JP5906844B2 - Computer parameter tuning program, method, and apparatus - Google Patents

Description

  The present invention relates to a program, method, and apparatus for tuning computer parameters.

  In order to optimize a business environment using a computer that executes an application program, it is necessary that parameters in the computer are appropriately set so that the performance of the computer can be sufficiently extracted.

  Dynamic program analysis is known as a technique for dynamically analyzing computer performance. In the dynamic program analysis, information related to a computer (for example, CPU execution time for various functions) is collected through execution of an application program. The collected information is analyzed, and the setting parameters of the computer are adjusted (particularly, parameter tuning related to the CPU) so that the performance of the computer can be maximized for the application program.

  A CPU profiler is used as a tool for collecting computer execution information. The CPU profiler is one of performance analysis tools, and collects various information (execution time of each function, etc.) at the time of executing an application program. By using this profiler, it is possible to statistically acquire computer execution information such as the frequency of calling various functions and the CPU execution time related to them. By analyzing the execution information of the computer, it is understood how to tune various parameters related to the execution of the CPU included in the computer. If each parameter is optimized by this parameter tuning, the computer's ability to execute the application program can be utilized more effectively. In this specification, statistically collecting computer execution information is called profiling. Information acquired by profiling is called profiling data.

  Note that the operating environment of the application program changes dynamically. For this reason, it is known that appropriate values of various parameter values also change corresponding to this change. Therefore, for example, when a situation (trouble) in which the response of the computer is delayed occurs during operation of the application program, parameter retuning work may be required in response to this situation. In many cases, retuning of this parameter can eliminate the response delay of the computer.

In this case, the computer (target machine) in which a problem such as a response delay has occurred is mostly used for business operations. Therefore, it is not realistic to find the optimum parameter by moving the parameter freely on the target machine currently in use and repeating trial and error. This is because there is a possibility that unexpected troubles may occur simultaneously with business operations. Therefore, in order to tune parameters of a target machine used for business operations, for example, it is often necessary to use a test machine having a computer environment of the target machine. In this case, for example, the following procedure is necessary.
(1) First, profiling of a running computer (target machine) executing an application program is performed. By performing this profiling, profiling data for the parameters at that time is acquired.
(2) Analyze this profiling data.
(3) Acquire target environment information (parameter value, usage DB environment, etc.).
(4) Build the environment of the target machine in the test machine.
(5) The parameter value is changed in the environment of the test machine, and an appropriate parameter value is determined by trial and error.
(6) The appropriate parameter value obtained in (5) above is set in the target machine.
(7) Check if the trouble has been improved on the target machine.

  In the above (3), it is necessary to analyze the function level of the application program and operating system. This analysis requires a high level of knowledge. That is, it is necessary to understand the characteristics of the function (the role of the function, the degree of influence of the function on the system, etc.), and it is very difficult to recognize the cause of the trouble.

  In many cases, the main body that uses the target machine is a different customer from the main body that performs parameter tuning (for example, a system vendor). For this reason, in (4) above, considering the security of the customer, it may be difficult (or limited) for the system vendor to obtain information on the environment of the target machine used by the customer. Many. In addition, even if some information about the environment is available, it is often difficult to construct the environment (it is very difficult to construct an exactly identical machine environment on a test machine). Also, it is often difficult to reproduce troubles such as response delays.

  In addition, it is very difficult to find out how the parameters are set so that trouble can be solved or improved, because it takes time to repeat trial and error.

  In the following description, a computer that is actually used for parameter tuning is referred to as a “target machine”, and a computer that includes the environment of the target machine is referred to as a “test machine”.

