JP2012133608A - Histogram generation device, method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide technique for generating a histogram with high speed.SOLUTION: A histogram generation device includes: determination means for, on the basis of object data for generating a histogram, determining a plurality of division candidate points that are to be candidates for division points to determine respective segments of a histogram; change value calculation means for respectively calculating change values of frequency of the object data at the respective division candidate points; selection means for selecting fewer numbers of division candidate points than the number of division candidate points determined by the determination means by sampling the plurality of division candidate points on the basis of the respective change values calculated by the change value calculation means; and histogram generation means for generating a histogram of the object data by using the division candidate points selected by the selection means.

Description

  The present invention relates to a histogram generation technique.

  A histogram is a kind of statistical graph having a frequency (frequency) on the vertical axis and a class on the horizontal axis, and expresses the distribution of data. The distribution of data is visually grasped by a histogram generated with respect to given data. In addition to these points, the histogram is applied to various industrial techniques.

  One application of histograms is database query optimization. In query optimization, the most efficient query plan is selected from a plurality of query plans for executing a query. In the query optimization process, a query is decomposed into a plurality of small sub-queries, and some execution plans are generated according to their application order.

  Since the database maintains a histogram of the frequency (number of data) corresponding to the attribute value of the data as statistical information, the number of results of each subquery should be estimated from the histogram without actually executing the query Can do. Therefore, in the query optimization process, the calculation cost of each execution plan is estimated using this estimated value, and the one with the lowest cost is selected as the optimal execution plan.

  In such an example of database query optimization, the accuracy of the histogram determines whether the execution plan is good or bad. For example, when the accuracy of the histogram is poor, an appropriate execution plan is not selected, and as a result, a longer time than the original time may be spent processing the query. Therefore, generating a highly accurate histogram is important for processing a query at high speed.

  As a method for generating such a histogram, there are a method based on a partition rule (see Non-Patent Document 1), a method based on model selection based on an MDL (Minimum Description Length) principle (see Non-Patent Documents 2 and 3), and the like. Exists.

  The former divides data into the number of bins desired by the user according to the partition rule. Here, the bin means an arbitrary section in which a set having values in the histogram is divided.

  As shown in Non-Patent Document 1, a plurality of rules have been proposed as the partition rule. For example, there are rules such as EQUI-WIDTH, EQI-DEPTH, MAX-DIFF, COMPRESED. The EQUI-WIDTH method generates a histogram by uniformly dividing a range of data values. The EQUI-DEPTH method generates a histogram so that the frequency of values is equal in each bin. The MAX-DIFF method divides in order from the place where the frequency difference between consecutive unique data values is large. The COMPRESSED method gives individual bins to frequently used values, and uses the EQUI-WIDTH and EQUI-DEPTH methods for the rest. However, in any of these partition rules, the user needs to specify the number of bins.

  On the other hand, the latter generates a histogram by estimating the probability density distribution generating the data from the data. Among them, model selection is performed according to the MDL principle, and a model having a complexity that best represents data is selected. For this reason, the latter can automatically select an appropriate number of bins even if the user does not specify the number of bins like the former.

Viswanath Poosala; Yannis E. Ioannidis; Peter J. Haas; Eugene J. Shekita, "Improved Histograms for Selectivity Estimation of Range Predicates", the 1996 ACM SIGMOD international conference Petri Kontkanen; Petri Myllymaki, "MDL Histogram Density Estimation", Eleventh International Conference on Artificial Intelligence and Statistics (AISTATS 2007), 21-24 Mar 2007 Rissanen J; Speed T.P; Yu B, "Density estimation by stochastic complexity", IEEE Transactions on Information Theory, Vol. 38, March 1992.

  However, the histogram generation method based on model selection as described above requires a large amount of calculation compared to the method based on the partition rule. This is because, for each bin number, a description length is calculated for measuring the quality of the division for all the bins.

  Usually, it is necessary to calculate the number of selection patterns for selecting a division point from a plurality of division candidate points. In the proposed method described above, the amount of calculation is reduced by dynamic programming. However, even with such a proposed method, the amount of calculation still increases as the number of candidate division points increases.

  An object of the present invention is to provide a technique for generating a histogram at high speed.

  Each aspect of the present invention employs the following configurations in order to solve the above-described problems.

  A 1st aspect is related with a histogram production | generation apparatus. A histogram generation device according to a first aspect includes: a determination unit that determines a plurality of division candidate points that are candidates for division points for determining each section of a histogram based on target data for generating a histogram; The change value calculation means for calculating the change value of the frequency of the target data at the division candidate points, and sampling the plurality of division candidate points based on the change values calculated by the change value calculation means, Selection means for selecting a smaller number of division candidate points than the number of division candidate points determined by the determination means, and histogram generation means for generating a histogram of the target data using the division candidate points selected by the selection means And comprising.

