WO2014101532A1

WO2014101532A1 - Method and device for analyzing program running performance

Info

Publication number: WO2014101532A1
Application number: PCT/CN2013/085302
Authority: WO
Inventors: 赵家程; 崔慧敏; 冯晓兵
Original assignee: 华为技术有限公司
Priority date: 2012-12-26
Filing date: 2013-10-16
Publication date: 2014-07-03
Also published as: CN103902443A; CN103902443B

Abstract

Provided are a method and device for analyzing program running performance, which relate to the technical field of networks, and can perform analysis on respective performance interference of multichannel programs running simultaneously, thereby improving the scheduling efficiency of resources and the utilization rate of hardware resources in the program running process. The method comprises: acquiring a feature vector of each program in a running state in a preset program set; acquiring respective performance interference parameters of at least two jointly running programs in the preset program set, the performance interference parameters being respective performance reduction rates of at least two jointly running programs when the feature vectors of at least two jointly running programs in the preset program set interact; and performing curve fitting on the feature vectors of at least two jointly running programs and the performance interference parameters of at least two jointly running programs, so as to generate a performance interference function model. The present invention is applied to program performance interference analysis.

Description

The present invention claims to be submitted to the Chinese Patent Office on December 26, 2012, and the application number is 201210576264.2, and the invention name is "a program operation performance analysis method and device, and the Chinese patent application is preferred. The entire contents of which are incorporated herein by reference.

The present invention relates to the field of network technologies, and in particular, to a method and an apparatus for analyzing program running performance.

BACKGROUND OF THE INVENTION Cloud computing is an Internet-based computing method in which hardware and software resources and information provided by the Internet are shared and provided to computers and other devices as needed. This requires a cloud computing service provider's server to have powerful computing power, and this powerful computing power is composed of a large number of on-chip multi-core processors. However, since the on-chip multi-core system shares various resources, in order to ensure the performance of the high-quality service (QoS) priority program, the high QoS priority program is usually prohibited from being shared with other programs. Running, making the utilization of hardware resources in the cloud computing environment very low. In order to solve this problem, the prior art generally analyzes the performance interference situation by combining the respective pressure scores of the two programs running together and the sensitivity curve of the performance to the pressure, so that the system can utilize the performance. The interference situation provides the decision basis for the scheduler, so that the program with low performance interference rate can run simultaneously with the high priority program, that is, the program with low performance interference rate and the high priority program exist in one data center at the same time, thereby improving hardware. Utilization of resources. In the process of implementing the above-mentioned program performance interference analysis, the inventors have found that at least the following problems exist in the prior art: The prior art can only analyze the performance interference situation of two commonly-run programs, and cannot be more than two common The running program analyzes the performance interference situation, but more and more programs running on the same on-chip multi-core processor nowadays can not effectively analyze the performance interference of more than two co-operating programs, which will affect the hardware. Utilization of resources.

