating the performance boundary bottleneck a for the nth time n and the n-1th performance boundary bottleneck a n-1 The difference between them is taken as the performance jitter difference A. If the performance jitter difference A of the specified number of times in the recent period is less than the preset threshold, the performance boundary bottleneck a for the nth time is determined. n Convergence, otherwise determine the nth performance boundary bottleneck a n Not converged yet.

[0020] Optionally, in step 4), calculating the final boundary bottleneck deduction result after deriving the gradient of the sum of the performance boundary bottleneck deduction results of the previous n times includes: first, calculating the partial derivative of the gradient of the sum of the performance boundary bottleneck deduction results of the previous n times according to the following formula;

[0021]

[0022] In the above formula, H i is the partial derivative of the gradient of the sum of the performance boundary bottleneck deduction results of the previous n times, f represents the calculation formula of a single performance bottleneck, a1~a n Represent the 1st to nth performance boundary bottlenecks respectively;

[0023] Then, the final boundary bottleneck deduction result is calculated according to the following formula:

[0024]

[0025] In the above formula, is the final boundary bottleneck deduction result, is the weight parameter, x n For the nth type of performance impact indicator data, if the number of iterations n is less than the preset threshold, the weight parameter of the nth iteration is The value is a preset constant value, otherwise the nth weight parameter The value is the partial derivative H of the gradient of the sum of the performance boundary bottleneck deduction results of the previous n-1 times i .

[0026] In addition, the present invention also provides a performance boundary bottleneck simulation and deduction system for a cloud computing system, comprising a microprocessor and a memory connected to each other, wherein the microprocessor is programmed or configured to execute the steps of the performance boundary bottleneck simulation and deduction method for a cloud computing system.

[0027] In addition, the present invention also provides a computer-readable storage medium, which stores a computer program, and the computer program is used to be executed by a computer device to implement the steps of the performance boundary bottleneck simulation and deduction method for a cloud computing system.

[0028] Compared with the prior art, the present invention mainly has the following advantages:

[0029] 1. The present invention includes collecting performance impact index data of the target cloud computing system, extracting the convolution kernel and introducing the convolutional neural network model; sequentially selecting a set of performance impact index data x n Input the convolutional neural network model to iteratively calculate the performance boundary bottleneck a for the nth time n ; If the nth performance boundary bottleneck a n If convergence occurs, the gradient of the sum of the previous n performance boundary bottleneck deduction results is derived and the final boundary bottleneck deduction result is calculated and output. Otherwise, iteration continues. The present invention can calculate the performance boundary bottlenecks of various cloud computing systems, including physical servers with different hardware combinations, server clusters with different numbers of hardware, server clusters with different numbers of hardware using different application software, and different teams using different application software in server clusters with different numbers of hardware.

[0030] 2. The present invention uses an iterative approach to train and use the model multiple times, and corrects and repairs the weight calculation in the event of human interference or system software and application software crashes, memory leaks and losses, thread locks, human operation errors, sudden power outages, abnormal network environments, etc., and has the advantages of good robustness and high reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 Schematic diagram of the basic process of the method of the embodiment of the present invention.

[0032] Figure 2 Schematic diagram of the architecture of a cloud computing system in an embodiment of the present invention.

[0033] Figure 3 Schematic diagram of the principle of extracting the convolution kernel in an embodiment of the present invention. DETAILED DESCRIPTION

[0034] Example 1:

[0035] like Figure 1 As shown, the performance boundary bottleneck simulation method for a cloud computing system in this embodiment includes:

[0036] 1) Collect N types of performance impact indicator data of the target cloud computing system and build a data warehouse;

[0037] 2) Traverse and select the nth type of performance impact indicator data x in the data warehouse n , the nth type of performance impact indicator data x n After normalization and conversion to an image, convolution pooling is performed to extract the convolution kernel.

[0038] 3) Substitute the convolution kernel into the preset convolutional neural network model, and iteratively calculate the performance boundary bottleneck a for the nth time through the convolutional neural network model n ;

[0039] 4) Determine the nth performance boundary bottleneck a n Convergence or whether n is less than N is established. If so, the gradient of the sum of the performance boundary bottleneck deduction results of the previous n times is derived and the final boundary bottleneck deduction result is calculated and output. Otherwise, jump to step 2) to continue iteration.

