JP5842704B2 - Estimation apparatus, program, and estimation method - Google Patents

Description

  The present invention relates to accurately estimating the execution time of a batch job.

  In a batch job network where batch jobs are connected to the network, in order to complete all batch jobs at the scheduled end time, it is necessary to predict the execution time of the batch job for the input data and set an appropriate start time. It has been.

  For example, it has been proposed to use a model that estimates execution time from the amount of input data. Create a model that estimates the output data amount and execution time from the input data amount of each past batch job, and estimate the output data amount and execution time from the current input data amount using the created model, and output data amount Is used as the amount of input data when estimating the execution time of subsequent batch jobs, and multiple models for obtaining batch job statistics are prepared. It has been proposed to determine from a measurement result using data, and to estimate a statistic at the time of processing large-scale data using the determined parameter.

  In addition, by extracting the program that reads the file or database that stores the input data items and performing an impact search on each program starting from the variable in the input record corresponding to each target item, the range of the input data item It is known to analyze the effect of changes in application maintenance on applications.

JP 2004-295731 A JP 2010-061417 A No. 2009-011057

Sturges, H.A., "The choice of a class interval", J. American Statistical Association, pp.65-66

  In the above prior art, the output data amount and the execution time are estimated only by the input data amount. In such an execution time estimation, in the case of a batch job that processes records of input data composed of a plurality of fields one by one, the output data amount and the execution time greatly vary depending on the distribution of input data values. There was a problem that the estimation error would increase.

Execution time estimation apparatus disclosed the value of the field of the input data that affects the output data for each class obtained by class classification, the frequency distribution table for setting the input power of classified said input data, each of該度number distribution table set the input frequency of the input data sampled in class, a storage unit for storing the frequency distribution map to obtain an output frequency of the output data by performing a predetermined process on the input data the sampling, stored in the storage unit For each class of the frequency distribution table , the setting unit for setting the input frequency of the target input data of the predetermined process, and a linear expression using the value of the frequency distribution map as a coefficient, The output frequency of the output data after the predetermined processing is acquired for each class, and the output frequency for the target input data is obtained by summing the acquired output frequencies for each class. Estimating a and an output power estimator.

  Therefore, an object of the present invention is to enable estimation of output data amount and execution time in a batch job with better accuracy.

  The disclosed execution time estimation device includes a frequency distribution table indicating a frequency distribution of input data for each class, a storage unit for storing a mapping from input data based on the frequency distribution table to result information by a predetermined process, and the storage An execution time for estimating the execution time of the predetermined process on the input data based on a calculation result using a mapping to the result information stored in the storage unit that receives the frequency distribution table stored in the storage unit And an estimation unit.

  Further, as means for solving the above problems, a program for causing a computer to function as the execution time estimation apparatus, an execution time estimation method, and a recording medium on which the program is recorded can be used.

  With the disclosed technology, it is possible to accurately estimate the execution time of a batch job even if the source information of the batch job is unknown.

It is a figure for demonstrating the execution time estimation method of the batch job which concerns on this Embodiment. It is FIG. (1) for demonstrating the frequency distribution map production | generation method. It is FIG. (2) for demonstrating the frequency distribution map production | generation method. It is FIG. (1) for demonstrating an execution time map production | generation method. It is FIG. (2) for demonstrating an execution time map production | generation method. It is a figure for demonstrating the output data frequency distribution estimation method. It is a figure for demonstrating the execution time estimation method. It is a figure for demonstrating the variable field determination method. It is a figure for demonstrating the influence calculation method. It is a figure which shows the function structural example of the system which concerns on this Embodiment. It is a figure which shows the hardware constitutions of a computer apparatus. It is a figure for demonstrating a frequency distribution map creation process. It is a figure for demonstrating real time mapping production processing. It is a figure for demonstrating an output data frequency distribution estimation process. It is a figure for demonstrating execution time estimation processing. It is a figure for demonstrating the variable field determination process. It is a figure for demonstrating the influence degree calculation process of output data. It is a figure for demonstrating an example of the batch process of temperature sensor data. It is a figure which shows the example of the schema of each data, and the estimated number of cases. It is FIG. (1) for demonstrating the example of determination of the variable field in 1st Example. It is FIG. (2) for demonstrating the example of determination of the variable field in 1st Example. It is a figure which shows the example of the three-dimensional frequency distribution in 1st Example. It is FIG. (1) for demonstrating the creation example of the frequency distribution map in 1st Example. It is FIG. (2) for demonstrating the creation example of the frequency distribution map in 1st Example. It is a figure for demonstrating the frequency distribution estimation example of the output data in 1st Example. It is FIG. (1) for demonstrating the creation example of the execution time map in 2nd Example. It is FIG. (2) for demonstrating the creation example of the execution time map in 2nd Example. It is a figure for demonstrating the example of estimation of the other execution time in 2nd Example. It is a figure for demonstrating an estimation error.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the present embodiment, for batch jobs that process records of input data composed of a plurality of fields one by one, the frequency distribution of output data using the frequency distribution mapping and the execution time mapping from the frequency distribution of the input data And the execution time. This embodiment makes it possible to estimate the output data amount and the execution time with high accuracy even when the source information of the batch job is unknown.

  FIG. 1 is a diagram for explaining a batch job execution time estimation method according to the present embodiment. In FIG. 1, in the batch job execution time estimation method, a field used as a variable (hereinafter referred to as a variable field) is determined for each of the frequency distribution 3 of the input data and the frequency distribution 4 of the output data, and the frequency distribution of the input data is determined. 3 and a frequency distribution table for each of the output data frequency distributions 4 are created (variable field determination method). The variable field determination method will be described later.

  The following is generated, and the frequency distribution 4 and the execution time 5 of the output data are estimated.

Generate the frequency distribution 3 of the input data The frequency distribution 3 of the input data is actually a multidimensional frequency distribution 3-2 composed of a plurality of variables, and the execution time estimation method of the batch job is such a multidimensional frequency. Handle distribution 3-2.

  Generate a frequency distribution map f.

    The frequency distribution map f indicates to which class of the frequency distribution 4 of the output data the records belonging to each class of the frequency distribution 3 of the input data are output.

  The execution time map g represents the average execution time of batch jobs per record for each class of the frequency distribution 3 of the input data.

  Next, a frequency distribution map generation method for generating the frequency distribution map f will be described. 2 and 3 are diagrams for explaining a frequency distribution map generation method. In FIG. 2, the number of records of input / output data in the schema 6 of known input / output data including information such as field names, types, and value ranges is estimated, and the method is performed by the batch job 2 (step S11). The estimated number of records is 1000, for example.

  The input / output data schema 6 is known and is information defining the data type of each item. Each data is defined by a field, a type, a value range, and the like. In the field, item names such as “sensor ID”, “date / time”, and “temperature” are shown. The data type of “sensor ID” is “int” and the value range [0,9999], the data type of “date and time” is “date”, the value range [0:00, 23:59], and the data type of “temperature” is “Float” and a range [0, 100] or the like.

  In the batch job 2, the variable field 7 is determined as a field having a certain degree of influence on the output data (step S12). It is assumed that the temperature field is determined as the variable field 7 of the determined input data.