  There are various parameters relating to the computer. One example is a parameter called sched_migration_cost. This parameter is a parameter on the premise of multi-CPU or multi-core. When a task wakes up from a sleep state, the task scheduler attempts to wake up the task with the same CPU or core that was used during the previous execution of the task. This is because if the same CPU or core is used, there is a high probability that the data used by the task at the previous execution remains in the cache used by the CPU. A state in which the task can be woken up again by the CPU used last time is called “cache hot”. The parameter sched_migration_cost is a parameter for setting a maximum value (the longest time) that can be waited for a task to be waked up to become “cache hot”. As the value of the parameter sched_migration_cost is larger, cache misses due to migration of tasks between CPUs can be suppressed, the cache hit rate can be improved, and improvement in throughput performance can be expected. However, since the waiting time for the cache hot determination increases, the response decreases. Conversely, the smaller the value of sched_migration_cost, the more actively the task moves to the idle CPU, and therefore an improvement in response is expected. In this case, however, the number of cache misses increases, resulting in a decrease in throughput.

  This sched_migration_cost needs to be set to an appropriate value according to the number of tasks waiting for execution on the target machine on which the application software is operating and other operating environments. For this reason, establishment of the technique which can set such a parameter value simply according to environmental change is desired. In addition, a parameter adjustment technique that does not require complicated work such as embedding a command for parameter adjustment in an application program or the like is desired.

  Conventionally, a technique for creating an application and its application model is known. That is, a program source in which a log output command for application model adjustment is embedded, a corresponding performance simulation source, and an initial parameter value for application model adjustment are included in advance, and a history of parameter values after adjustment can be accumulated. By combining these software components, creating an application and application model, comparing the logs obtained by executing them, automatically adjusting the parameters of the application model according to the actual measurement result of the application, and the adjustment result Is reflected in the application model, and in addition to the history of the original software component, there is known a technique for changing the parameter initial value if necessary (Patent Document 1).

  There is also known a technique for automatically selecting a function that contributes more to performance evaluation based on a program operation history of a computer system. In other words, it accepts the operation history, assigns the function to the tree-structured node using the function call relationship included in the accepted operation history, treats the execution time or memory usage as quantitative information, and quantitative information associated with the function A technique is known in which a function is extracted under a condition in which the sum of the numbers exceeds the product of quantitative information associated with the whole and a predetermined ratio (Patent Document 2).

JP 2002-215423 A JP 2010-198133 A

  In one aspect, an object of the present invention is to enable easy tuning of computer parameters.

  According to an embodiment, there is provided a program for determining a setting parameter of the target machine using a test machine in which software that operates on the target machine is mounted, the test against a change in a value of the setting parameter of the test machine Measuring a machine behavior, generating a test model of the test machine from the change in the configuration parameter and the measured behavior, and the behavior of the target machine at a first value of the configuration parameter; The relationship between the behavior of the target machine and the behavior model is identified from the behavior of the behavior model at the first value of the setting parameter, and the target is determined from the behavior model and the relationship. Calculating a second value of the setting parameter for the machine, causing the computer to execute a process; Gram is provided.

  According to the embodiment, the computer parameters can be easily tuned.

It is a functional block diagram of the parameter tuning device in one embodiment of the present invention. It is a figure which shows the processing flow in one Embodiment of this invention. It is a figure which shows the profiling data of the test machine in one Embodiment. It is a figure which shows the dependence parameter with respect to the function in one Embodiment, and a function time prediction coefficient. It is a figure which shows the process which calculates the environmental correction coefficient in one Embodiment. It is a figure which shows the predicted value of the profiling data of the target machine in one Embodiment. It is a figure which shows the hardware constitutions of one Embodiment.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  FIG. 1 shows a functional block diagram of a parameter tuning apparatus according to an embodiment of the present invention. The target machine 102 is a physical machine or a virtual machine in which an application program is operating and provides a service of the application program. This target machine 102 is assumed to be operating with a specific parameter value (current parameter value 104).

  The test machine 112 is installed with an application that runs on the target machine 102. Furthermore, the operating environment such as an operating system, hardware configuration, peripheral devices, and necessary databases is ideally the same as that of the target machine 102. is there. As described above, in reality, it may not be possible to make the operating environment the same. For this reason, it is desirable to maintain the same environment as much as possible. The parameter tuning device 150 is a device whose main purpose is to output an appropriate parameter value for the target machine 102 based on the behavior of the test machine 112. The parameter tuning device 150 may be realized as a program that runs on a computer. This program may run on the test machine 112 or may run on another computer. Further, the parameter tuning apparatus may output an estimated execution time value 186 of the application program of the target machine 102.

