JP2012234448A - Software performance sequentially-predicting method and predicting system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a software performance sequentially-predicting method and a predicting system method that achieve performance improvement and quality improvement by avoiding performance measurement through embedded software execution on an actual embedded system and by making the determination through predictive estimate of performance in large-scale software development conducted by multiple people.SOLUTION: With respect to software that is executed before registration of source code change, a software performance sequentially-predicting system, at the time of software development, executes steps of: calculating a performance reduction ratio of an input source code; and providing notification of a determination result externally when the performance reduction ratio is equal to or greater than a predetermined value.

Description

  The present invention relates to a software performance sequential prediction method and system, and more particularly, to a source code registration method and a performance management method during software development.

  With the increasing functionality and functionality of embedded systems, the amount of description of embedded software running on embedded processors, which are the main components of embedded systems, is growing. Ensuring the functionality and quality of embedded software has become a critical issue due to this high functionality and multi-functionality. Furthermore, in addition to high functionality and multi-functionality, embedded systems also include performance requirements such as real-time responsiveness.

  In the current embedded software development field, along with high functionality and multi-functionality, in order to ensure functional quality, there is a tendency to prioritize complex functional implementation, and performance measures such as real-time performance are implemented after functional implementation May be. Therefore, a sufficient time for performance measures cannot be ensured, and a problem that performance is not achieved may occur.

  For performance measures, run the embedded software on the actual embedded system, identify the location that is the performance bottleneck based on the performance profile results, implement the performance measures for the embedded software, and then again on the embedded system. This is realized by an iterative process of executing the embedded software that measures the above and measuring the performance. There are two problems with this method.

  First, an embedded system that actually executes embedded software is required for performance measurement, which increases the development cost. The second problem is that since it is necessary to repeatedly execute execution with performance measurement of embedded software, the development period is prolonged due to the execution time.

  In response to these issues, a method has been proposed in which embedded software is not executed on an actual embedded system, but the performance of the software, such as the processing time or the storage capacity required for processing, is evaluated in the state of the source code. .

  Patent Document 1 discloses a “system performance prediction method and method based on performance measurement of software components” that performs performance prediction in units of software components. The software component includes a communication component, an application component, a database component, and the like, and has a database in which performance information such as CPU usage time is defined for each component. In this method, the total processing time when a plurality of components are continuously executed is predicted based on the performance information stored in the database. According to Patent Document 1, it is described that the performance of the modified software can be predicted without executing the modified embedded software on the actual embedded system. Furthermore, by measuring the performance again with the modified software and maintaining the pair of the modified software and its performance information in the database, the latest software performance can always be managed, and a new modification is made. It is described that it is possible to maintain high prediction accuracy with respect to the performance of the software.

JP 2004-220453 A

  As described above, the code size of embedded software is increasing with the increase in functionality and functionality of embedded systems. As the code scale increases, the number of embedded software developers required to realize one embedded system increases, and one series of functions (in other words, modules) in an embedded system is developed by multiple developers. Development style has become mainstream. In particular, when the development scale is large, a plurality of software developers, such as offshore development, may develop a series of functions by being dispersed both spatially and temporally.

  The performance prediction method shown in Patent Document 1 performs performance prediction for one software modification, and when performance degradation does not occur, the modified software is registered in the database, and again based on the modified software. This is a method for measuring performance and registering the performance information in a database. In this method, when multiple people modify a series of functions, it is necessary to measure the performance in units of individual software modifications. As a result, the modified embedding by the actual embedding system, which is the above-mentioned problem, is required. The performance measurement through the execution of the software must be performed repeatedly.

  In this way, in large-scale software development with multiple people, it is necessary to measure the performance by executing embedded software on an actual embedded system, and as a result, the occurrence of rework at the later stage of the design process, and consequently Increasing development time and development costs are inevitable.

  The problem to be solved by the present invention is to avoid the performance measurement by executing the embedded software on the actual embedded system, and to determine the performance based on the estimated estimation of the performance, thereby enabling the performance improvement and the quality improvement. The object is to provide a prediction method and a prediction system method.