Summary of the invention Embodiments of the present invention provide a program running performance analysis method and apparatus, which are capable of analyzing performance interference of multiple concurrently running programs, thereby improving resource scheduling efficiency and utilization of hardware resources during program running. In order to achieve the above objective, the embodiment of the present invention uses the following technical solutions: In a first aspect, a method for analyzing program performance is provided, including: acquiring feature vectors of each program in an operating state in a preset program set; Obtaining respective performance interference parameters of at least two programs that are commonly run in the preset program set, where the performance interference parameters are interactions of feature vectors of at least two programs that are commonly run in the preset program set Determining a rate of decline of respective performances of at least two co-operating programs; curve fitting a feature vector of the at least two co-operating programs with performance interference parameters of the at least two co-operating programs to generate a performance interference function model. In a first possible implementation, according to the first aspect, after the generating the performance interference function model, the method further includes: acquiring feature vectors of at least two programs in an operating state; and operating according to the at least two programs The lower feature vector and the performance interference function model calculate respective performance interference parameters of the at least two programs in an operating state. In a second possible implementation, in combination with the first aspect or the first possible implementation, the feature vector of the at least two cooperating programs and the performance of the at least two cooperating programs The interference parameter is subjected to curve fitting, and the performance interference function model is generated, including: adding, to the spatial coordinate system, the respective performance interference parameters of the at least two commonly-run programs in the preset program set acquired; a data amount of all performance interference parameters in the spatial coordinate system and a feature vector setting function form of the program in the preset assembly; in the spatial coordinate system, according to the set function form A feature vector of at least two co-operating programs interferes with performance of the at least two co-operating programs The parameters are curve-fitted to obtain a fitted curve, and a performance interference function model is generated. In a third possible implementation manner, according to the second possible implementation manner, the data quantity according to all performance interference parameters added to the space coordinate system and the program in the preset program set Before the feature vector setting function form, the method further includes: repeatedly acquiring respective performance interference parameters of the at least two programs that are commonly run in the preset program set, and adding the re-obtained performance interference parameter to the space coordinate system Until the number of performance interference parameters in the spatial coordinate system reaches a predetermined threshold. In a fourth possible implementation, in combination with the first aspect or the first possible implementation manner, or the second possible implementation manner, or the third possible implementation manner, the feature vector is a program in a running state. The competitive characteristics of the shared resources, including: shared cache, shared prefetcher, shared memory, shared bandwidth, and shared input and output devices. The second aspect provides a program running performance analyzing apparatus, including: a feature vector acquiring unit, configured to acquire a feature vector of each program in an operating state in a preset program set; and a parameter acquiring unit, configured to acquire the a performance interruption parameter of each of the at least two programs running in the preset program, the performance interference parameter being the program of the at least two co-operating programs in the preset program set acquired by the feature vector acquisition unit a rate of decrease of respective performances of the at least two co-operating programs when the feature vector interacts; a curve fitting unit, configured to acquire a feature vector of the at least two co-operating programs acquired by the feature vector acquiring unit The performance interference parameter of the at least two co-operating programs acquired by the parameter obtaining unit performs curve fitting to generate a performance interference function model. In a first possible implementation, according to the second aspect, the device further includes: the feature vector acquiring unit, configured to acquire feature vectors of at least two programs in an operating state; and a parameter calculating unit, configured to: And acquiring, according to the feature vector obtained by the feature vector acquiring unit, the feature vector of the at least two programs in an operating state and the curve fitting unit The performance interference function model calculates respective performance interference parameters of the at least two programs in an operating state. In a second possible implementation, in combination with the second aspect or the first possible implementation, the curve fitting unit includes: a parameter storage subunit, configured to acquire the pre-obtained by the parameter obtaining unit a performance interference parameter of each of the at least two co-operating programs is added to the spatial coordinate system; a function setting sub-unit is configured to store all performances of the sub-unit into the spatial coordinate system according to the parameter a data amount of the interference parameter and a feature vector setting function form of the program in the preset program acquired by the feature vector acquiring unit; a curve fitting subunit, configured to perform according to the space coordinate system The function form set by the function setting subunit is configured to acquire the feature vector of the at least two cooperating programs acquired by the feature vector acquiring unit and the at least two acquired by the parameter acquiring unit The performance of the running program interferes with the parameters to perform curve fitting, obtains a fitting curve, and generates a performance interference function model. In a third possible implementation manner, according to the second possible implementation manner, the device further includes: the parameter acquiring unit, configured to repeatedly acquire at least two jointly running in the preset program set The respective performance interference parameters of the program, and the re-acquired performance interference parameters are added to the spatial coordinate system by the parameter storage sub-unit, until the number of the performance interference parameters in the spatial coordinate system reaches a predetermined width value. In a fourth possible implementation manner, in combination with the first aspect or the first possible implementation manner, or the second possible implementation manner, or the third possible implementation manner, the feature vector is that the program is shared in the running state. The competitive characteristics of the resources include: shared cache, shared prefetcher, shared memory, shared bandwidth, and shared input and output devices. A program running performance analysis method and apparatus provided by an embodiment of the present invention, by curve fitting a feature vector and a performance interference parameter of at least two programs that are commonly run, The performance interference function model is used to analyze the performance interference of multiple simultaneous programs by the performance interference function model, thereby improving the efficiency of resource scheduling and the utilization of hardware resources during the running process.