  Based on the range of the input / output data and the estimated value of the number of records, the class number and class width of each frequency distribution are determined, and the frequency distribution table 3f is created (step S13). For example, for the temperature field determined as the variable field 7, the class width is 0 ° C. to 5 ° C. for the class (1), the class width is 5 ° C. to 10 ° C. for the class (2), and the class (3 ) For the class (4), the class width is 15 ° C. to 20 ° C., for the class (5), the class width is 20 ° C. to 25 ° C., and so on. The number of classes and class width are determined.

  Then, for the input data, a predetermined number (for example, 100) of random data is generated for each class of the frequency distribution table 3f (step S14). The frequency of the frequency distribution table 3f indicates the number of generated random data.

  In FIG. 3, the random input data generated for each class of the frequency distribution table 3f in step S14 is input to the batch job 2 (step S15). For each class, batch job 2 processes the records one by one, and counts how many input data of a predetermined number (100) of each class are output to which class of frequency distribution 4 of the output data (step S16). ). The counted value is recorded in the input / output correspondence table 2a.

After processing all input data, the value of the input / output correspondence table 2a is divided by the number of cases (100) for each class to obtain a frequency distribution map f (step S17).

Next, an execution time map generation method for generating the frequency distribution map g will be described. 4 and 5 are diagrams for explaining the execution time map generation method. In FIG. 4, the number of input / output data records in the schema 6 of known input / output data including information such as field names, types, and value ranges is estimated, and processing is performed by the batch job 2 (step S21). The estimated number of records is 1000, for example. The schema 6 of input / output data is the same as the schema shown in FIG.

  In the batch job 2, the variable field 7 is determined as a field having a certain degree of influence on the output data (step S22). It is assumed that the temperature field is determined as the variable field 7 of the determined input data.

  Based on the range of the input / output data and the estimated value of the number of records, the class number and class width of each frequency distribution are determined, and the frequency distribution table 3f is created (step S23). For example, for the temperature field determined as the variable field 7, the class width is 0 ° C. to 5 ° C. for the class (1), the class width is 5 ° C. to 10 ° C. for the class (2), and the class (3 ) For the class (4), the class width is 15 ° C. to 20 ° C., for the class (5), the class width is 20 ° C. to 25 ° C., and so on. The number of classes and class width are determined.

  Then, a predetermined number (for example, 100) of random data is generated for each class of the frequency distribution table 3f for the input data (step S24). The frequency of the frequency distribution table 3f indicates the number of generated random data.

  In FIG. 5, the random input data generated for each class of the frequency distribution table 3f in step S24 is input to the batch job 2 (step S25). For each class, batch job 2 processes records one by one, measures the execution time at that time, and corresponds to the total execution time of the input data of the predetermined number (100) of each class Record in Table 2b (step S26).

  After processing all input data, the value of the execution time correspondence table 2b is divided by the number of cases (100) for each class to obtain an execution time map g (step S27).

  Next, an output data frequency distribution estimation method for estimating the frequency distribution of output data will be described. FIG. 6 is a diagram for explaining an output data frequency distribution estimation method. In the output data frequency distribution estimation method shown in FIG. 6, the frequency distribution table 3f-2 of the actual input data 2d is created (step S31). The frequency distribution table 3f-2 is a table in which the number of cases is set to the frequency of the frequency distribution table 3f based on the actual input data 2d. Class (1) is “50” frequency, Class (2) is “92” frequency, Class (3) is “81” frequency, Class (4) is “73” frequency, and Class (5) is “42” frequency. Is set, and the frequency distribution table 3f-2 in which the total number of actual input data 2d is “338” is created.

  Then, the frequency of each class of the output data is calculated with respect to the frequency for each class in the frequency distribution table 3f by a linear expression using each value of the frequency distribution map f as a coefficient (step S32). The frequency distribution estimated value 4f of the output data estimated with respect to the frequency distribution table 3f-2 of the actual input data 2d is output. For frequency distribution table 3f, class (1) is "0" frequency, class (2) is "0" frequency, class (3) is "81" frequency, class (4) is "73" frequency, and class (5) obtains a frequency distribution estimated value 4f indicating "42" frequency. The output data amount predicted for the actual input data 2d is the total number “196”.

  Next, an execution time estimation method for estimating the execution time will be described. FIG. 7 is a diagram for explaining the execution time estimation method. In the execution time estimation method shown in FIG. 7, a frequency distribution table 3f-2 of actual input data 2d is created (step S41). Creation of the frequency distribution table 3f-2 is the same as step S31 in FIG.

Then, the execution time is calculated by a linear expression using each value of the execution time map g as a coefficient for the frequency in the frequency distribution table 3f-2 of the same class (step S42). From the frequency distribution table 3f-2 and the execution time map g shown in FIG.
Estimated execution time = 0.1 * 50 + 0.1 * 92 + 0.2 * 81
+ 0.2 * 73 + 0.2 * 42
And an execution time estimated value 5 (53.4 msec) is obtained based on the frequency distribution table 3f-2 of the actual input data 2d.

  Next, a variable field determination method will be described. FIG. 8 is a diagram for explaining a variable field determination method. In the variable field determination method shown in FIG. 8, two types of data sets A that can be used for any field and data sets B1 and B2 of input data for a certain field a are created.

  The data set A is a data set in which the number of records set for the estimated number of fields is set using normal random numbers having an average or standard deviation. The two types of data sets B1 and B2 are set to the same value as that of the data set A except for the field a for each record, and only the field a is a normal random number that is different from the data set A or has a standard deviation. It is a data set in which the estimated number of records is changed.

  The data set A and the data sets B1 and B2 are respectively processed by the batch job 2, the output data 2e based on the batch job 2 is compared, and how much they differ is quantified as the degree of influence on the output data 2e. A field whose degree indicates a predetermined value or more is determined as a variable field. The degree of influence on the output data 2e is calculated based on the number of cases, the number of fields having different values, and the like.

  In FIG. 8, of the output data 2e, the result after the processing of the batch job 2 for the data set A is indicated by the data set A-2, and the result after the processing of the batch job 2 for the data set B1 is the data set B1-2. The result after processing of batch job 2 for data set B2 is indicated by data set B2-2.

  The output data A-2 after processing of the batch job 2 of the data set A is a reference. The data set B1 is a data set in which the “sensor ID” field is changed, and the output data B1-2 after processing of the batch job 2 shows an example in which there is no change in values other than the “sensor ID” field. . The data set B2 is a data set in which the “temperature” field is changed, and the output data B2-2 after the processing of the batch job 2 shows an example in which the data amount of the data set A is changed (increased or decreased). . The data amount is indicated by the number of records.

  Next, an influence calculation method for calculating the influence α on the output data will be described. FIG. 9 is a diagram for explaining an influence degree calculation method. As shown in FIG. 9, in the influence calculation method, first, the number of records of output data is compared (Comparison I). When the number of records is different from the reference output data A-2 based on the judgment of the comparison I, “1” is set to the influence degree α on the output data.

  When the number of records of the output data A-2 of the data set A is 300, whereas the number of records of the output data B2-2 of the data set B2 in which the “temperature” field is changed is 200, The change in the “temperature” field is determined to have an influence degree α = 1 on the output data. If the number of records of the output data B2-2 is less than or greater than the number of records of the output data A-2 serving as a reference, the degree of influence α is “1”.