  Representative examples of the invention disclosed in the present application are as follows. The software performance sequential prediction method of the present invention is a software performance sequential prediction method for predicting the performance of software accompanying a change in source code by a software performance prediction system, wherein the software performance prediction system is a database in which the software is registered. And is configured to be connectable to a plurality of terminals, accepting input of initial source code or changed source code accompanying development of the software from at least one of the plurality of terminals, as source code information The ratio of the performance of the old and new source code is calculated by comparing the performance of the software by the modified source code with the performance of the software by the old source code included in the registered source code information. And the reduction ratio of the performance There determines whether more than a predetermined value, when the performance reduction ratio exceeds the predetermined value, and notifies the determination result to the outside.

  According to the present invention, performance measurement due to execution of embedded software is avoided, and whether or not the generated change information can be adopted is determined by predicting performance. As a result, even in software development that is a joint work of a plurality of people, a change in performance is managed in each software correction unit, and a performance deterioration portion among a plurality of correction portions can be easily realized. Therefore, it is possible to suppress the performance degradation of the embedded software and reduce the development cost.

It is a figure for demonstrating the structure of the whole software performance prediction system as a 1st Example of this invention. It is the figure which showed the detail of the database structure part which registers | registers the produced | generated change information in the database in 1st Example, and produces | generates source code information. It is a figure for demonstrating two types of information stored in a database in a 1st Example. It is a figure which shows the concept of the initial source code and change source code which were registered into the database by the version management system etc. at the time of software development in this invention. It is a figure for demonstrating the structure of the determination part in a 1st Example. It is a figure for demonstrating the structure of the change information production | generation part in a 1st Example. 2 shows an overall operation flow of the software performance prediction system in the first embodiment. The operation | movement flow of the determination process in a 1st Example is shown. The operation | movement flow of the database registration process in a 1st Example is shown. It is a figure for demonstrating the whole structure of the software performance prediction system which becomes a 2nd Example of this invention. It is a figure for demonstrating the structure of the determination part in a 2nd Example. It is the figure which showed the structure of the change information generation part in a 2nd Example. It is the figure which showed the operation | movement flow of the software performance prediction system which becomes the 3rd Example of this invention. It is a figure for demonstrating the structure of the determination part in a 3rd Example. It is the figure which showed the output screen which displays a performance increase / decrease ratio to a project manager in the software performance prediction system which becomes the 4th Example of this invention.

Here, in order to solve the above-described problems, an outline of a typical embodiment of the invention disclosed in the present application will be described regarding a software performance prediction method and system.
The present invention relates to software that runs on a CPU, which is one of the components of an embedded system, and at the time of software development, the developer uses the initial source code and the previous source code (old source code) that changed the source code. The present invention relates to a method of registering new changed software (hereinafter, changed source code) in a database such as a version control system. Specifically, when a change source code is requested to be registered to the version control system, the change location of this time is changed in the past by referring to a database storing past change history data (hereinafter referred to as change information). If there is a history, the information stored in the change information is compared with the current change, and the performance is estimated and estimated without performing the performance measurement. Assembler instructions are used for comparison. With this method, developers can prevent performance degradation without performing performance measurement when changing source code, reducing total system design man-hours and improving software quality.

  The software performance prediction system according to the present invention receives the changed source code as input. The determination unit uses the source code and various information stored in a database to be described later to determine whether or not the changed source code has a performance impact (increase / decrease in processing time) due to the change. A determination result as a result is output, and the determination result is input to a communication unit which is a display device for notifying the software developer. In addition, the determination unit has additional functions such as what changes have been made in the source code compared to the source code registered in the past, and what kind of performance impact (increase / decrease in processing time) Means for generating change information indicating whether or not The change information is input to the database construction unit, and the database construction unit has means for performing format conversion and the like in order to store the change information in the database.

  The determination unit receives the changed source code and the source code registered in the past, determines which function of the changed source code has changed from the difference between the two inputs, and outputs the changed function name. Also, the modified source code is compiled and the number of assembler instructions for the target modified function is calculated. Based on the name of the changed function, the determination unit predicts the performance increase / decrease (increase / decrease in processing time) caused by the target modification from the performance increase / decrease ratio information among the number of function unit assembler instructions stored in the database and the change information. And output as change information. When the obtained performance increase / decrease ratio exceeds a certain value, the registrant is notified. Thereby, remarkable deterioration of performance can be prevented in advance.