BRIEF DESCRIPTION OF DRAWINGS FIG. 1 is a flowchart of a method for analyzing program running performance according to an embodiment of the present invention. FIG. 2 is a flowchart of another program running performance analysis method according to an embodiment of the present invention; FIG. 3 is a structural flowchart of a program running performance analyzing apparatus according to an embodiment of the present invention; FIG. Another flowchart of the program running performance analysis device is shown in FIG. 5 is a structural flowchart of a program running performance analysis device according to another embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A method and apparatus for analyzing program running performance according to an embodiment of the present invention will be described in detail below with reference to the accompanying drawings. Nowadays, with more and more programs running on the same on-chip multi-core processor, in order to ensure the performance of high-priority programs, it is usually easy to prohibit the high-priority program from running together with other programs, resulting in a very low Resource utilization. Therefore, in order to solve the problem of low resource utilization, a method for predicting the performance interference of a multi-channel program running together is needed, thereby scheduling the program, so that the high-priority program can have a low performance interference rate. Programs run together to increase hardware resource utilization. The embodiment of the present invention provides a program running performance analysis method. Referring to FIG. 1, before performing the method provided by the embodiment of the present invention, a suitable assembly is first determined, and a suitable assembly needs to be strongly representative of the overall input program, so when selecting Need to follow certain principles. First of all, the assembly should be extensive, that is, the behavioral features of the program should cover a wide range, and need to cover all aspects of the program feature vector. Specifically, it refers to the program in the assembly. It should cover different types such as computationally intensive and intensive intensive. Secondly, the assembly should be targeted, that is, the program in the assembly should be adjusted according to the behavior characteristics of the input program. For example, if the input application is mostly online, the program is centralized. The program should also be based on the online service program. Specifically, the specific steps of the program running performance analysis method are as follows:

101. The program running performance analysis device acquires a feature vector of each program in a running state in a preset program set. Optionally, the feature vector described in this embodiment is a competitive feature of the shared resource in the running state of the program, where the shared resource includes: a shared cache, a shared prefetcher, a shared memory, a shared bandwidth, and Shared input and output devices. Specifically, step 101 is mainly for quantifying the behavior characteristics of the input program, and the behavior characteristics of the input program are represented here by feature vectors. The feature vector of the program refers to the demand characteristics of the shared resource when the program is running. The shared resource includes: shared cache, shared prefetcher, shared memory, shared bandwidth broadband, and shared input and output devices. Specifically, in terms of the competing characteristics of the shared resources at runtime, the program's competitive features for the shared cache can use the program cache hit rate, the number of cache misses per unit time, or the number of cache misses per million instructions. To characterize; the competitive features of the shared prefetcher can be characterized by the number of prefetches per unit time or the number of prefetches per million instructions; the bandwidth characteristics of the shared bandwidth are characterized by bandwidth traffic; The contention of the memory is characterized by the amount of memory consumed; the competing characteristics of the shared input and output device are characterized by the number of times the input and output read and write requests are made, and the number of bytes per read and write request. After quantifying the demand characteristics of each shared resource, these quantized values together constitute the feature vector of the program. The competitive feature (ie, feature vector) of each program in the preset assembly on the shared resource is obtained by means of a suitable program profiling, and can be described as ^C '(c' ¹ 'c' ² ' + + +' ^c )(' = '' ). Wherein, since the accuracy of the present invention is related to the scale of the experimental data, the size of the assembly directly affects the accuracy of the resulting performance interference function model, and the larger the assembly size, the more accurate the performance interference function model is. High, but at the same time, the longer the program needs to be parsed, so it can be adjusted according to the size of the assembly. The balance between the accuracy of the performance model and the profiling time.