  If the number of records in the output data is equal as a result of the comparison (I), fields other than the changed target field are compared among all records in the output data (Comparison II). When it is determined by comparison (I) that the number of records in the output data is equal to the number of records in the reference output data A-2, fields other than the target field are compared among all records in the output data.

  The influence degree α is calculated by the following formula.

α = (number of fields with different values) / (number of fields in all records)
In the output data B1-2 of the data set B1 in which the “sensor ID” field is changed, the number of fields of all records is to multiply the number of records “300” by the number of items “4” excluding “sensor ID”. Obtained by. The number of fields in all records is 300 * 4 = 1200. In the output data B1-2, since there is no change in the reference output data A-2 except for the changed “sensor ID” field, the number of fields having different values is “0”. Therefore, the influence degree α = 0/1200 = 0 is obtained.

  In the case of different output data C with 480 different fields, the number of fields in all records is 300 * 4 = 1200. Since the number of fields having different values is “480”, the degree of influence α = 480/1200 = 0.4 is obtained.

  Both the output data B1-2 and the output data C have “300” records, but the influence α is different.

Hereinafter, a system 1000 that executes a batch job execution time estimation method will be described. FIG. 10 is a diagram illustrating a functional configuration example of the system according to the present embodiment. In FIG. 10, the system 1000 mainly includes a frequency distribution generation unit 40 and an execution time estimation unit 50. The degree number distribution generating unit 40, an execution time estimation unit 50 is mounted on a separate computer device, and the respective frequency distribution generating unit may be execution time estimation. Alternatively, the frequency distribution generation unit 40 and the execution time estimation unit 50 may be mounted on the same computer device. The frequency distribution generation unit 40 and the execution time estimation unit 50 are realized by the CPU 11 described later executing a corresponding program.

  The frequency distribution generation unit 40 generates a frequency distribution table 3f for the variable field 7 that affects the output data by processing the input data to the batch job 2, and outputs the output data amount or / and the execution time based on the frequency distribution table 3f. A processing unit that creates a frequency distribution map f or / and an execution time map g to be estimated. A variable field determination unit 41, a frequency distribution table generation unit 42, a frequency distribution mapping generation unit 43, and an execution time mapping generation unit 44. And have.

  The variable field determination unit 41 is a processing unit that determines a variable field based on the degree of influence α on the output data after the batch job. When the input field is generated based on the input / output data schema 6, the variable field determination unit 41 creates a data set in which a certain field is changed, and calculates the degree of influence α on the output data after the batch job. The variable field determination unit 41 determines a variable field based on the calculated influence degree α.

  The frequency distribution table generation unit 42 refers to the schema 6 of the input / output data, and based on the range of the variable field 7 determined by the variable field determination unit 41 and the estimated value of the number of records, This is a processing unit that determines the class and class width and creates the frequency distribution table 3f.

  The frequency distribution map creation unit 43 is a processing unit that creates a frequency distribution map f by performing batch job 2 on random input data 2d based on the frequency distribution table 3f. The execution time map creation unit 44 is a processing unit that creates an execution time map g by performing batch job 2 on random input data 2d based on the frequency distribution table 3f.

  The execution time estimation unit 50 is a processing unit that estimates the execution time using the frequency distribution table 3f-2 of the actual input data and the frequency distribution map f or / and the execution time map g, and creates an input data frequency distribution. Unit 51, execution time estimation unit 52, and output data frequency distribution estimation unit 53.

The input data frequency distribution creation unit 51 is a processing unit that creates the frequency distribution table 3f-2 by setting the frequency in the frequency distribution table 3f with the actual number of input data.

  The execution time estimation unit 52 is a processing unit that estimates the execution time based on the frequency distribution table 3f-2 created by setting the frequency in the frequency distribution table 3f by the input data frequency distribution creation unit 51. The execution time estimation unit 52 calculates the execution time estimation value 5 based on the sum of frequencies in the frequency distribution table 3f-2. Alternatively, the execution time estimation unit 52 calculates the execution time estimated value 5 by a linear expression using each value of the execution time map g as a coefficient.

  Based on the frequency distribution table 3f-2 created by the input data frequency distribution creation unit 51 using the frequency distribution table 3f, the output data frequency distribution estimation unit 53 is a linear expression using each value of the frequency distribution map f as a coefficient. Is a processing unit that creates the frequency distribution estimated value 4f of the output data by calculating the frequency of each stage of the output data.

  A hardware configuration of the computer apparatus 10 that implements the frequency distribution generation unit 40 and the execution time estimation unit 50 will be described. FIG. 11 is a diagram illustrating a hardware configuration of the computer apparatus. In FIG. 11, a computer device 10 is a terminal controlled by a computer, and includes a CPU (Central Processing Unit) 11, a main storage device 12, an auxiliary storage device 13, an input device 14, a display device 15, The output device 16, a communication I / F (interface) 17, and a drive 18 are connected to the bus B.

  The CPU 11 controls the computer device 10 according to a program stored in the main storage device 12. The main storage device 12 uses RAM (Random Access Memory), ROM (Read-Only Memory), etc., and is obtained by a program executed by the CPU 11, data necessary for processing by the CPU 11, and processing by the CPU 11. Stored data and the like are stored. A part of the main storage device 12 is allocated as a work area used for processing by the CPU 11.

  The auxiliary storage device 13 uses a hard disk drive and stores data such as programs for executing various processes. A part of the program stored in the auxiliary storage device 13 is loaded into the main storage device 12 and executed by the CPU 11, whereby various processes are realized. The storage unit 130 includes the main storage device 12 and / or the auxiliary storage device 13.

  The input device 14 includes a mouse, a keyboard, and the like, and is used for a user to input various information necessary for processing by the computer device 10. The display device 15 displays various information required under the control of the CPU 11. The output device 16 has a printer or the like and is used for outputting various types of information in accordance with instructions from the user. The communication I / F 17 is a device that is connected to, for example, the Internet, a LAN (Local Area Network), etc., and controls communication with an external device.

  A program that realizes processing performed by the computer apparatus 10 is provided to the computer apparatus 10 by a storage medium 19 such as a CD-ROM (Compact Disc Read-Only Memory). That is, when the storage medium 19 storing the program is set in the drive 18, the drive 18 reads the program from the storage medium 19, and the read program is installed in the auxiliary storage device 13 via the bus B. . When the program is activated, the CPU 11 starts its processing according to the program installed in the auxiliary storage device 13. The medium for storing the program is not limited to a CD-ROM, and any medium that can be read by a computer may be used. As a computer-readable storage medium, in addition to a CD-ROM, a portable recording medium such as a DVD disk or a USB memory, or a semiconductor memory such as a flash memory may be used.

  Further, a program that realizes processing performed by the computer device 10 may be provided from an external device via the communication I / F 17. Alternatively, the program may be provided to an external device, and each process described below may be realized by the external device. Communication by the communication I / F 17 is not limited to wireless or wired.

  When the computer device 10 is a device that implements the frequency distribution generation unit 40, the input / output data schema 6 and the like are stored in the storage unit 130. When the computer device 10 is a device that implements the execution time estimation unit 50, the storage unit 130 includes the input data 2d, the frequency distribution map f, the frequency distribution table 3f, and the frequency distribution data 34f including the frequency distribution estimation value 4f, the execution time. The mapping g and the like are stored.