In this way, according to the present invention, information for each individual change on the source code is stocked and combined with static analysis information or already acquired execution information, so that newly added changes can be performed. Or predicting the effect of the effect, it is possible to reduce the cost of performance measurement and to notify the developer before registration, so that significant performance degradation can be prevented in advance.
Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

  An outline of a software performance prediction method according to an embodiment of the present invention will be described first. The present invention relates to software that runs on a CPU, which is one of the components of an embedded system, and registers the software changed by the developer (hereinafter referred to as changed source code) in a database such as a version control system when the software is developed. It is about the method. When registering the change source code in the version control system, if there is a risk that the performance will be significantly degraded by estimating the impact on the performance of the change caused by the change source code, the software developer or version Notify the administrator of the management system of the impact of performance degradation and enable performance degradation to be prevented before registering the changed source code. Specifically, when a change source code is requested to be registered to the version control system, the change location of this time is changed in the past by referring to a database storing past change history data (hereinafter referred to as change information). If there is a history, the information stored in the change information is compared with the current change, and the performance is estimated and estimated without performing the performance measurement. Assembler instructions are used for comparison. The assembler instruction is a machine language for causing the CPU to perform processing described in a form that is easy for humans to understand, and is a low-level language with finer processing granularity than a high-level language such as C language. By using this assembler instruction, it is possible to estimate the performance effect (increase in processing time / decrease in throughput) due to small changes. With this method, developers can prevent performance degradation without performing performance measurement when changing source code, reducing total system design man-hours and improving software quality.

  Next, a software performance prediction method according to an embodiment of the present invention will be described in detail with reference to FIGS.

  FIG. 1 is a diagram for explaining the configuration of a software performance prediction system 10 according to the present invention. The software performance prediction system 10 is a system that predicts performance influence (increase / decrease in processing time) due to a change in order to suppress significant performance deterioration due to a change in source code. A plurality of software developers Pa to Pn terminals 20 (20-1 to 20-n) are connected to the software performance prediction system 10 via a network 40. The plurality of software developers are members who develop embedded software including one series of functions, and exist in a spatially and temporally distributed manner. The plurality of terminals 20-1 to 20-n belong to any of a plurality of development bases C1 to Cn that are spatially dispersed. Further, the terminal 30 of the project manager M is also connected to the software performance prediction system 10.

  The software performance prediction system 10 receives the initial source code 202, the modified source code 500, and other information necessary for software development from the terminal 20 and the terminal 30, and is registered and stored in the database 200 described later.

  The software performance prediction system 10 includes a database configuration unit 100, a determination unit 300, and a notification unit 400 as functions realized by executing a program on a computer. The terminals 20 and 30 access the software performance prediction system 10 and edit and input the initial source code 202, the modified source code 500 and other information necessary for software development, and the software performance prediction system 10 necessary for software development. It has a function to acquire information.

  The determination unit 300 of the software performance prediction system 10 uses the initial source code 202, the changed source code 500, and various information stored in the database 200, so that the changed source code 500 has a performance impact (processing time) due to the change. The determination result 700 is output, and the determination result is output to the terminal 20 of the software developer or the terminal 30 of the project manager M via the input / output unit 900. It inputs into the communication part 400 for notifying. This determination result is also displayed on the input / output unit 900.

  Further, a further function of the determination unit 300 is that the change source code 500 has been changed compared to the registered source code registered in the past, and what performance influence ( A means for generating change information 250 indicating whether the processing time is increased or decreased. The change information 250 is input to the database construction unit 100, and the database construction unit 100 has means for performing format conversion and the like in order to store the change information 250 in the database 200.

  FIG. 2 is a diagram showing details of the database configuration unit 100. The database configuration unit 100 has a function of registering the initial source code 202, the changed source code 500, and the generated changed information 250 in the database 200 and generating the source code information 201. The database construction unit 100 includes an information input / output control unit 101 that controls data input / output, a compile unit 102 that generates a call graph 106 and an assembler instruction 107 from the initial source code 202, and a call graph output from the compile unit 102 A function structure information generation unit 103 that generates a reference relationship (function structure information 203) from 106, and a function unit assembler that extracts the function unit assembler instruction number 204 by dividing the assembler instruction 107 output from the compilation unit 102 into function unit assembler instructions The instruction number extraction unit 104 is configured.