  Further, the functional configuration of the parameter tuning device 150 will be described with reference to FIG. The parameter tuning device 150 preferably includes a behavior measuring unit 162, a behavior model generating unit 164, a relationship specifying unit 172, and a parameter value calculating unit 182. The behavior measuring unit 162 may change various parameters for the test machine 112 (163), execute CPU profiling each time, and acquire the CPU behavior 114 as profiling data. The CPU behavior 114 (profiling data) and parameter values are sent to the behavior model generation unit 164. The behavior model generation unit 164 models the behavior (CPU profiling data) of the test machine 112 with respect to parameter changes. As an example of this model, the function time prediction coefficient 166 may be generated. Details of the function time prediction coefficient 166 will be described later.

  This model is a model of the test machine 112. However, the test machine 112 is a machine that matches the application software of the target machine 102 and other environments as much as possible. Therefore, it should be noted that this model can be regarded as a model of the behavior of the target machine 102.

  Then, an estimated value 165 of the model behavior when the current parameter value 104 of the target machine 102 is applied to this model is calculated and sent to the relationship identifying unit 172. The relationship identification unit 172 executes CPU profiling of the target machine 102 and acquires the behavior 106 of the target machine 102, that is, profiling data of the target machine 102. The relationship identifying unit 172 identifies the relationship between the model behavior estimation value 165 and the behavior 106 of the target machine 102, that is, the profiling data of the target machine 102. As an example of this relationship, an environmental correction coefficient 174 is calculated. Details of the environment correction coefficient 174 will be described later. Note that the environment of the target machine 102 is constantly changing. For this reason, it should be noted that the role of the relationship specifying unit 172 described above can be regarded as including the role of bringing the model generated by the behavior model generation unit 164 closer to the environment where the current target machine 102 is placed. Should.

  The function time prediction coefficient 166 and the environment correction coefficient 174 are sent to the parameter value calculation unit 182. The parameter value calculation unit 182 preferably includes a parameter optimization unit 184. The parameter optimization unit calculates an appropriate parameter value 183 for the target machine 102 from the input function time prediction coefficient 166, the environment correction coefficient 174, and other information, and provides it to the target machine 102. Further, the parameter optimization unit may output an execution time predicted value of the application program in the target machine 102 when the appropriate parameter is applied to the target machine 102. Details of the method of calculating the appropriate parameter value 183 will be described later.

  FIG. 2 is a diagram showing a processing flow in one embodiment of the present invention. As a premise of this flow, an application program that operates on the target machine 102 is mounted on the test machine 112, and an environment such as an operating system, hardware, peripheral devices, and a database to be used is mounted. And As described above, the environment of the test machine 112 is preferably the same as the environment of the target machine 102, but there may be different parts.

  In step 202, the value of the setting parameter of the test machine 112 is changed, and the profiling data (behavior) of the test machine 112 is measured and acquired.

  In step 204, a behavior model (function time prediction coefficient) of the test machine 112 is generated. A function time prediction coefficient as an example of the behavior model will be described later.

  In step 206, profiling data (behavior) of the target machine 102 at the current parameter values is obtained.

  In step 208, profiling data of the behavior model of the test machine 112 is calculated using the current setting parameters of the target machine 102.

In step 210, the relationship (environment correction coefficient) between the acquired profiling data of the target machine 102 and the calculated profiling data estimation value of the behavior model of the test machine 112 is calculated. As described above, since the test machine 112 is mounted with the application of the target machine 102 and various environments are very similar to the target machine 102, the behavior model of the test machine 112 is the model of the target machine 102. It can be seen that it is a product. Therefore, it is estimated that the estimated profiling data calculated by applying the current setting parameter of the target machine 102 to the model approximates the profiling data acquired from the target machine 102. However, as already described, the behavior of the target machine 102 can be dynamically changed to some extent due to various factors. Therefore, in this step 210, in order to bring the model closer to the current behavior of the target machine 102, the estimated profiling data calculated by applying the current setting parameter of the target machine 102 to the model, and the target machine 102 Find the relationship with the acquired profiling data. As a specific example of this relationship, an environmental correction coefficient may be obtained. Details of the environment correction coefficient will be described later.