[0040] like Figure 2 As shown, in this embodiment, the target cloud computing system includes multiple servers, such as servers for R&D personnel, servers for testers, servers for designers, servers for human resources personnel, servers for accounting and finance personnel, etc. The hardware configurations of each server are different, and they can be server clusters or single-point servers. The N types of performance impact indicator data in step 1) include three types: server hardware performance information, server software performance information, and cloud desktop software performance information, and each type of performance impact indicator data includes at least one performance impact indicator data. In this embodiment, the server hardware performance information includes the hardware configuration of the server and the number of servers, and the hardware configuration includes CPU, memory, disk, motherboard, and network card (optional); the server software performance information includes the CPU usage information per second, memory usage information per second, disk I / O information per second, thread list details per second, network data information per second, CPU and memory usage details of each thread, and boot duration (optional); the cloud desktop software performance information includes the CPU usage information per second, memory usage information per second, disk I / O information per second, thread list details per second, network data information per second, CPU and memory usage details of each thread, and boot duration (optional) on the cloud desktop system.

[0041] In this embodiment, server hardware performance information, server software performance information, and cloud desktop software performance information are used as performance impact indicator data to construct a data warehouse. The data warehouse uses a MySQL database to store performance impact indicator data. In step 2 of this embodiment, when collecting server software performance information and cloud desktop software performance information from the target cloud computing system, a performance collection program is run on the cloud desktops of each team, including R&D personnel, testers, designers, human resources personnel, and accounting and finance personnel. This data is collected and stored in the MySQL database to form the data warehouse.

[0042] like Figure 3 As shown, the data collected from the cloud desktop is stored in a data warehouse in the form of a MySQL database. The data warehouse includes performance impact indicator data such as CPU, memory, disk I / O, network data, etc. for different hardware configurations, different cluster numbers, different team departments, and different application scenarios. This article traverses the data warehouse to select the nth type of performance impact indicator data x n , the nth type of performance impact indicator data x n After normalization and conversion into an image, the convolution kernel is extracted through convolution pooling (the corresponding network structure includes convolution layer and pooling layer). Through the above method, the performance impact index data x of the nth category can be obtained. n The weights of various performance influencing indicator data can be obtained, such as the convolution kernel (weight) of the number of servers, CPU, memory, disk, motherboard, network card, etc.

[0043] In this embodiment, in step 3), the convolutional neural network model is used to iteratively calculate the performance boundary bottleneck a for the nth time. n The function expression is:

[0044] a n =f(w1*x1+w2*x2+…+w n *x n )

[0045] In the above formula, a n is the nth performance boundary bottleneck, f is the calculation formula of a single performance bottleneck, x1~x n They are the performance impact indicator data of categories 1 to n, w1~w n The convolution kernels corresponding to the performance impact index data of the 1st to nth categories are respectively. In this embodiment, the function expression of the single performance bottleneck calculation formula f is: f = x n *n1 / e, where x n is the nth type of performance impact indicator data, and the nth type of performance impact indicator data x nThe value of is normalized to between 0 and 1, n1 represents the number of nodes that simultaneously collect N types of performance impact indicator data of the target cloud computing system, and e represents the total performance of the nth type of performance impact indicator data.

[0046] Because changes in each type of performance-influencing indicator data will have a single-dimensional impact on system performance, the function expression of the convolutional neural network model in this embodiment is:

[0047] a n =f(w1*x1+w2*x2+…+w n *x n )

[0048] In the above formula, a n represents the nth performance bottleneck, f represents the calculation formula of a single performance bottleneck, x1~x n Respectively represent the performance impact index data of the 1st to nth input convolutional neural network model, w1~w n They represent the convolution kernel values of the 1st to nth iterations of the convolutional neural network model, and the function expression of the calculation formula f of the single performance bottleneck is: f = x n *n1 / e, where x n The performance impact index data of the convolutional neural network model is input for the nth time. The value of the performance impact index data is normalized to between 0 and 1. n1 represents the number of nodes that simultaneously collect the server software performance information and the cloud desktop software performance information of the target cloud computing system, and e represents the total amount of performance.