  Further, when the frequency distribution generation unit 40 and the execution time estimation unit 50 are realized by one computer apparatus 10, the computer apparatus 10 corresponds to the entire system 1000.

  The frequency distribution map creating process by the frequency distribution map creating unit 43 will be described. FIG. 12 is a diagram for explaining the frequency distribution mapping process. In the frequency distribution mapping process shown in FIG. 12, the CPU 11 acquires the schema of input data from the schema 6 of input / output data (step S101). The input data schema selected by the user is read from the input / output data schema 6.

  The CPU 11 executes a variable field determination process for determining a variable field of the input data (step S102). Thereafter, the CPU 11 executes a frequency distribution table generation process for creating the frequency distribution table 3f of the input data (step S103). The number of classes and the class width are determined by the frequency distribution table generation process.

  The CPU 11 generates random input data for each class of the frequency distribution 3 of the input data (step S104), processes it one by one in the batch job 2, and outputs to which class of the frequency distribution 4 of the output data. Is counted (step S105).

  After completing the processing in batch job 2 for all generated input data, the CPU 11 divides the counted value for each class by the total number of input data (frequency for each class) to create a frequency distribution map g. (Step S106). Then, the CPU 11 ends the frequency distribution mapping process.

  The real-time map creation process by the real-time map creation unit 43 will be described with reference to FIG. FIG. 13 is a diagram for explaining real-time mapping creation processing. In the real-time mapping creation process shown in FIG. 13, the CPU 11 acquires the schema of input data from the schema 6 of input / output data (step S111). The input data schema selected by the user is read from the input / output data schema 6.

  The CPU 11 executes a variable field determination process for determining a variable field of the input data (step S112). After determining the variable field, the CPU 11 executes a frequency distribution table generation process for generating the frequency distribution table 3f of the input data (step S113). The number of classes and the class width are determined by the frequency distribution table generation process.

  The CPU 11 generates random input data for each class of the frequency distribution 3 of the input data (step S114), processes it one by one in a batch job, and measures the execution time for each item (step S115).

  Then, the CPU 11 creates an execution time map g from the average execution time for each class (step S116).

  The output data frequency distribution estimation processing by the output data frequency distribution estimation unit 53 will be described with reference to FIG. FIG. 14 is a diagram for explaining the output data frequency distribution estimation process. The output data frequency distribution estimation processing by the output data frequency distribution estimation unit 53 differs between the case where the batch job 2 is a single unit and the case where the batch job 2 is multistage.

  FIG. 14A describes output data frequency distribution estimation processing for a batch job 2 when the batch job 2 is a single unit. 14A, the CPU 11 reads the frequency distribution table 3f-2 of the actual input data 2d and the frequency distribution map f of the batch job 2 from the storage unit 130 (step S131), and the frequency distribution map f. The frequency of each class of the output data is calculated by a linear expression with each value as a coefficient (step S132).

  Then, the CPU 11 generates the frequency distribution estimated value 4f of the output data for the calculated frequency of each class (step S133), and ends the process when the batch job 2 is a single unit.

  In FIG. 14B, processing when the batch job 2 has multiple stages will be described. In FIG. 14B, the CPU 11 generates a frequency distribution table 3f of input data (step S141).

  The CPU 11 executes output data frequency distribution estimation processing for the batch job (step S142). After generating the output data frequency distribution estimated value 4f in the output data frequency distribution estimation process of the batch job, the CPU 11 determines whether or not a subsequent batch job exists (step S143).

  When determining that there is a subsequent batch job, the CPU 11 sets the frequency distribution estimated value 4f of the output data of the batch job in the frequency distribution table 3f of the input data (step S144), returns to step S142, and the same as described above. Repeat the process. On the other hand, if it is determined that there is no subsequent batch job, the CPU 11 ends this process.

  The execution time estimation process by the execution time estimation unit 52 will be described with reference to FIG. FIG. 15 is a diagram for explaining the execution time estimation process. The execution time estimation processing by the execution time estimation unit 52 differs between the case where the batch job 2 is a single unit and the case where the batch job 2 is multistage.

  FIG. 15A describes the execution time estimation process of the batch job 2 when the batch job 2 is a single unit. In FIG. 15A, the CPU 11 reads the frequency distribution table 3f-2 of the actual input data 2d and the frequency distribution map f of the batch job 2 from the storage unit 130 (step S151), and executes the execution time map g. The execution time is calculated by a linear expression with each value as a coefficient (step S152). Then, the CPU 11 ends this process.

  In FIG. 15B, processing when the batch job 2 has multiple stages will be described. In FIG. 15B, the CPU 11 generates a frequency distribution table 3f of input data (step S161).

  The CPU 11 executes the execution time estimation process for the batch job (step S162). After calculating the execution time in the execution time estimation process of the batch job, the CPU 11 determines whether there is a subsequent batch job (step S163).

  When determining that there is a subsequent batch job, the CPU 11 adds the execution time of the batch job to the total execution time of the batch job network (step S164), returns to step S162, and repeats the same processing as described above. On the other hand, if it is determined that there is no subsequent batch job, the CPU 11 ends this process.

  The variable field determination process in step S102 of FIG. 12 and step S112 of FIG. 13 will be described with reference to FIG. FIG. 16 is a diagram for explaining the variable field determination process. In the variable field determination process shown in FIG. 16, the CPU 11 determines whether it is the terminal data (step S171). If it is determined that the data is the end data, the CPU 11 ends this process. On the other hand, if it is determined that the data is not the terminal data, a data set A is generated using normal random numbers having an average or standard deviation for each field of the input data 2d (step S172).

  Then, the CPU 11 sets each record to the same value as the first field of the data set A except for the field a, and changes only the field a using a normal random number having an average or standard deviation different from that of the data set A. Data set B is generated (step S173).

  The CPU 11 processes the data sets A and B by the batch job 2 and obtains the data sets C and D as outputs (step S174). The CPU 11 calculates the influence of the output data using the data sets C and D (step S175).

  The CPU 11 determines whether or not the influence level α of the output data is greater than or equal to a certain level (step S176). If not, the CPU 11 proceeds to step S178. On the other hand, if the value is above a certain level, the CPU 11 determines the field a as a variable field (step S177), and changes the field a to the variable field of the input data 2d for the preceding batch job (step S178). Thereafter, the CPU 11 ends this process.

  The output data influence degree calculation processing in step S175 of FIG. 16 will be described with reference to FIG. FIG. 17 is a diagram for explaining the influence calculation processing of output data. In the output data influence calculation process shown in FIG. 17, the CPU 11 compares the data sets C and D (step S181), and determines whether the number of records in the data set C and the data set D is different (step S181). S182). This comparison process corresponds to the comparison I in FIG.

  If the number of records matches, the CPU 11 sets “1” as the influence α of the output data (step S183), and ends this process.

  On the other hand, if the number of records is different, the CPU 11 further determines whether there is a field with a different value other than the field a whose value has been changed (step S184). If there is no field with different values, the CPU 11 calculates the influence α of the output data by dividing the number of fields with different values between the data sets C and D by the number of fields of all records (step S185). The process ends. On the other hand, if there are fields with different values, the CPU 11 ends this process.