  In step 212, an appropriate parameter of the target machine 102 is calculated using information including the behavior model (function time prediction coefficient) of the test machine 112 and the relationship (environment correction coefficient). It may be understood that the current target machine 102 is modeled by information including the behavior model (function time prediction coefficient) of the test machine 112 and the relationship (environment correction coefficient). By using these pieces of information, it is possible to simulate the behavior (profiling data) of the target machine 102 when the parameters are changed. An appropriate parameter of the target machine 102 is calculated by this simulation. As an appropriate parameter calculation method, a parameter that minimizes the total execution time of each function acquired by profiling data may be determined using a solution to an optimization problem. Details of the parameter calculation will be described later.

  In step 214, the obtained appropriate parameter value may be used to calculate and output a prediction of the execution time of the application program in the target machine 102. By examining this prediction, it is possible to predict how much trouble such as response delay will be improved. Then, the obtained appropriate parameter value is applied to the target machine 102.

  In the above-described embodiment, the behavior model of the test machine 112 is generated, but the test machine 112 itself may be used instead of the behavior model.

  FIGS. 3A to 3D are diagrams showing profiling data of the test machine 112 according to an embodiment. More specifically, the content of the profiling data of the test machine 112 acquired by the behavior measuring unit 162 in FIG. 1 is shown. In addition, the contents of the process of step 202 in FIG. 2 are shown.

  FIG. 3A shows the profiling data of the functions X, Y, Z, and U when the values of the parameters A, B, and C in the test machine 112 are 10, 15, and 0, respectively.

  Profiling data is represented by the CPU execution time of each function. In this case, the profiling data is acquired with the execution times of the function X, the function Y, the function Z, and the function U being 200 ms, 400 ms, 60 ms, and 100 ms, respectively. The parameter values in FIG. 3A are assumed to be default values. The default parameters may be stored in a storage unit (memory 715 in FIG. 7) in the test machine 112.

  Hereinafter, the function execution time is referred to as function time.

  FIG. 3B shows profiling data in which only the parameter A is changed from 10 to 0 in the test machine 112. In this case, it can be seen that only the function Z changes from 60 ms to 160 ms.

  FIG. 3C shows profiling data in which the parameter B is changed from 15 to 30. In this case, it can be seen that the functions X and Y change from 200 ms to 320 ms and from 400 ms to 100 ms, respectively.

  FIG. 3D shows profiling data in which only the parameter C is changed from 0 to 10. In this case, it can be seen that the functions X and Z change from 200 ms to 250 ms and from 60 ms to 30 ms, respectively.

FIG. 4 shows dependent parameters and function time prediction coefficients for a function in one embodiment. That is, FIG. 4 summarizes the results shown in FIG. From FIG. 4, it is possible to generate a linear model that receives parameter values as inputs and outputs estimated values of profiling data related to each function. In FIG. 4, the function time prediction coefficient is the slope of the linear model. If the profiling data is (t _X , t _Y , t _Z , t _U ) ^T and the parameter values are (A, B, C) ^T , FIG. 4 can be expressed by the following equation (1). T means transposition of the matrix.

と表すことができる。式（１）は、テストマシン１１２の線形モデルである。すなわちパラメータ値（Ａ，Ｂ，Ｃ）^Ｔに、具体的な値を代入することにより、線形モデルに基づくテストマシン１１２のプロファイリングデータ（ｔ_Ｘ，ｔ_Ｙ，ｔ_Ｚ，ｔ_Ｕ）^Ｔの推定値が取得できる。

It can be expressed as. Equation (1) is a linear model of the test machine 112. That is, by substituting specific values for the parameter values (A, B, C) ^T , the estimated values of the profiling data (t _X , t _Y , t _Z , t _U ) ^T of the test machine 112 based on the linear model Can be obtained.