[0049] In this embodiment, the nth performance boundary bottleneck a in step 4) n Convergence means: first calculate the performance boundary bottleneck a for the nth time n and the n-1th performance boundary bottleneck a n-1 The difference between them is taken as the performance jitter difference A; then the performance boundary bottleneck a of the nth time is calculated according to the following formula n Corresponding Softmargin loss function:

[0050]

[0051] In the above formula, n is the number of iterations, max is the maximum value function, a n is the nth performance boundary bottleneck, A is the performance jitter difference, x n is the nth type of performance impact indicator data, a is the regularization parameter, and b is the intercept parameter; if the value of the calculated Softmargin loss function is less than the preset threshold, the nth performance boundary bottleneck a is determined. n Convergence, otherwise determine the nth performance boundary bottleneck a nNot converged yet. Because not all software and hardware performance parameter data are linear, there are various situations in the actual cloud desktop operation that affect the actual value. In this embodiment, the data point loss function across the dividing line - Softmargin loss function is used to determine the nth performance boundary bottleneck a n Whether it converges can ensure the convergence of the convolution kernel (weight) of various performance-influencing index data.

[0052] In this embodiment, in step 4), calculating the final boundary bottleneck deduction result after derivation of the gradient of the sum of the performance boundary bottleneck deduction results of the previous n times includes: first, calculating the partial derivative of the gradient of the sum of the performance boundary bottleneck deduction results of the previous n times according to the following formula;

[0053]

[0054] In the above formula, H i is the partial derivative of the gradient of the sum of the performance boundary bottleneck deduction results of the previous n times, f represents the calculation formula of a single performance bottleneck, a1~a n Represent the 1st to nth performance boundary bottlenecks respectively;

[0055] Then, the final boundary bottleneck deduction result is calculated according to the following formula:

[0056]

[0057] In the above formula, is the final boundary bottleneck deduction result, is the weight parameter, x n For the nth type of performance impact indicator data, if the number of iterations n is less than the preset threshold, the weight parameter of the nth iteration is The value is a preset constant value, otherwise the nth weight parameter The value is the partial derivative H of the gradient of the sum of the performance boundary bottleneck deduction results of the previous n-1 times i In this embodiment, the weight parameter of the nth time is When the value is a preset constant value, the preset constant value is obtained by reading the configuration information from the ini file.

[0058] Furthermore, this embodiment provides a performance boundary bottleneck simulation and deduction system for a cloud computing system, comprising an interconnected microprocessor and memory, the microprocessor being programmed or configured to execute the steps of the aforementioned performance boundary bottleneck simulation and deduction method for a cloud computing system. Furthermore, this embodiment provides a computer-readable storage medium storing a computer program, which is configured to be executed by a computer device to implement the steps of the aforementioned performance boundary bottleneck simulation and deduction method for a cloud computing system.

[0059] Example 2:

[0060] This embodiment is basically the same as the first embodiment, with the main difference being that in step 4) the nth performance boundary bottleneck a n Convergence is determined in different ways. In this embodiment, the nth performance boundary bottleneck a in step 4) n Convergence means: calculating the performance boundary bottleneck a for the nth time n and the n-1th performance boundary bottleneck a n-1 The difference between them is taken as the performance jitter difference A. If the performance jitter difference A of the specified number of times in the recent period is less than the preset threshold, the performance boundary bottleneck a for the nth time is determined. n Convergence, otherwise determine the nth performance boundary bottleneck a n The advantage of this method is that it is computationally simple and convenient.

[0061] Furthermore, this embodiment provides a performance boundary bottleneck simulation and deduction system for a cloud computing system, comprising an interconnected microprocessor and memory, the microprocessor being programmed or configured to execute the steps of the aforementioned performance boundary bottleneck simulation and deduction method for a cloud computing system. Furthermore, this embodiment provides a computer-readable storage medium storing a computer program, which is configured to be executed by a computer device to implement the steps of the aforementioned performance boundary bottleneck simulation and deduction method for a cloud computing system.

[0062] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, may be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including the instruction device, which implements the function specified in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0063] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the concept of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A performance boundary bottleneck simulation method for a cloud computing system, characterized by: include: 1) Collect the target cloud computing system N Classify performance impact indicator data and build a data warehouse; 2) Traverse and select the first n Class performance impact indicator data x n , will n Class performance impact indicator data x n After normalization and conversion to an image, convolution pooling is performed to extract the convolution kernel. 3) Substitute the convolution kernel into the preset convolutional neural network model, and iteratively calculate the first n performance bottleneck a n ; 4) Determine the n performance bottleneck a n Convergence or n Less than N Is it established? If so, n After taking the gradient derivative of the sum of the performance boundary bottleneck deduction results, the final boundary bottleneck deduction result is calculated and output. Otherwise, jump to step 2) to continue iteration; Step 4) Lieutenant General n The final boundary bottleneck deduction result is calculated by taking the gradient derivative of the sum of the performance boundary bottleneck deduction results. The calculation includes: First, the previous bottleneck deduction result is calculated according to the following formula: n The partial derivative of the gradient of the sum of the performance boundary bottleneck deduction results; ， In the above formula, For the front n The partial derivative of the gradient of the sum of the performance boundary bottleneck deduction results, f The calculation formula for a single performance bottleneck is: a 1~ a n Respectively represent the 1st to n Secondary performance bottlenecks; Then, the final boundary bottleneck deduction result is calculated according to the following formula: = In the above formula, is the final boundary bottleneck deduction result, is the weight parameter, x n For the n Class performance affects indicator data, if the number of iterations n If the value is less than the preset threshold, n The weight parameter The value is a preset constant value, otherwise n The weight parameter The value is the front n-1 The partial derivative of the gradient of the sum of the performance boundary bottleneck deduction results .