Hereinafter, a case where the present embodiment is applied to batch processing of temperature sensor data will be described with reference to FIG. FIG. 18 is a diagram for explaining an example of batch processing of temperature sensor data. In the example of batch processing of temperature sensor data shown in FIG. 18, the average temperature of the room A is obtained based on data of the temperature sensors 8 arranged in the building (hereinafter referred to as temperature sensor data). The entire building is 25 × 25 (m ² ). The temperature sensor data is transmitted every 0.1 ° C., and the transmission interval is irregular.

  The temperature sensor data is collected at one place and stored in the temperature sensor data D1 of the entire building. The field items of the temperature sensor data D1 are “sensor ID”, “date / time”, “temperature”, “x”, “y”, and the like.

  The batch job J1 extracts the temperature sensor data D2 of the room A. For example, when the position information of the room A is represented by coordinates (8, 8) and coordinates (18, 16) which are the vertices on the diagonal line of the room A in the x and y directions shown in FIG. Among the sensors identified by (1), a record indicating the interior of the room A is extracted. The field items of the temperature sensor data D2 are the same as the field items of the temperature sensor data D1.

  The batch job J2 calculates an average temperature for each position based on the temperature sensor data D2 of the room A extracted by the batch job J1. Average temperature data D3 at each position indicated by xy is output. The items of the field of the average temperature data D3 are “x”, “y”, “average temperature”, and the like.

  FIG. 19 is a diagram illustrating an example of the schema of each data and the estimated number of cases. In FIG. 19, the schema 6-1 of the temperature sensor data D1 defines each data by field, type, value range, and the like. For each of the “sensor ID”, “date / time”, “temperature”, “x”, and “y” fields, a type, a value range, and the like are defined.

  The schema 6-2 of the temperature sensor data D2 defines each data by field, type, value range, and the like. For each of the “sensor ID”, “date / time”, “temperature”, “x”, and “y” fields, a type, a value range, and the like are defined.

  The schema 6-3 of the average temperature data D3 defines each data with fields, types, value ranges, and the like. For each of the “x”, “y”, and “average temperature” fields, a type, a value range, and the like are defined.

  Assume that the estimated number of cases of each data D1, D2, and D3 is 1000 cases.

  Hereinafter, a description will be given of a first embodiment in which the execution time is estimated using only the frequency distribution map f based on the schemas 6-1 to 6-3 and the estimated number of cases.

The procedure in the first embodiment is as follows.
(1) The frequency distribution estimated value 4f of the output data (temperature difference sensor data D2) is calculated from the frequency distribution table 3f-2 of the input data (temperature difference sensor data D1) using the frequency distribution map f, and the frequency distribution estimated value is calculated. The output data amount of batch job J1 (data amount of temperature difference sensor data D2) is obtained from 4f.
(2) The execution time is estimated from the output data amount (data amount of the temperature difference sensor data D2) using the method disclosed in Patent Document 1, and the execution time estimated value 5 is obtained. The output data amount (data amount of the temperature difference sensor data D2) can be easily obtained by the sum of the frequencies of the frequency distribution estimated value 4f.
(3) The frequency distribution estimated value 4f of the output data (temperature difference sensor data D2) is used as the frequency distribution table 3f of the input data of the subsequent batch job J2. As described above, the execution time of the subsequent batch job J2 is estimated using the method disclosed in Patent Document 1 from the output data amount (the output data amount of the average temperature data D3 at each position).

  An example of determining the variable field in the first embodiment will be described with reference to FIG. 20 and 21 are diagrams for explaining an example of determining a variable field in the first embodiment. 20, in temperature sensor data D1, reference temperature sensor data D1-0, temperature sensor data D1-1 in which only “sensor ID” is changed with respect to temperature sensor data D1-0, and temperature sensor data 1000 pieces of temperature sensor data D1-2 in which only “x” is changed with respect to D1-0 are prepared.

  The temperature sensor data D1-1 in which only the “sensor ID” is changed and the temperature sensor data D1-2 in which only “x” is changed with respect to the temperature sensor data D1-0 will be described below as an example. Temperature sensor data D1 is prepared in which only fields other than “ID” and “x” are changed.

  By processing the reference temperature sensor data D1-0 with the batch job J1, the temperature sensor data D2-0 of the room A is extracted from the temperature sensor data D1-0 and used as the reference data after extraction.

  By processing the temperature sensor data D1-1 in which only the “sensor ID” is changed by the batch job J1, the temperature sensor data D2-1 of the room A is extracted from the temperature sensor data D1-1, and other than “sensor ID”. Is compared with the temperature sensor data D2-0. In this example, since all values other than “sensor ID” are the same, “0” is set to the influence degree α (sensor ID).

  By processing the temperature sensor data D1-2 in which only “x” is changed by the batch job J1, the temperature sensor data D2-2 of the room A is extracted from the temperature sensor data D1-2, and a value other than “x”. Is compared with the temperature sensor data D2-0. In this example, since all values other than “x” are the same, “0” is set to the influence degree α (x).

  As described above, the degree of influence α is also calculated for the temperature sensor data D1 in which only the fields other than “sensor ID” and “x” are changed.

  Furthermore, in FIG. 21, a field a in which the influence degree α is a certain value (for example, 0.3) or more is defined as a variable field. From the influence α based on the batch job J1, “x” and “y” are set in the variable field of the temperature sensor data D1.

  Further, “temperature”, “x”, and “y” are set in the variable field of the temperature sensor data D2 from the influence α based on the subsequent batch job J2. Here, since the temperature sensor data D1 and the temperature sensor data D2 are data sets in which the field items match, the condition is that the variable field of D1 includes the variable field of D2. Therefore, “temperature” which is a difference from the variable field of D2 is added to the variable field of D1.

  Then, “x”, “y”, and “average temperature” are set in the variable field of the average temperature data D3 that is the terminal data.

  Next, an example of creating the frequency distribution table 3f of the input data D1 will be described. First, the number of classes and the class width are determined using the Sturges formula (Non-Patent Document 1). By setting the estimated number of cases to 1000, the number k of the variable field “temperature” of the temperature sensor data D1 is obtained by the calculation of Equation 1.

  Further, by calculation of Equation 2, approximately 9.09 is obtained as the class width h of the variable field “temperature”.

  For each of the other variable fields “x” and “y”, the class number k and the class width h are calculated as described above.

  The frequency distribution 3f of the temperature sensor data D1 is represented by a three-dimensional frequency distribution 38 as shown in FIG. FIG. 22 is a diagram illustrating an example of a three-dimensional frequency distribution. In FIG. 22, the three-dimensional frequency distribution 38 is represented by the dimensions of a “temperature” class 38a, an “x” class 38b, and a “y” class 38c.

  Each class of the variable field “temperature” is divided by a class width h = 9.09 to show a data example such as a class 38a of “temperature”. Each class of “x” is divided by a class width h = 4.54 to show an example of data such as a class 38b of “x”. Each class of “y” is divided by a class width h = 4.54, and shows a data example such as class “c” of “y”.

Next, an example of creating the frequency distribution map f in the first embodiment will be described. 23 and 24 are diagrams for explaining an example of creating a frequency distribution map in the first embodiment. FIG. 23A shows a frequency distribution table 3f based on the three-dimensional frequency distribution 38 of FIG. In the frequency distribution table 3f, the frequency “100” is set for each combination of the variable fields “temperature”, “x”, and “y”.