1 02. The program running performance analysis device acquires respective performance interference parameters of at least two programs that are commonly run in a preset program set, where the performance interference parameter is a feature vector of at least two programs that are commonly run in a preset program set. The rate of decline in performance of each of the at least two co-operating programs. Specifically, if m programs are randomly selected from the program (the selected programs can be repeatedly selected) as a workload, the performance degradation ratio of each program when the m programs are run together is detected, and the obtained m groups are obtained. The data is added to the spatial coordinate system. Wherein, the performance interference parameter of the program can be expressed as a function ^ ¹ ''...' - '): ⁷ ^'^ ¹ '...'^" ¹ - ') , the meaning of the function formula indicates that at least two are running together The performance of the program interferes with the function vector of the program itself and the function vector of other programs running together. Among them, Ρ^ ^'^' ··· ^»- ¹ ) means ^ 'and program '...' ¹ performance changes during joint operation, and the order of '...' - ¹ does not affect the value of the performance interference parameter (ie, the arbitrarily changing the order of the feature vectors other than c'' does not affect the performance change of c''). 1 03. The program running performance analysis device performs curve fitting on a feature vector of at least two commonly running programs and a performance interference parameter of at least two commonly running programs to generate a performance interference function model. Further, as shown in FIG. 2, step 203 specifically includes:

1 03a, the program running performance analysis device adds the respective performance interference parameters of at least two commonly-run programs in the preset preset program to the space coordinate system. Optionally, before step 103b, the method further includes: the program running performance analysis device repeatedly acquiring the performance interference parameters of the at least two commonly running programs in the preset program assembly, and adding the re-obtained performance interference parameters to the spatial coordinates So, until the number of performance interference parameters in the spatial coordinate system reaches a predetermined threshold. Specifically, repeatedly obtaining performance interference parameters of at least two programs that are commonly run in a preset program set, and performing performance interference of each of the performance interference parameters of each of the at least two commonly-run programs obtained by running each time The parameter is added to the space coordinate system until the data amount of the performance interference parameter in the space coordinate system reaches a certain threshold value set in advance, and only the density of the performance interference parameter in the space coordinate system reaches a certain threshold value. , In order to achieve a high accuracy of the fitted fitting curve, of course, the higher the density of the performance interference parameters in the spatial coordinate system, the higher the accuracy of the fitted fitting curve.

103b. The program running performance analyzing device sets the function form according to the data amount of all the performance interference parameters added to the space coordinate system and the feature vector of the program in the preset assembly. Specifically, before performing the curve fitting, the function form of the function to be fitted is first determined, and the function form first satisfies the function characteristic of the performance interference parameter function, and then according to the data of the performance interference parameter added in the space coordinate system. The quantity and the feature vector of the program are flexibly selected, and the function form may use a polynomial function or an exponential function, etc., or a function form in which two functions are combined.

103c, the program running performance analysis device performs curve fitting on the feature vector of the at least two co-operating programs and the performance interference parameter of the at least two co-operating programs according to the set function form in the space coordinate system to obtain a fitting curve. , Generate a performance interference function model. Wherein, the independent variable of the performance interference function model is a value of a feature vector of at least two programs that are commonly operated, and the function value is a performance degradation rate ^Pif of the program of interest in at least two programs running together ^(t i ¹

And you can use different fitting tools, such as Ma tl ab (matrix lab), Or ig in (scientific drawing, data analysis software) and other graphic data analysis tools. Optionally, the method further includes: after obtaining the performance interference function model, performing performance interference analysis on any at least two programs that are commonly run according to the performance interference function model.

104. The program running performance analysis device acquires a feature vector of at least two programs in an operating state. Perform a program profiling of all the procedures mentioned in step 104 to collect the feature vectors of all the programs parsed by the process.