2. The performance boundary bottleneck simulation method for a cloud computing system according to claim 1 is characterized in that: In step 1) N The performance impact indicator data includes three categories: server hardware performance information, server software performance information, and cloud desktop software performance information, and each category of performance impact indicator data includes at least one performance impact indicator data.

3. The performance boundary bottleneck simulation method for a cloud computing system according to claim 2 is characterized in that: The server hardware performance information includes the hardware configuration of the server and the number of servers. The hardware configuration includes part or all of the CPU, memory, disk, motherboard, and network card; the server software performance information includes the CPU usage information per second, memory usage information per second, disk I / O information per second, thread list details per second, network data information per second, CPU and memory usage details of each thread, and part or all of the boot duration on the server; the cloud desktop software performance information includes the CPU usage information per second, memory usage information per second, disk I / O information per second, thread list details per second, network data information per second, CPU and memory usage details of each thread, and part or all of the boot duration on the cloud desktop system.

4. The performance boundary bottleneck simulation method for a cloud computing system according to claim 1, characterized in that: In step 3), the convolutional neural network model is used to iteratively calculate the n performance bottleneck a n The function expression is: a n = f ( w 1* x 1+ w 2* x 2+ … + w n * x n ) In the above formula, a n For the n Secondary performance bottlenecks, f is the calculation formula for a single performance bottleneck, x 1~ x n They are No. 1~ n Class performance impact indicator data, w 1~ w n They are No. 1~ n The convolution kernel corresponding to the class performance impact indicator data.

5. The performance boundary bottleneck simulation method for a cloud computing system according to claim 4, characterized in that: Formula for calculating single performance bottleneck f The function expression is: f = x n * n 1 / e ,in, x n For the n Class performance impact indicator data, and the n Class performance impact indicator data x n The value of is normalized to between 0 and 1. n 1 means collecting the target cloud computing system's N The number of nodes that affect the index data of the class performance, e Indicates the n Class performance affects the total amount of performance of indicator data.

6. The performance boundary bottleneck simulation method for a cloud computing system according to claim 5, characterized in that: Step 4) n performance bottleneck a n Convergence means: first calculate the n performance bottleneck a n Hedi n- 1 performance boundary bottleneck a n-1 The difference between them is the performance jitter difference A Then calculate the first n performance bottleneck a n Corresponding Softmargin loss function: , In the above formula, n is the number of iterations, max is the maximum value function, a n For the n Secondary performance bottlenecks, A For poor performance jitter, x n For the n Class performance impact indicator data, is the regularization parameter, b is the intercept parameter; if the calculated value of the Softmargin loss function is less than the preset threshold, it is determined that n performance bottleneck a n Converge, otherwise determine n performance bottleneck a n Not converged yet.

7. The performance boundary bottleneck simulation method for a cloud computing system according to claim 5, characterized in that: Step 4) n performance bottleneck a n Convergence means: Calculate the n performance bottleneck a n Hedi n- 1 performance bottleneck a n-1 The difference between them is the performance jitter difference A If the performance jitter of the specified number of consecutive times is poor A are all less than the preset threshold, then the n performance bottleneck a n Converge, otherwise determine n performance bottleneck a n Not converged yet.

8. A performance bottleneck simulation system for a cloud computing system, comprising a microprocessor and a memory connected to each other, characterized in that: The microprocessor is programmed or configured to execute the steps of the performance boundary bottleneck simulation method for a cloud computing system as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and the computer program is used to be executed by a computer device to implement the steps of the performance boundary bottleneck simulation and deduction method for a cloud computing system as described in any one of claims 1 to 7.

Images