  The sample data 39 corresponding to the frequency (100 pieces) indicated for each combination is generated. For example, for a combination of the “temperature” “0-9.09” class, the “x” “0-4.54” class, and the “y” “0-4.54” class (1002, 12:20). , 2.1, 3, 2), etc. are generated.

  Then, the sample data 39 is processed one by one by the batch job J1, and the value counted in the input / output correspondence table 2a to which class of the estimated value 4f of the output data is output is recorded.

  As illustrated in FIG. 23B, for convenience, class names of A, B, C, D, and E are given to some of the class combinations. FIG. 23C shows an example of the input / output correspondence table 2a using the class name of FIG. 23B in the case of the batch job J1.

  In the input correspondence table 2a including a matrix of class names A to E showing the correspondence between the temperature sensor data D1 which is input data to the batch job J1 and the temperature sensor data D2 after the batch job J1, input (D1) Is output 38 times from the class B to the class B of the output (D2), is output 100 times from the class C of the input (D1) to the class C of the output (D2), and is output from the class B of the input (D1) ( It is output 64 times to class B of D2).

  A frequency distribution map f as shown in FIG. 24 is obtained from the input correspondence table 2a shown in FIG. FIG. 24 is a diagram illustrating an example of a frequency distribution map in the first embodiment. In FIG. 24, the frequency distribution map f of the batch job J1 is obtained by dividing each value of the input correspondence table 2a shown in FIG. 23C by the frequency for each class (100 cases).

  Next, a frequency distribution estimation example of output data in the first embodiment will be described. FIG. 25 is a diagram for explaining an example of estimating the frequency distribution of output data in the first embodiment. In FIG. 25A, a frequency distribution table 3f-2 of actual input data D1 is created. The created frequency distribution table 3f-2 includes, for example, the frequency of class A is “50”, the frequency of class B is “92”, the frequency of class C is “81”, the frequency of class D is “73”, and the class The frequency of E indicates “42” or the like.

  Then, the frequency of each class of the frequency distribution of the output data is estimated by a linear expression using each value of the frequency distribution map f as a coefficient. FIG. 25B shows an example of creating the frequency distribution estimated value 4f of the output data D2 using the frequency distribution map f shown in FIG.

The frequency of the class B of the data D2 is expressed as D2 [B],
D2 [B] = ... + 0 * D1 [A] + 0.38 * D1 [B] + 0 * D1 [C]
+ 0 * D1 [D] + 0 * D1 [D] +.
Is required.

  By performing the same calculation for the other classes, the frequency distribution estimated value 4f can be obtained.

  Then, the output data amount is calculated using the obtained frequency distribution estimated value 4f.

Output data amount = ... + D2 [A] + D2 [B] + D2 [C] + D2 [D]
+ D2 [E] + ...
It is obtained as follows. Based on the output data amount of the obtained temperature difference sensor data D2, the estimated execution time 5 of the batch job J1 is calculated using the method of Patent Document 1.

Further, the frequency distribution estimated value 4f is set in the frequency distribution table 3f-2 of the input data of the subsequent batch job J2, and the same processing as described above is performed, so that the output data amount of the average temperature data D3 at each position is used to determine the subsequent batch. it is possible to calculate the execution time estimate value 5 job J2.

  The second embodiment for estimating the execution time using the frequency distribution map f and the execution time map g will be described below.

The procedure in the second embodiment is as follows.
(1) The frequency distribution estimated value 4f of the output data (temperature difference sensor data D2) is calculated from the frequency distribution table 3f-2 of the input data (temperature difference sensor data D1) using the frequency distribution map f.
(2) The execution time is estimated from the frequency distribution table 3f-2 of the input data (temperature difference sensor data D1) using the execution time map g, and the execution time estimated value 5 is obtained.
(3) The frequency distribution estimated value 4f of the output data (data amount of the temperature difference sensor data D2) is used as the frequency distribution table 3f of the input data of the subsequent batch job J2. The execution time of the subsequent batch job J2 is estimated using the execution time map g as described above.

  Since the determination of the variable field in the second embodiment is as described in the first embodiment, its detailed description is omitted.

An example of creating the execution time map g in the second embodiment will be described. 26 and 27 are diagrams for explaining an example of creating an execution time map in the second embodiment. FIG. 26A shows a frequency distribution table 3f based on the three-dimensional frequency distribution 38 of FIG. The frequency distribution table f based on the three-dimensional frequency distribution 38 of FIG. 22 is used, which is the same as when the frequency distribution map 3f is created. In the frequency distribution table 3f, the frequency “100” is set for each combination of the variable fields “temperature”, “x”, and “y”.

Similarly to the creation of the frequency distribution map f , sample data 39 corresponding to the frequency (100) indicated for each combination is generated. For example, for a combination of the “temperature” “0-9.09” class, the “x” “0-4.54” class, and the “y” “0-4.54” class (1002, 12:20). , 2.1, 3, 2), etc. are generated.

  Then, the sample data 39 is processed one by one with the batch job J1, and the value obtained by counting the execution time 5 in the execution time correspondence table 2b is recorded.

  As illustrated in FIG. 26B, for convenience, class names of A, B, C, D, and E are given to some of the class combinations. FIG. 26C shows an example of the execution time correspondence table 2b using the class names of FIG. 26B in the case of the batch job J1. The execution time correspondence table 2b shows 100 total execution times for each class.

  An execution time map g as shown in FIG. 27 is obtained from the execution time correspondence table 2a shown in FIG. FIG. 26D illustrates an execution time map g of the batch job J1. In FIG. 26D, the execution time map g of the batch job J1 is obtained by dividing each value of the execution time correspondence table 2b shown in FIG. 26C by the frequency for each class (100 cases).

  FIG. 27 is a diagram for explaining an example of estimating the execution time in the second embodiment. FIG. 27A shows an example of creating the frequency distribution table 3f-2 of actual input data D1. The frequencies “50”, “92”, “81”, “73”, and “42” are shown for each of the classes A, B, C, D, and E which are part of the frequency distribution table 3f-2. It is an example.

  Then, the execution time estimated value 5 is calculated using the execution time map g illustrated in FIG. The execution time map g shown in FIG. 27 (B) corresponds to the execution time map g (FIG. 26 (D)) of the batch job J1. An execution time estimated value 5 is calculated by a linear expression using each value of the execution time map g as a coefficient. For example, the execution time t [J1] of the batch job J1 is calculated as follows.

t [J1] =... + 0.1 * D1 [A] + 0.3 * D1 [B]
+ 0.3 * D1 [C] + 0.3 * D1 [D] + 0.1 * D1 [E]
= 0.1 * (1125-92-81-73)
+ 0.3 * (92 + 81 + 73)
= 161.7
However, the actual number of records of the temperature sensor data D1 is 1125. The average execution time of classes other than B, C, and D is all 0.1 ms.

  A case where the number of records N of the input data is the same but the estimated time is different due to different frequency distributions will be described with reference to FIGS.

  FIG. 28 is a diagram for explaining another estimation example of the execution time in the second embodiment. In FIG. 28, it is assumed that the record number N of the input data D1 and D1 ′ is 1125.

  In the frequency distribution table 3 f-2 of the input data D 1, a part of the frequency distribution table 3 f-2 indicates “50” frequency, the class B indicates “92” frequency, the class C indicates “81” frequency, and the class D indicates “ 73 ”frequency, and class E indicates“ 42 ”frequency.