1 05. The program running performance analysis device calculates the respective properties of at least two programs in the running state according to the feature vector and the performance interference function model of at least two programs in the running state. Can interfere with parameters. The program running performance analysis method provided by the embodiment of the present invention generates a performance interference function model by curve fitting the feature vector and the performance interference parameter of at least two commonly operated programs, and then uses the performance interference function model to multi-channel The performance of each program running at the same time is analyzed to improve the efficiency of resource scheduling and the utilization of hardware resources during the running of the program. An embodiment of the present invention provides a program running performance analyzing apparatus. Referring to FIG. 3, the program running performance analyzing apparatus 3 includes: a feature vector acquiring unit 3 1 , a parameter acquiring unit 32 , and a curve fitting unit 3 3 , wherein The feature vector obtaining unit 31 is configured to acquire feature vectors of each program in an operating state in a preset program set. The parameter obtaining unit 32 is configured to acquire a performance interference parameter of each of the at least two programs that are commonly run in the preset program set, where the performance interference parameter is at least two common in the preset program set acquired by the feature vector acquiring unit 32. The feature vector of the running program interacts with the rate of decline of the respective performance of at least two programs that are running together. The curve fitting unit 3 3 is configured to curve the feature vector of the at least two commonly-run programs acquired by the feature vector acquiring unit 31 and the performance interference parameter of at least two commonly-run programs acquired by the parameter acquiring unit 32. Combine, generate a performance interference function model. Optionally, the device further includes: a parameter calculation unit 34, wherein: the feature vector obtaining unit 31 is further configured to acquire feature vectors of at least two programs in an operating state. The parameter calculation unit 34 is configured to calculate at least two programs in the running state according to the feature vector of the at least two programs acquired by the feature vector acquiring unit 31 in the running state and the performance interference function model fitted by the curve fitting unit 33 The respective performance interference parameters. Optionally, the feature vector mentioned in the embodiment of the present invention is a competitive feature of the shared resource in the running state of the program, where the shared resource includes: a shared cache, a shared prefetcher, a shared memory, a shared bandwidth, and Shared input and output devices. Further, as shown in FIG. 4, the curve fitting unit 3 3 further includes: a parameter storage sub-unit 331, a function setting sub-unit 332, and a curve-fitting sub-unit 333, wherein: the parameter storage sub-unit 331 is configured to use at least two of the preset programs acquired by the parameter obtaining unit 32 to operate together. The respective performance interference parameters of the program are added to the spatial coordinate system. The function setting sub-unit 332 is configured to set, according to the data quantity of all performance interference parameters added to the space coordinate system by the parameter storage sub-unit 331 and the feature vector of the program in the preset program set acquired by the feature vector acquiring unit 31. The form of the function. The curve fitting sub-unit 333 is configured to acquire the feature vector and parameter acquiring unit 32 of the at least two commonly-run programs acquired by the feature vector acquiring unit 31 according to the function form set by the function setting sub-unit 332 in the space coordinate system. The performance interference parameters of at least two commonly-run programs are obtained by curve fitting, and a fitting curve is obtained to generate a performance interference function model. Further, the parameter obtaining unit 32 is further configured to repeatedly obtain the performance interference parameters of the at least two programs that are commonly run in the preset program set, and add the re-acquired performance interference parameters to the space through the parameter storage sub-unit 331. Coordinate system, until the number of performance interference parameters in the space coordinate system reaches a predetermined threshold. The program running performance analysis device provided by the embodiment of the present invention generates a performance interference function model by curve fitting the feature vector and the performance interference parameter of at least two commonly-run programs, and then multi-channels through the performance interference function model. The performance of each program running at the same time is analyzed to improve the efficiency of resource scheduling and the utilization of hardware resources during the running of the program. FIG. 5 is a schematic structural diagram of a program running performance analyzing apparatus according to still another embodiment of the present invention. The program running performance analyzing apparatus 5 includes at least one processor 51, a memory 52, a communication bus 53, and at least one communication interface 54. The communication bus 53 is used to implement connection and communication between the above components, and the communication interface 