  3 and 4, the slope (function time prediction coefficient) of each linear model is obtained by sampling between two points, but a plurality of parameters are given to the test machine 112 to obtain a plurality of profiling data, The slope (function time prediction coefficient) may be obtained by the least square method or the like.

  In the above example, the linear model is used. However, the test machine 112 may be modeled using a nonlinear model such as a quadratic function or a cubic function. If the vector of profiling data is t and the parameter vector is α, the model f (*) of the test machine 112 that outputs the vector t of profiling data with the parameter vector α as an input is represented by the following general formula (2): Can be expressed as

  FIG. 5 shows a process for calculating the environmental correction coefficient in one embodiment. This process is a specific example of the process of the relationship specifying unit 172 in FIG. Moreover, it is a specific example of the process of step 210 in FIG. That is, the relationship between the profiling data of the target machine 102 and the calculated profiling data estimated value of the behavior model of the test machine 112 is calculated as the environmental correction coefficient.

In FIG. 5, 502 is profiling data at the current parameter value (A, B, C) ^T = (10, 20, 10) ^T in the target machine 102. Reference numeral 504 denotes an estimated value of the profiling data of the behavior model of the test machine 112 when the same parameter value as that of the target machine 102 is used. There is a case where there is a difference between both profiling data due to factors such as environmental changes. In order to absorb this difference and bring the behavior model of the test machine 112 closer to the behavior of the target machine 102, an environment correction coefficient 506 may be introduced. The calculation formula of the environmental correction coefficient 508 indicates the calculation method and the value of the environmental correction coefficient. That is, the ratio of the values of the corresponding functions of the profiling data is calculated. By multiplying the estimated value of the profiling data obtained from the behavior model of the test machine 112 by the environmental correction coefficient, the estimated value of the profiling data of the target machine 102 can be calculated more appropriately.

By using the equation (1) and the environment correction coefficient 508 in FIG. 5, the estimated values (t _{X ′} , t _{Y ′} , t _{Z ′} , t _{U ′} ) ^T of the profiling data of the target machine 102 are obtained from the parameter values. It can be represented by the following formula (3).

  Further, assuming that the environment correction coefficient vector is R, the predicted value t ′ of the profiling data vector of the target machine 102 is expressed by the following general formula (4) using the general formula (2).

  FIG. 6 shows the predicted value of the profiling data of the target machine 102 in one embodiment. Specifically, the total value of the profiling data when the parameter value P (A, B, C) is changed variously using the expression (3) is shown. For example, in order to improve the response delay trouble of the target machine 102, it may be effective to obtain a parameter value that minimizes the total value of predicted values of profiling data. In FIG. 6, the total value when the parameter value is P (10,15,0) is 1819, which is the minimum value. Therefore, the parameter value (10, 15, 0) may be selected as an appropriate value for the parameter of the target machine 102.

  It should be noted that as a general calculation formula for obtaining the minimum value of the total of profiling data, a parameter vector α = α ′ satisfying the following general formula (5) may be obtained using formula (4).

そして、式（５）によって表される最小値が、ターゲットマシン１０２のアプリケーションプログラムの実行時間予測値としても利用することができる。

Then, the minimum value represented by Expression (5) can be used as an execution time prediction value of the application program of the target machine 102.

  Expression (5) is an expression for obtaining the minimum value of the total value of the profiling data, but an evaluation function that minimizes the total value obtained by multiplying each of the profiling data may be used. Alternatively, focusing on only one or more specific profiling data, the parameter may be obtained so as to minimize the value.

  As a method for obtaining the parameters, it is desirable to predetermine the range that each parameter can take (parameter constraint conditions). Thereby, it is possible to avoid selecting an unrealistic parameter value.