  In the frequency distribution table 3f-2 ′ of the input data D1 ′, in a part thereof, the class A indicates “20” frequency, the class B indicates “15” frequency, the class C indicates “24” frequency, and the class D Indicates “30” frequency, and class E indicates “25” frequency.

  In addition, the execution time for each class indicated by the execution time map g created in advance shows “0.1” msec in class A, “0.3” msec in class B, and “C” in class C. “0.3” msec is shown, “0.3” msec is shown in class D, and “0.1” msec is shown in class E.

In such a case, when the linear expression for estimating the execution time by the conventional method not using the frequency distribution table 3f-2 of the input data and the execution time mapping is t [J1] = 0.1N + 50,
t [J1] = 0.1 * 1125 + 50 = 162.5
As required.

On the other hand, the estimated execution time 5 for the input data D1 and D1 ′ according to the present embodiment is
For input data D1,
t [J1] = 0.1 * (1125-92-81-73)
+ 0.3 * (92 + 81 + 73) = 161.7
It becomes.

In the case of input data D1 ′,
t [J1] = 0.1 * (1125-15-24-30)
+ 0.3 * (15 + 24 + 30) = 126.3
It becomes.

  As described above, in the present embodiment, different execution time estimation values 5 are obtained for the input data D1 and the input data D1 ′, whereas in the conventional method, when the number of input data items is the same, one is obtained. Only one execution time estimate 5 is obtained. That is, the conventional method means that the estimated execution time includes an error. On the other hand, in the present embodiment, the estimation error can be made smaller and the execution time estimated value 5 can be obtained with high accuracy. .

A difference (estimation error) in the estimated value of execution time illustrated in FIG. 28 will be described with reference to FIG. FIG. 29 is a diagram for explaining the estimation error. In FIG. 29, a batch job J1 is a batch job that extracts data in the range [° C. ₁ , ° C. ₂ ] and performs a predetermined process.

  In this example, it is assumed that the input data amount N corresponding to the number of records is 1125. Even with the same input data amount N, the frequency distribution of temperature may be different. The input data D1 indicates a frequency distribution 29a (D1 frequency distribution table 3f-2 in FIG. 28), the input data D1 ′ indicates a frequency distribution 29b (D1 ′ frequency distribution table 3f-2 ′ in FIG. 28), and the like. is there. For example, the input data D1 is summer data, and the input data D1 ′ is winter data.

  As described above, the estimated execution time value according to the conventional method has the same result at t [J1] = 162.5 with respect to the input data D1 and the input data D1 ′. Regardless of the difference between the frequency distribution 29a and the frequency distribution 29b, the same execution time estimated value is calculated.

  When this embodiment is applied, even if the input data amount N is the same 1125 cases, when the frequency distribution 29a is obtained, the frequency distribution 29a (frequency distribution table 3f-2 of D1 in FIG. 28) is obtained. A frequency distribution 29c relating to the output data amount after the batch job J1 is obtained. Further, t [J1] = 161.7 is obtained by a linear expression using the execution time map g.

  Further, when the frequency distribution 29b (the frequency distribution table 3f-2 ′ of D1 ′ in FIG. 28) is obtained, the frequency distribution 29d related to the output data amount after the batch job J1 is obtained based on the frequency distribution 29b. Further, t [J1] = 16.3 is obtained by a linear expression using the execution time map g.

  The difference between the estimated value t [J1] = 161.7 and the estimated value t [J1] = 16.3 corresponds to an execution time estimation error in the case of the conventional method.

  Further, in the conventional method, when the output data amount is estimated from the input data amount, an estimation error also occurs in the output data amount. In the batch job network, the output data becomes input data of the subsequent batch job. Therefore, when the estimation error of the output data amount is large, the estimation error is accumulated for each batch job. In addition, since it affects not only the estimation of the output data amount but also the estimation of the execution time, it may not be possible to estimate an appropriate execution time.

  On the other hand, in the present embodiment, even if the input data amount N is the same number, the output data amount and the execution time are accurately estimated based on each of the frequency distribution 29a and the frequency distribution 29b. Therefore, even in the batch job network, the influence of the estimation error can be reduced, and the output data amount and the execution time can be estimated with high accuracy.

  A third embodiment for estimating the execution time using only the execution time map g will be described below.

  In the third embodiment, when a batch job is composed of only a single batch job, the execution time is estimated by using the frequency distribution map f from the frequency distribution table 3f-2 of the input data, and the execution time is estimated. A value of 5 is obtained.

  In the third embodiment, the determination of the variable field is the same as that in the first embodiment and the detailed description thereof is omitted. The creation of the execution time map g is the same as that in the second embodiment and the detailed description thereof is omitted.

  As described above, in this embodiment, since the output data amount and the execution time are estimated based on the distribution of the input data values, the estimation is performed as compared with the conventional method in which the output data amount and the execution time are estimated only by the input data amount. The error can be reduced.

  In this embodiment, for a batch job 2 that processes input data records composed of a plurality of fields one by one, the execution time using the frequency distribution map or / and the execution time map from the frequency distribution of the input data. By estimating the execution time, the execution time can be accurately estimated even if the source information, the setting file, etc. regarding the batch job 2 are unknown. Further, the frequency distribution of the output data can be acquired from the frequency distribution of the input data and the frequency distribution map, and the output data amount can be estimated with high accuracy.

  The present invention is not limited to the specifically disclosed embodiments, and various modifications and changes can be made without departing from the scope of the claims.