54 is used for connecting and communicating with external devices. The communication bus 53 can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus, or an extended industry standard architecture (Extended Industry Standard). Architecture, referred to as EISA) bus. The communication bus 53 can be divided into an address bus, a data bus, a control bus, and the like. For ease of representation, only one thick line is shown in Figure 5, but it does not mean that there is only one bus or one type of bus. The processor 51 can be: a central processing unit (CPU), an application specific integrated circuit (ASIC), a digital signal processor (DSP), an off-the-shelf programmable gate array (FPGA), or the like. Programmable logic device. The communication interface 54 is used to connect the program running performance analysis device and the communication network, and the communication network includes: an Ethernet, a radio access network (RAN), a wireless local area network (WLAN), or the like. . The memory 52 can be any available medium that can be accessed by a computer, including but not limited to: read only memory (ROM), random access memory (RAM), or disk storage, flash memory. A programmable memory such as a programmable read only memory or an electrically erasable programmable memory, a register, or the like. The program code that needs to be executed is stored in the memory 52, and the program code may specifically include: a feature vector acquiring unit 521, a parameter obtaining unit 522, and a curve fitting unit 523. The processor 51 is configured to execute the unit stored in the memory 52. When the unit is executed by the processor 51, the following functions are implemented: a feature vector obtaining unit 521, configured to acquire each program in a preset program set. The feature vector in the running state. The parameter obtaining unit 522 is configured to acquire a performance interference parameter of each of the at least two programs that are commonly run in the preset program set, where the performance interference parameter is at least two common in the preset program set acquired by the feature vector acquiring unit 521. The feature vector of the running program interacts with the rate of decline of the respective performance of at least two programs that are running together. The curve fitting unit 523 is configured to perform curve fitting on the feature vector of the at least two commonly-run programs acquired by the feature vector acquiring unit 521 and the performance interference parameter of the at least two commonly-run programs acquired by the parameter acquiring unit 522 , Generate a performance interference function model. Optionally, the memory 52 further includes: a parameter calculation unit 524, where: The feature vector obtaining unit 521 is further configured to acquire feature vectors of at least two programs in an operating state. The parameter calculation unit 524 is configured to calculate at least two programs in the running state according to the feature vector of the at least two programs acquired by the feature vector acquiring unit 521 in the running state and the performance interference function model fitted by the curve fitting unit 523. The respective performance interference parameters. Optionally, the feature vector mentioned in the embodiment of the present invention is a competitive feature of the shared resource in the running state of the program, where the shared resource includes: a shared cache, a shared prefetcher, a shared memory, a shared bandwidth, and Shared input and output devices. Further, optionally, the curve fitting unit 523 further includes: a parameter storage subunit, a function setting subunit, and a curve fitting subunit, wherein: the parameter storage subunit, the pre-acquisition obtained by the parameter obtaining unit 522 The performance interference parameters of at least two programs running together are added to the spatial coordinate system. a function setting subunit, a data amount for all performance interference parameters added to the spatial coordinate system according to the parameter storage subunit, and a feature vector setting function of the program in the preset program acquired by the feature vector obtaining unit 521 form. a curve fitting sub-unit, configured to obtain, in the spatial coordinate system, the feature vector of the at least two commonly-running programs acquired by the feature vector acquiring unit 521 according to the function form set by the function setting sub-unit and the parameter acquiring unit 522 The performance interference parameters of at least two co-operating programs are curve-fitted to obtain a fitted curve, and a performance interference function model is generated. Further, the parameter obtaining unit 522 is further configured to repeatedly obtain performance interference parameters of at least two programs that are commonly run in the preset program set, and add the re-acquired performance interference parameters to the space coordinates through the parameter storage subunit. System, until the number of performance interference parameters in the spatial coordinate system reaches a predetermined threshold. The program running performance analysis device provided by the embodiment of the present invention generates a performance interference function model by curve fitting the feature vector and the performance interference parameter of at least two commonly-run programs, and then multi-channels through the performance interference function model. The performance of each program running at the same time is analyzed to improve the efficiency of resource scheduling during the running of the program. And utilization of hardware resources.