  Then, each parameter may be changed by a predetermined increment, and a combination of the changed parameters may be comprehensively given in Expression (5) to search for a vector value of a parameter having a minimum total value. Also, starting from the parameter values found in this exhaustive search (or starting from any parameter value), search for the vector value of the parameter that can take the minimum value using linear programming or nonlinear optimization techniques. May be.

  The parameters in the present embodiment include various parameters on the computer, and are not limited to parameters for the CPU.

  FIG. 7 shows a hardware configuration of one embodiment. The target machine, test machine, and parameter tuning apparatus include a CPU 710, a memory 715, an input device 720, an output device 725, an external storage device 730, a portable recording medium drive device 735, and a network connection device 745. Each device is connected by a bus 750. Further, the portable recording medium driving device 735 can read and write the portable recording medium 740. The network connection device 745 may be connected to the Internet 760 and / or a dedicated line 761.

  Note that the program of this embodiment can be stored in the portable recording medium 740. The portable recording medium 740 refers to one or more non-transitory, tangible storage media having a structure. Illustrative examples of the portable recording medium 740 include a magnetic recording medium, an optical disk, a magneto-optical recording medium, and a nonvolatile memory. Magnetic recording media include HDDs, flexible disks (FD), magnetic tapes (MT) and the like. Optical disks include DVD (Digital Versatile Disc), DVD-RAM, CD-ROM (Compact Disc-Read Only Memory), CD-R (Recordable) / RW (ReWritable), and the like. Magneto-optical recording media include MO (Magneto-Optical disk).

  In the invention relating to the method and the program described in the claims, the order of the processes may be changed as long as there is no contradiction, or a plurality of processes may be performed simultaneously. It goes without saying that these embodiments are also included in the technical scope of the invention described in the claims.

The following notes are disclosed regarding the above embodiment.
(Appendix 1)
A program for determining a setting parameter of the target machine using a test machine in which software that operates on the target machine is installed,
Measuring the behavior of the test machine with respect to a change in the value of a setting parameter of the test machine;
From the change in the setting parameter and the measured behavior, a behavior model of the test machine is generated,
Between the behavior of the target machine at the first value of the setting parameter and the behavior of the behavior model at the first value of the setting parameter, between the behavior of the target machine and the behavior model. Identify the relationship
Calculating a second value of the configuration parameter for the target machine from the behavior model and the relationship;
A program that causes a computer to execute processing.
(Appendix 2)
The program according to claim 1, wherein the behavior model receives a setting parameter value and outputs an estimated value of the behavior of the test machine.
(Appendix 3)
The program according to appendix 1 or 2, wherein the behavior is an execution time of one or more functions included in the software.
(Appendix 4)
The program according to any one of appendices 1 to 3, wherein the setting parameter is a parameter for setting an operating environment of a CPU of a target machine or a CPU of the test machine.
(Appendix 5)
The process of calculating the second value includes a process of obtaining a value of the setting parameter of the target machine for minimizing a predicted value of time for executing software on the target machine. 4. The program according to any one of four.
(Appendix 6)
A method for determining a setting parameter of the target machine using a test machine equipped with software that operates on the target machine,
Measuring the behavior of the test machine with respect to a change in the value of a setting parameter of the test machine;
From the change in the setting parameter and the measured behavior, a behavior model of the test machine is generated,
Between the behavior of the target machine at the first value of the setting parameter and the behavior of the behavior model at the first value of the setting parameter, between the behavior of the target machine and the behavior model. Identify the relationship
Calculating a second value of the configuration parameter for the target machine from the behavior model and the relationship;
A method having processing.
(Appendix 7)
The method according to claim 6, wherein the behavior model receives a setting parameter value and outputs an estimated value of the behavior of the test machine.
(Appendix 8)
The method according to claim 6 or 7, wherein the behavior is an execution time of one or more functions included in the software.
(Appendix 9)
The method according to any one of appendices 6 to 8, wherein the setting parameter is a parameter for setting an operating environment of a CPU of a target machine or a CPU of the test machine.
(Appendix 10)
The process of calculating the second value includes a process of obtaining a value of the setting parameter of the target machine for minimizing a predicted value of time for executing the software on the target machine. The method according to any one of 9.
(Appendix 11)
A device that determines a setting parameter of the target machine using a test machine in which software that operates on the target machine is mounted,
A behavior measuring unit for measuring the behavior of the test machine with respect to a change in a value of a setting parameter of the test machine;
A behavior model generating unit that generates a behavior model of the test machine from the change in the setting parameter and the measured behavior;
Between the behavior of the target machine at the first value of the setting parameter and the behavior of the behavior model at the first value of the setting parameter, between the behavior of the target machine and the behavior model. A relationship identifying unit that identifies the relationship between
A parameter value calculation unit for calculating a second value of the setting parameter for the target machine from the behavior model and the relationship;
Having a device.
(Appendix 12)
The apparatus according to claim 11, wherein the behavior model receives a set parameter value and outputs an estimated value of the behavior of the test machine.
(Appendix 13)
The apparatus according to claim 11 or 12, wherein the behavior is an execution time of one or more functions included in the software.
(Appendix 14)
The apparatus according to any one of appendices 11 to 13, wherein the setting parameter is a parameter for setting an operating environment of a CPU of a target machine or a CPU of the test machine.
(Appendix 15)
The setting parameter value calculation unit includes a parameter optimization unit that calculates a value of the setting parameter of the target machine for minimizing a predicted value of time for executing software on the target machine. 14. The apparatus according to any one of fourteen.