  When the change information 250 is input, the information input / output control unit 101 registers it in the database 200. When the initial source code 202 or the modified source code 500 is input, it is passed to the compilation unit. When the function structure information 203 and the function unit assembler instruction count 204 are input, they are registered in the database 200 as source code information 201 together with the initial source code 202 and the modified source code 500. The compiling unit 102 compiles the input initial source code 202 and the changed source code 500 to generate a call graph 106 and an assembler instruction 107. The function structure information generation unit 103 outputs information as to how many times each function is called from the input call graph 106 as function structure information 203. The function unit assembler instruction number extraction unit 104 divides the assembler instruction number into function units from the input assembler instruction 107 and outputs the result as function unit assembler instruction number 204.

  FIG. 3A is a diagram for explaining three types of information stored in the database 200. Source code information 201 as first information includes initial source code 202 and changed source code 500 registered in the past, function structure information 203 indicating the function structure of the source code 202 and changed source code 500, and each function. The number of assembler instructions 204 for each function derived from the compilation result of The function structure information 203 includes the number of function calls for each function. The change information 250 as the second information includes a change function name 251 indicating the function name changed in the change source code 500, and a modified function assembler instruction indicating the number of assembler instructions obtained by compiling the change function. And a performance increase / decrease ratio information 253 indicating the performance to be increased or decreased by this modification. In the database 200, other information 260 that is not held in the source code information 201 and the change information 250 and that is related to performance increase / decrease, determination results, and the like is also stored as appropriate.

  FIG. 3B is a diagram showing the concept of the initial source code and the changed source code registered in the database by the version management system or the like during software development in the present invention. The source code is sequentially updated to the latest version as the software development progresses. The source code can be decomposed into a plurality of functions Fa to Fx. One or a plurality of functions are combined into one module, and one embedded software that satisfies various required requirements is completed by these modules A to N. Each of the functions Fa to Fx is developed and updated by one or a plurality of developers in the same space or distributed space, at the same time or at different times. That is, a single function of the source code is developed by multiple people.

  FIG. 4 is a diagram for explaining the configuration of the determination unit 300 of the software performance prediction system 10. The determination unit 300 includes a change function extraction unit 301, an assembler instruction number extraction unit 302, a function call number calculation unit 303, and a change information generation unit 304. The change function extraction unit 301 receives the latest change source code 500 and the initial source code 202 or the change source code 500 registered in the past, and determines which function of the change source code 500 from the difference between the old and new input source codes. On the other hand, it is determined whether a change has occurred, and a change function name 251 is output. The assembler instruction count extraction unit 302 compiles the changed source code 500 based on the changed function name 251 and calculates the assembler instruction count 252 of the target changed function. The function call count calculation unit 303 calculates the function call count 307 of the function to be changed from the function structure information 203 based on the change function name 251. The change information generation unit 304 includes a change function name 251, a change function assembler instruction count 252, a function call count 307, a function unit assembler instruction count 204 stored in the database 200, and performance increase / decrease ratio information in the change information 250. 253 is input, and an increase / decrease in performance (increase / decrease in processing time) caused by target correction is predicted and output as change information 250.

  FIG. 5 is a diagram for explaining a means for generating change information 250 by the change information generation unit 304. The change information generation unit 304 includes a subtracter 320, a change function influence degree calculation unit 321, a multiplier 322, an adder 323, and a performance deterioration detection unit 327. The changed function assembler instruction number 252 generated by the assembler instruction number extraction unit 302 and the function unit assembler instruction number 204 stored in the database 200 are input to the subtractor 320, and a difference value 324 of the assembler instruction number generated by the change is obtained. Generated. Also, the function unit assembler instruction count 204 and the changed function name 251 are input to the changed function influence calculation unit 321. The change function influence degree calculation unit 321 generates a change function influence degree index 325 indicating how much the change function indicated by the change function name 251 occupies all functions. The effective range of the change function influence index 325 is 0 to 1, and when the target change function has little influence on all functions, the value approaches 0 and is very large for all functions. The value approaches 1 when it has an effect. In the description of this embodiment, for ease of explanation, the change function influence index 325 will be described as the number of assembler lines of the function to be changed with respect to the number of assembler instructions of all functions. However, the generation method of the change function influence index 325 is not limited with respect to the generation method, which is an example.