The above is only the specific embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily think of changes or substitutions within the technical scope of the present invention. It should be covered by the scope of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

Rights request

1. A method for analyzing program running performance, characterized by including: obtaining the feature vector of each program in the running state in a preset assembly; obtaining at least two co-running programs in the preset assembly The performance interference parameter of each program, the performance interference parameter is the degradation rate of the respective performance of the at least two co-running programs when the feature vectors of at least two co-running programs in the preset program interact; The characteristic vectors of the at least two co-running programs are curve-fitted with the performance interference parameters of the at least two co-running programs to generate a performance interference function model.

2. The method according to claim 1, characterized in that, after generating the performance interference function model, it further includes: obtaining the feature vectors of at least two programs in the running state; The characteristic vectors under and the performance interference function model calculate the respective performance interference parameters of the at least two programs in the running state.

3. The method according to claim 1 or 2, characterized in that: performing curve fitting on the characteristic vectors of the at least two co-running programs and the performance interference parameters of the at least two co-running programs. , generating a performance interference function model, including: adding the obtained performance interference parameters of at least two co-running programs in the preset assembly to the spatial coordinate system; according to the parameters added to the spatial coordinate system The data amount of all performance interference parameters and the feature vectors of the programs in the preset assembly set a functional form; in the spatial coordinate system, the at least two co-running programs are configured according to the set functional form. The feature vector of the program is curve-fitted with the performance interference parameters of the at least two co-running programs to obtain a fitting curve and generate a performance interference function model.

4. The method according to claim 3, characterized in that, the data volume of all performance interference parameters added to the spatial coordinate system and the feature vector setting of the program in the preset assembly set Before the functional form, also include: Repeatedly obtain the performance interference parameters of at least two co-running programs in the preset program set, and add the re-acquired performance interference parameters to the spatial coordinate system until the performance interference parameters in the spatial coordinate system are The number of interference parameters reaches a predetermined threshold.

5. The method according to any one of claims 1 to 4, characterized in that the feature vector is the competition feature of the program for shared resources in the running state, and the shared resources include: shared cache, shared Prefetcher, shared memory, shared bandwidth, and shared input and output devices.

6. A program running performance analysis device, characterized by including: a feature vector acquisition unit, used to acquire the feature vector of each program in the running state in a preset assembly set; a parameter acquisition unit, used to acquire all The performance interference parameters of at least two co-running programs in the preset program set are the performance interference parameters of at least two co-running programs in the preset program set obtained by the feature vector acquisition unit. The degradation rate of the respective performance of the at least two co-running programs when the feature vectors of the at least two co-running programs interact; a curve fitting unit, used to calculate the feature vectors of the at least two co-running programs obtained by the feature vector acquisition unit Perform curve fitting with the performance interference parameters of the at least two co-running programs acquired by the parameter acquisition unit to generate a performance interference function model.

7. The device according to claim 6, characterized in that, the device further includes: the feature vector acquisition unit, further used to acquire the feature vectors of at least two programs in a running state; a parameter calculation unit, used to Calculate the running status of the at least two programs based on the feature vectors of the at least two programs in the running state obtained by the feature vector acquisition unit and the performance interference function model fitted by the curve fitting unit. Below are their respective performance interference parameters.

8. The device according to claim 6 or 7, characterized in that, the curve fitting unit includes: a parameter storage subunit, used to store the preset value obtained by the parameter acquisition unit. The performance interference parameters of at least two co-running programs in the assembly are added to the spatial coordinate system; a function setting subunit is used to store all the performance interference parameters added by the subunit to the spatial coordinate system according to the parameters. The amount of data and the feature vector setting function form of the program in the preset assembly set obtained by the feature vector acquisition unit; a curve fitting subunit, used to calculate the function according to the function in the spatial coordinate system The functional form set by the setting sub-unit is adapted to the feature vectors of the at least two co-running programs acquired by the feature vector acquisition unit and the at least two co-run programs acquired by the parameter acquisition unit. Perform curve fitting on the performance interference parameters of the program, obtain the fitting curve, and generate a performance interference function model.

9. The device according to claim 8, characterized in that, the device further includes: the parameter acquisition unit, further configured to repeatedly acquire the respective performance of at least two co-running programs in the preset program set interference parameters, and add the re-obtained performance interference parameters to the spatial coordinate system through the parameter storage subunit until the number of performance interference parameters in the spatial coordinate system reaches a predetermined threshold.

10. The device according to any one of claims 6 to 9, characterized in that the feature vector is the competition feature of the program for shared resources in the running state, and the shared resources include: shared cache, shared prefetcher, shared memory, shared bandwidth, and shared input and output devices.