102 target machine 112 test machine 150 parameter tuning device 162 behavior measuring unit 164 behavior model generating unit 172 relation specifying unit 182 parameter value calculating unit

Claims (7)

A program for determining a setting parameter of the target machine using a test machine in which software that operates on the target machine is installed,
Measuring the behavior of the test machine with respect to a change in the value of a setting parameter of the test machine;
From the change in the setting parameter and the measured behavior, a behavior model of the test machine is generated,
From the behavior of the target machine at the first value of the setting parameter and the estimated value of the behavior by the behavior model at the first value of the setting parameter, the behavior of the target machine and the behavior by the behavior model Identify the relationship between the estimate of
Calculating a second value of the configuration parameter for the target machine from the behavior model and the relationship;
A program that causes a computer to execute processing.   The program according to claim 1, wherein the behavior model receives a setting parameter value and outputs an estimated value of the behavior of the test machine.   The program according to claim 1, wherein the behavior is an execution time of one or more functions included in the software.   The program according to any one of claims 1 to 3, wherein the setting parameter is a parameter for setting an operating environment of a CPU of a target machine or a CPU of the test machine.   The process of calculating the second value includes a process of obtaining a value of the setting parameter of the target machine for minimizing a predicted value of time for executing software on the target machine. 5. The program according to any one of 4 to 4. A method for determining a setting parameter of the target machine using a test machine equipped with software that operates on the target machine,
Measuring the behavior of the test machine with respect to a change in the value of a setting parameter of the test machine;
From the change in the setting parameter and the measured behavior, a behavior model of the test machine is generated,
From the behavior of the target machine at the first value of the setting parameter and the estimated value of the behavior by the behavior model at the first value of the setting parameter, the behavior of the target machine and the behavior by the behavior model Identify the relationship between the estimate of
Calculating a second value of the configuration parameter for the target machine from the behavior model and the relationship;
A method having processing. A device that determines a setting parameter of the target machine using a test machine in which software that operates on the target machine is mounted,
A behavior measuring unit for measuring the behavior of the test machine with respect to a change in a value of a setting parameter of the test machine;
A behavior model generating unit that generates a behavior model of the test machine from the change in the setting parameter and the measured behavior;
From the behavior of the target machine at the first value of the setting parameter and the estimated value of the behavior by the behavior model at the first value of the setting parameter, the behavior of the target machine and the behavior by the behavior model A relationship identifying unit that identifies a relationship between the estimated value of
A parameter value calculation unit for calculating a second value of the setting parameter for the target machine from the behavior model and the relationship;
Having a device.

Images