  The generated difference value 324 of the number of assembler instructions, the change function influence index 325, and the function call count 307 of the target change function are multiplied by the multiplier 322, and the performance increase / decrease ratio information 253 is output. The performance increase / decrease ratio information 253 indicates the ratio of the increase / decrease value of the performance caused by the target correction.

  The performance increase / decrease ratio information 253 accumulates a history as the performance increase / decrease ratio information 253 in the database 200 every time a source code is registered. By this accumulation, it is possible to hold the performance increase / decrease caused by each past source code modification. By adding the performance increase / decrease ratio information 253 registered in the past and the newly generated performance increase / decrease ratio information 253 by the adder 323, the current correction indicates the increase or decrease in performance from the original source code 202 or 500. , Change information (X) 250 can be obtained. Data relating to the performance reduction ratio in the change information (X) 250 is compared with a predetermined value α by the performance deterioration detection unit 327, and when X> α, it is detected as performance deterioration and a determination result 700 for notifying that is detected 700. The information is generated and notified to related parties via the corresponding terminal and recorded in the database 200. If no performance deterioration is detected, information on the source code of the database 200 is updated based on the change information (X) 250. The change information (X) 250 includes data related to the performance increase ratio, but this information is preferably notified to the parties concerned, but it is not essential to notify the information one by one. Regarding the use of data related to the performance increase ratio, for example, a record is left in the database 200, and the project manager M is allowed to determine whether or not to reflect in other parts of the source code having the same function. good.

  FIG. 6 shows an operation flow of the software performance prediction system 10 according to the present invention. This software performance prediction system operates when the software developer registers the changed source code in the version management system. When the input of the source code is accepted (step S100), it is determined whether or not the source code is changed (S101). When the input is the modified source code 500, the process proceeds to step S102, and when the input is the initial source code 202, the process proceeds to step S108 for generating the modified information 600. When a registration request for the changed source code 500 is issued (step S102), the changed source code 500 is input to the determination unit 300 described later (step S103). The determination unit 300 compares the change source code 500 with the change information 600 registered in the database 200 described later (step S104). When the performance reduction ratio X of the change information generated by the determination unit 300 in step S104 exceeds a certain value α (step S105), the registrant is notified of the possibility of performance degradation (step S106). Even if a certain modified source code 500 reduces performance, it may be unavoidably adopted as a part of a series of functions or embedded software depending on the state of the process. This determination should be made with comprehensive consideration of performance improvement, quality improvement, rework in the post-process of the design process, and the determination is left to the registrant or project manager M. If the modified source code 500 with reduced performance is adopted, the process proceeds to step S108 for generating modified information. If the modified source code 500 whose performance is reduced is not adopted, a corresponding instruction from the registrant or the project manager M is accepted, such as using a previous version of the source code as an alternative to the modified source code 500. . Data relating to the performance increase ratio included in the performance increase / decrease ratio information 253 is also registered in the database 200. The change information and the response instruction information created in this way are registered in the database 200 (step S110), and the series of processing ends.

  FIG. 7 shows a detailed operation flow of the determination process S104 of FIG. First, the changed function name 251, the function call count 307, and the changed function assembler instruction count 252 are extracted from the input changed source code 500. Next, a difference value 324 between the number of changed function assembler instructions 252 and the number of function unit assembler instructions 204 in the database 200 is obtained (step S1031). A change function influence index 325 is obtained from the number of function unit assembler instructions 204 and the change function name 251 (step S1032). The performance increase / decrease ratio information 253 is obtained from the difference value 324 obtained in step S1031, the change function influence index 325 obtained in step S1032, and the function call count 307 (S1033). The obtained performance increase / decrease ratio information 253 and the database performance increase / decrease ratio information 253 are added together with the change function name 251 and the number of correction function assembler instructions 252 to generate change information 250 (S1034). For example, when the input change source code is changed for the function name funcA, and the number of assembler instructions is changed to increase 5 instructions from the number of function unit assembler instructions in the change information of funcA in the database, the performance increase / decrease ratio is as follows: Ask for. Multiply the increased function by 5 instructions and the change function influence index (value between 0 and 1) indicating the specific gravity of the target function in all functions and the number of function calls (n times). The change function influence index is a ratio of the total number of assembler lines of the change function to the total number of assembler lines of the source code. The performance increase / decrease ratio of the function name funcA registered in the database is added to the obtained performance increase / decrease ratio to obtain the final performance increase / decrease ratio due to the change.