Regarding the embodiment including the first to third examples, the following additional notes are disclosed.
(Appendix 1)
A frequency distribution table showing the frequency distribution of the input data for each class, and a storage unit for storing a mapping from the input data based on the frequency distribution table to result information by a predetermined process;
The execution time of the predetermined process for the input data is estimated based on a calculation result using a mapping to the result information stored in the storage unit that receives the frequency distribution table stored in the storage unit. An execution time estimation device having an execution time estimation unit.
(Appendix 2)
The mapping stored in the storage unit is a frequency distribution map by the predetermined processing from the frequency distribution table of the input data to the frequency distribution estimated value of the output data,
An output data frequency distribution estimation unit that obtains the frequency distribution estimated value of the output data by using the frequency distribution map stored in the storage unit, which has the frequency distribution table of the input data as an input;
The execution time estimation device according to appendix 1, further comprising an execution time estimation unit that estimates an execution time based on an output data amount based on a sum of frequencies indicated by the estimated frequency distribution estimated value.
(Appendix 3)
The mapping stored in the storage unit is an execution time mapping by the predetermined process from the frequency distribution table of the input data,
The execution according to appendix 1, further comprising an execution time estimation unit that estimates an execution time by a linear expression using each value of the execution time map stored in the storage unit as a coefficient of the frequency of the same class of the frequency distribution table of the input data. Time estimation device.
(Appendix 4)
The storage unit further stores a frequency distribution map by the predetermined processing from the frequency distribution table of the input data to the frequency distribution estimated value of the output data,
Output data frequency distribution estimation that obtains a frequency distribution estimated value of the output data by a linear expression using each value of the frequency distribution map stored in the storage unit as a coefficient of the frequency of the same class of the frequency distribution table of the input data The execution time estimation apparatus according to supplementary note 3 having a unit.
(Appendix 5)
The output data frequency distribution estimation unit sets the frequency distribution estimation value of the output data acquired by a linear expression using the frequency distribution map in the frequency distribution table of the input data for the subsequent process of the predetermined process The execution time estimation apparatus described.
(Appendix 6)
The execution time estimation device according to any one of appendices 1 to 5, further comprising an input data frequency distribution creation unit that creates a frequency distribution table indicating the frequency of input data for each class in the storage unit.
(Appendix 8)
A frequency distribution table indicating the frequency distribution of the input data for each class stored in the storage unit is used as an input, and mapping from the input data based on the frequency distribution table stored in the storage unit to result information by a predetermined process is performed. Calculate
A program that causes a computer to execute a process of estimating an execution time of the predetermined process for the input data based on a result of the calculation.
(Appendix 9)
An execution time estimation method executed by a computer,
A frequency distribution table indicating the frequency distribution of the input data for each class stored in the storage unit is used as an input, and mapping from the input data based on the frequency distribution table stored in the storage unit to result information by a predetermined process is performed. Calculate
An execution time estimation method for estimating an execution time of the predetermined process for the input data based on a result of the calculation.
(Appendix 10)
A frequency distribution table showing the frequency distribution of the input data for each class, and a storage unit for storing a mapping from the input data based on the frequency distribution table to result information by a predetermined process;
A frequency distribution generation unit for storing the frequency distribution table and the mapping;
Based on a calculation result using a mapping to the execution information stored in the storage unit that receives the frequency distribution table stored in the storage unit, an execution time of the predetermined process for the input data is estimated. A system having an execution time estimation unit.
(Appendix 11)
The frequency distribution generation unit
A variable field determination unit that acquires a schema of input / output data and determines a variable field that affects output data by the predetermined processing among the fields of the input data based on the schema;
Frequency distribution for determining a class number and a class width based on the range of the variable field determined by the variable field determination unit and the estimated number of records, and generating a frequency distribution table of the input data in the storage unit A table generator,
The system according to claim 10, further comprising: a mapping creation unit that creates the mapping to the execution information.
(Appendix 12)
The variable field determination unit includes:
The standard input data set including the estimated number of records set by using a normal random number with an average or standard deviation for each field value, and the same value as the first field of the standard input data set other than the target field, A generating unit that generates an input data set in which the target field is changed using a normal random number having an average or standard deviation different from the reference input data set;
By comparing the reference output data set obtained by performing the predetermined processing on the reference input data set and the output data set obtained by performing the predetermined processing on the input data set, the target field Calculating a degree of influence of the predetermined processing on the output data set;
The system according to claim 11, further comprising: a determining unit that determines the target field as the variable field when the influence degree is equal to or greater than a predetermined value.
(Appendix 13)
The variable field determination unit includes:
When the number of records of the reference output data set and the output data set is different, a first setting unit that sets a maximum value for the degree of influence,
In the comparison between the output data set and the reference output data set, when there is a field having a different value other than the target field, a second setting unit that sets a ratio of the number of fields having a different value to the number of fields of all records as the influence The system according to appendix 12, which has:
(Appendix 14)
The mapping creation unit
A data generation unit for generating random data for each frequency for each class of the frequency distribution table of the input field;
A map creation unit that creates the map showing the frequency distribution by counting the class in which the output data is processed one by one in the predetermined process, and dividing the count result by the frequency of the class;
14. The system according to any one of appendices 11 to 13, comprising:
(Appendix 15)
The mapping creation unit
A data generation unit for generating random data for each frequency for each class of the frequency distribution table of the input field;
A mapping creation unit that creates the mapping indicating the execution time of each class by averaging the execution times measured and processed for each class in the predetermined process;
14. The system according to any one of appendices 11 to 13, comprising:

2 Batch job 6 I / O data schema 11 CPU
12 Main storage device 13 Auxiliary storage device 14 Input device 15 Display device 16 Output device 17 Communication I / F
18 drive 19 storage medium 40 frequency distribution generation unit 41 variable field determination unit 42 frequency distribution table generation unit 43 frequency distribution mapping generation unit 44 execution time mapping generation unit 50 execution time estimation unit 51 input data frequency distribution generation unit 52 execution time estimation unit 53 Output data frequency distribution estimation unit 130 Storage unit 1000 System f Frequency distribution mapping g Execution time mapping

Claims (5)

The value of the field of the input data that affects the output data for each class obtained by class classification, the classified and frequency distribution table to set the input frequency of the input data, the input data sampled in each class of該度number distribution table A storage unit for storing a frequency distribution map that sets an input frequency and obtains an output frequency of output data by performing predetermined processing on the sampled input data ;
A setting unit for setting the input frequency of the target input data of the predetermined processing for each class of the frequency distribution table stored in the storage unit ;
The output frequency of the output data after the predetermined processing with respect to the target input data is acquired for each class by a linear expression using the value of the frequency distribution map as a coefficient, and the acquired output frequencies for each class are totaled. it is, estimator that having a <br/> an output power estimator for estimating the output power with respect to the target input data. The storage unit stores the predetermined processing execution time indicates an execution time by mapping per input data for each class frequency distribution table of the input data,
By obtaining the execution time after the predetermined processing for the target input data for each class by a linear expression using the value of the execution time map as a coefficient, and summing the obtained execution times for each class, The estimation apparatus according to claim 1, further comprising: an execution time estimation unit that estimates the execution time for the target input data . The storage unit stores the predetermined processing execution time indicates an execution time by mapping per input data for each class frequency distribution table of the input data,
Using said output power of said each class obtained by the output power estimating unit as the target input data of hierarchical class each, by a linear expression that the coefficient values of the mapping the execution time stored in the storage unit, Execution time estimation for estimating the execution time for the target input data by acquiring the execution time after the predetermined processing for the target input data for each class and summing the acquired execution times for each class The estimation apparatus according to claim 1, further comprising a unit. The value of the field of the input data affects the output data stored in the storage unit for each class obtained by class classification, as inputs frequency distribution table for setting the input power of the classified the input data, stored in the storage unit It has been set the input frequency of the input data sampled in each class of該度number distribution table, to calculate a frequency distribution map to obtain an output frequency of the output data by performing a predetermined process on the input data the sampled Memorize in the memory ,
For each class of the frequency distribution table stored in the storage unit, set the input frequency of the target input data of the predetermined processing,
The output frequency of the output data after the predetermined processing for the target input data is obtained for each class by a linear expression using the value of the frequency distribution map as a coefficient,
The program which makes a computer perform the process which estimates the said output frequency with respect to the said target input data by totaling the said output frequency for every said class acquired . An execution time estimation method executed by a computer,
The value of the field of the input data affects the output data stored in the storage unit for each class obtained by class classification, as inputs frequency distribution table for setting the input power of the classified the input data, stored in the storage unit It has been set the input frequency of the input data sampled in each class of該度number distribution table, to calculate a frequency distribution map to obtain an output frequency of the output data by performing a predetermined process on the input data the sampled Memorize in the memory ,
For each class of the frequency distribution table stored in the storage unit, set the input frequency of the target input data of the predetermined processing,
The output frequency of the output data after the predetermined processing for the target input data is obtained for each class by a linear expression using the value of the frequency distribution map as a coefficient,
An estimation method for estimating the output frequency for the target input data by summing the acquired output frequencies for each of the classes .

Images