  FIG. 8 shows a detailed operation flow of the database registration process S110 of FIG. First, it is determined whether the input is the source code 202 or the change information 250 (S1100). When the input is the initial source code 202 and the update source code 500, an assembler instruction and a call graph are extracted (step S1101). The function structure information 203 and the function unit assembler instruction count 204 are generated from the extracted assembler instruction and call graph (S1002). The source code 202 and 500, the generated function structure information 203 and the function unit assembler instruction count 204 are registered in the database 200 as source code information 201. On the other hand, when the input is the change information 250 or the like (S1100), necessary information is separately registered in the database 200.

  According to the present embodiment, when the update source code is received, the performance between the old and new versions of the source code is determined based on the prediction. In order to provide software performance sequential prediction technology that can be used for embedded software development, enabling performance improvement and quality improvement, and reducing rework in the later stages of the design process. it can.

  Next, a performance sequential prediction method and system as a second embodiment of the present invention will be described. The performance is predicted using a dynamic function call in addition to the input to the determination unit of the first embodiment. In the first embodiment, the number of function calls is used when calculating the performance increase / decrease ratio. In this embodiment, however, the use of the number of dynamic function calls enables performance prediction in line with execution.

  FIG. 9 is a diagram for explaining the configuration of the software performance prediction system of this embodiment. As described in the first embodiment, the software performance prediction system 10 is a system that predicts a performance influence (increase / decrease in processing time) due to a change in order to suppress significant performance deterioration due to a change in source code. The software performance prediction system 10 receives the initial source code 202, the changed source code 500, and the execution trace 800 as inputs. The determination unit 300 ′ uses the initial source code 202, the changed source code 500, the execution trace 800, and various information stored in the database 200, which will be described later. A determination result 700 that is a result of the determination is output, and the determination result is input to the communication unit 400 for notifying the software developer. Other parts not specifically mentioned are denoted by the same reference numerals as those in the first embodiment, and their details are omitted.

  FIG. 10 is a diagram for explaining the configuration of the determination unit 300 ′ of the software performance prediction system according to the present embodiment. The function call calculation unit 303 'counts how many times the target function is dynamically called from the changed function name and the execution trace, and outputs the dynamic function call number 307'. The change information generation unit 304 receives the change function assembler increase / decrease number 252, the function unit assembler instruction number 204, the change function name 251, the dynamic function call count 307 ′, and the performance increase / decrease ratio information 253 as inputs, and outputs the change information 250. . Other parts that are not particularly mentioned are denoted by the same reference numerals and their details are omitted.

  FIG. 11 is a diagram illustrating obtaining the performance increase / decrease ratio using the dynamic call count 307 ′ in the multiplier 322 ′. Other parts not specifically mentioned are denoted by the same reference numerals as those in the first embodiment, and their details are omitted.

  According to the present embodiment, when the update source code is received, the performance between the old and new versions of the source code is determined based on the prediction. In order to provide software performance sequential prediction technology that can be used for embedded software development, enabling performance improvement and quality improvement, and reducing rework in the later stages of the design process. it can.

Next, a performance sequential prediction method and system as a third embodiment of the present invention will be described.
FIG. 12 shows an operation flow when past change information of the change function does not exist in the database. If there is no data, the change information of the function whose function function count matches the change function influence index is used as a substitute for the change information in the change function. It is checked whether past change information of the change function of the input change source code exists in the database (step S1035). If there is no past change information of the change function in the database, a function whose change function influence index matches the number of function calls is checked (S1036). If there is a matching function, the change information of the function is substituted (S1037). This is based on the premise that in general, when the notation is similar, there is a high probability of expressing the same function. As a result, performance prediction is possible even when the change information of the change function in the input change source code does not exist in the database. In other steps, parts that are not particularly mentioned are denoted by the same reference numerals and their details are omitted.

  FIG. 13 is a diagram for explaining the configuration of the determination unit 300 of the software performance prediction system 10 according to the third embodiment. The function call number calculation unit 303 and the change information generation unit 304 use function change information whose function function number matches the change function influence index and the function unit assembler instruction number 204 ′ stored in the database 200, The change information generating unit 304 is configured to input the performance increase / decrease ratio information 253 ′ of the change information 250 ′, predict the performance increase / decrease (increase / decrease in processing time) caused by the target correction, and output the change information 250. In other respects, parts not particularly mentioned are denoted by the same reference numerals and their details are omitted.

  According to the present embodiment, when the update source code is received, even if the past change information of the change function does not exist in the database, the performance of the update source code is determined using the alternative information, and there is a problem of performance degradation. If there is an error, the author is notified in advance when writing the updated source code, so that the performance and quality can be improved, and the rework in the later stage of the design process is reduced. Software performance sequential prediction technology that can be used for software development can be provided.

  Next, a performance sequential prediction method and system as a fourth embodiment of the present invention will be described. FIG. 14 is a diagram illustrating a method for outputting information for the project manager in the determination unit according to the first embodiment. The software sequential performance prediction system operates when the project developer registers the changed source code in the repository. In this embodiment, when there is a registration request, a list of performance increase / decrease ratios is generated for each developer, each module, and each development base as information for the administrator. In addition, a project plan is input, and if the performance increase / decrease ratio is greater than a predetermined value, the project manager is notified. This makes it possible to prevent significant performance degradation due to changes.

  According to this embodiment, in large-scale software development that is developed by a plurality of people, the performance between the old and new versions of the source code is determined based on the prediction, and the project manager is notified to each developer, each module, or the development base. Providing software performance sequential prediction technology that improves performance and quality, and reduces rework in the later stages of the design process, because information such as the presence or absence of performance degradation for each terminal is provided to each terminal. Can do.

10 software performance prediction system 20 (20-1 to 20-n) software developer terminal 30 project manager terminal 40 network 100 database construction unit 101 information input / output control unit 102 compilation unit 103 function structure information generation unit 104 function unit assembler Instruction extractor 106 Call graph 107 Assembler instruction 200 Database 201 Source code information 202 Initial source code 203 Function structure information 204 Number of assembler instructions per function 250 Changed information 251 Changed function name 252 Number of modified function assembler instructions 253 Performance increase / decrease ratio information 300 Determination unit 300 'determination unit 301 changed function extraction unit 302 assembler instruction number extraction unit 303 function call number calculation unit 304 change information generation unit 307 function call count 320 subtractor 321 change function influence degree calculation Part 322 multiplier 323 adder 324 difference values 325 change function influence degree indicating 400 notifying unit 500 changes the source code 700 determination result.

Claims (15)

A software performance sequential prediction method for predicting software performance associated with a source code change by a software performance prediction system,
The software performance prediction system includes a database for registering the software, and is configured to be connectable to a plurality of terminals.
Accepting input of initial source code or changed source code accompanying development of the software from at least one of the plurality of terminals, registering as source code information in the database,
Compare the performance of the software by the modified source code with the performance of the software by the old source code included in the registered source code information, and calculate the performance reduction ratio of the old and new source code,
Determining whether the reduction ratio of the performance exceeds a predetermined value;
When the performance reduction ratio exceeds the predetermined value, the determination result is notified to the outside. The software performance sequential prediction method according to claim 1,
The software is embedded software including a series of functions,
The performance reduction ratio of the old and new source code is calculated by predicting a change in processing time due to the change of the source code and performing estimation instead of performance measurement. The software performance sequential prediction method according to claim 1,
The source code is one that is sequentially updated to the latest version by adding change information from the changed source code to the initial source code,
The source code includes a plurality of decomposable functions,
Receiving the input of the changed source code for each function from at least one of the plurality of terminals existing spatially or temporally distributed;
A software performance sequential prediction method, wherein the determination result is notified to the terminal that has received the input. The software performance sequential prediction method according to claim 1,
Output information on the result of the performance reduction ratio of the old and new source code according to one of the categories of each software developer corresponding to the plurality of terminals, each development base, and each module constituting the software. A software performance sequential prediction method. The software performance sequential prediction method according to claim 1,
The software performance prediction system includes a database construction unit, a determination unit, and a notification unit,
Holding data input from the terminal in the database;
Convert the input data and register it in the database,
The determination unit predicts performance based on the input data, determines whether the performance of the software deteriorates from the predicted value,
The software performance sequential prediction method, wherein the notification unit notifies the terminal that has received the input of the deterioration of the performance. The software performance sequential prediction method according to claim 5,
In the database, the source code, function structure information, and function unit assembler instructions are stored as the source code information, and a change function name, a modified function assembler instruction number, and a performance increase / decrease ratio information are stored as the change information. Software performance sequential prediction method. The software performance sequential prediction method according to claim 6,
The source code information includes the initial source code or the modified source code registered in the past, the function structure information indicating the function structure thereof, and the number of assembler instructions for each function derived from the compilation result for each function. First information including:
The changed function name indicating the changed function name in the changed source code, the modified function assembler instruction number indicating the number of assembler instructions obtained by compiling the changed function, and the performance reduction ratio information And second information,
The software performance sequential prediction method, wherein the function structure information of the first information includes the number of function calls for each function. The software performance sequential prediction method according to claim 6,
The determination unit includes a change function extraction unit, an assembler instruction number extraction unit, a function call calculation unit, and a change information generation unit,
The change function extraction unit identifies the change location of the change source code,
The assembler instruction number extraction unit calculates the assembler instruction number of the modified source code,
The function call calculation unit calculates a function call,
A software performance sequential prediction method, wherein the change information generation unit generates change information. The software performance sequential prediction method according to claim 8,
In the change information generation unit,
The determination unit includes a subtractor, a change function influence calculation unit, a multiplier, and an adder.
In the subtractor, find the difference between the number of assembler instructions of the change function and the number of function unit assembler instructions,
In the change function influence calculation unit, a change function influence index is obtained from the number of function unit assembler instructions and the change function name,
In the multiplier, the performance increase / decrease ratio information is obtained from the difference value obtained by the subtractor, the change function influence index and the number of function calls,
The software performance sequential prediction method, wherein the adder obtains the change information from the performance increase / decrease ratio information. The software performance sequential prediction method according to claim 5,
In the database construction unit, the information input / output control unit controls input / output of information, the compile unit outputs assembler instructions and a call graph from the source code, and the function structure information generation unit generates function structure information from the call graph. A software performance sequential prediction method, wherein a function unit assembler instruction extraction unit extracts the number of assembler instructions in function units from assembler instructions. The software performance sequential prediction method according to claim 8,
When the change information of the change function in the past does not exist in the database, the change information of the function whose change function influence index matches the number of function calls is used as a substitute for the change information in the change function. Performance sequential prediction method. A software performance prediction system that predicts software performance associated with source code changes using a software performance prediction system,
A database for registering the software, and configured to be connectable to a plurality of terminals;
A function of accepting an input of an initial source code or a modified source code from at least one of the plurality of terminals and registering it in the database as source code information;
A function of calculating the performance reduction ratio of the old and new source code by comparing the performance of the software by the changed source code with the performance of the software by the old source code included in the registered source code information;
A function of determining whether the reduction ratio of the performance exceeds a predetermined value;
A software performance sequential prediction system comprising a function of notifying the determination result to the outside when the performance reduction ratio exceeds the predetermined value. The software performance sequential prediction system according to claim 12,
The software is embedded software including a series of functions,
A database construction unit for converting the data input from the terminal and registering it in the database;
A determination unit that predicts performance based on the input data and determines whether the performance of the software deteriorates from the predicted value;
A notification unit that notifies the terminal of the input of the deterioration of the performance to the outside;
The determination of the deterioration of the performance of the old and new source code is calculated based on an increase in processing time due to the change of the source code. The software performance sequential prediction system according to claim 13,
The determination unit
A change function extraction unit for identifying a change location of the change source code;
An assembler instruction number extraction unit for calculating the assembler instruction number of the modified source code;
A function call calculation unit for calculating a function call;
A software performance sequential prediction system comprising: a change information generation unit that generates change information. The software performance sequential prediction system according to claim 14,
In the change information generation unit,
The determination unit includes a subtractor, a change function influence calculation unit, a multiplier, and an adder.
In the subtractor, find the difference between the number of assembler instructions of the change function and the number of function unit assembler instructions,
In the change function influence calculation unit, a change function influence index is obtained from the number of function unit assembler instructions and the change function name,
In the multiplier, the performance increase / decrease ratio information is obtained from the difference value obtained by the subtractor, the change function influence index and the number of function calls,
The software performance sequential prediction system, wherein the adder obtains the change information from the performance increase / decrease ratio information.

Images