  The second aspect relates to a histogram generation method. In the histogram generation method according to the second aspect, the computer determines a plurality of division candidate points that are candidates for division points for determining each section of the histogram based on the target data for generating the histogram. By calculating the change value of the frequency of the target data at the division candidate points, and sampling the plurality of division candidate points based on the calculated change values, the number of division candidate points determined above is determined. A small number of division candidate points are selected, and a histogram of the target data is generated using the selected division candidate points.

  As another aspect, a histogram generation program that causes a computer to realize any of the above-described configurations may be used, or a computer-readable storage medium that records such a program may be used.

  According to each aspect described above, a technique for generating a histogram at high speed can be provided.

The conceptual diagram which shows the structural example of the relational database system in 1st Embodiment. The figure which shows the purchase table as an example of a data storage part. The figure which shows the goods table as an example of a data storage part. The figure which shows the example of a histogram. The figure which shows the example of a histogram storage table. The flowchart which shows the operation example at the time of the histogram generation of the relational database system in 1st Embodiment. The figure which shows notionally the example of the calculation method of the change value of a division | segmentation candidate point. The figure which shows notionally the example of calculation of the change value with respect to a division | segmentation candidate point. The figure which shows the example of an execution plan. The figure which shows the example of an execution plan. The figure which shows the 2nd example of a histogram storage table. The figure which shows the example of the record extracted as a result of the data inquiry process. The conceptual diagram which shows the structural example of the relational database system in 2nd Embodiment. The figure which shows the example of the processing time data stored in a processing time storage part. The flowchart which shows the operation example at the time of the histogram production | generation of DB system in 2nd Embodiment.

  Embodiments of the present invention will be described below. In addition, embodiment mentioned below is an illustration and this invention is not limited to the structure of the following embodiment.

  The histogram generation device according to the present embodiment includes a determination unit that determines a plurality of division candidate points that are candidates for division points for determining each section of the histogram, based on target data for generating a histogram, The change value calculation unit that calculates the change value of the frequency of the target data at each division candidate point, and the determination by sampling the plurality of division candidate points based on each change value calculated by the change value calculation unit A selection unit that selects a smaller number of division candidate points than the number of division candidate points determined by the unit, and a generation unit that generates a histogram of target data using the division candidate points selected by the selection unit. Prepare.

  In the histogram generation device, a change value of the frequency of the target data is calculated at each division candidate point for determining each section of the histogram, and a plurality of division candidate points are sampled and narrowed based on the change value. A histogram of the target data is generated by using the candidate points. That is, the candidate division points used when the histogram is finally generated are set to a number smaller than the number of candidate division points determined based on the target data.

  Therefore, according to the above-described histogram generation device, a histogram can be generated at high speed even when a histogram generation method or the like based on model selection in which the amount of calculation increases depending on the number of candidate division points is adopted. it can.

  Hereinafter, the details of the above-described embodiment will be described. Each of the following embodiments is an example when the configuration of the above-described histogram generation device is applied to a relational database system. The configuration of the above-described histogram generation device is not limited to application to the relational database system, and can be applied to various modes for generating a histogram. For example, the configuration of the histogram generation device can be applied to a distributed database system. In this system, in order to make the data load uniform among the databases, the load is made uniform based on the distribution of data. Generation of a histogram used for predicting the distribution of the data may be realized by the configuration of the histogram generation apparatus.

[First Embodiment]
〔System configuration〕
FIG. 1 is a conceptual diagram showing a configuration example of a relational database system (hereinafter simply referred to as a DB system) 10 in the first embodiment. The DB system 10 in the first embodiment includes a data storage unit 100, a histogram generation system 200, a parameter designation unit 300, a histogram storage unit 400, a data inquiry unit 500, and the like.

  The DB system 10 has, for example, a CPU (Central Processing Unit), a memory (RAM (Random Access Memory), a ROM (Read Only Memory), etc.), an input / output interface, etc., and a program stored in the memory is executed by the CPU. As a result, the above-described processing units are realized. The data storage unit 100 and the histogram storage unit 400 are realized on a memory. Note that this embodiment does not limit the hardware configuration of the DB system 10.

  The data storage unit 100 is a relational database that stores arbitrary data. 2 and 3 are diagrams showing examples of the data storage unit 100. FIG. FIG. 2 shows a purchase table (purchase T) for storing a history of product purchase by the purchaser. Each record of the purchase table stores a product ID for specifying the purchased product, a purchaser ID for specifying the purchaser, and date / time information for specifying the purchase time.

  FIG. 3 shows a product table (product T) for storing product information. Each record of the product table stores the product ID, product name, and price. The product ID stored in the product table corresponds to the product ID stored in the purchase table.

  The histogram generation system 200 is a processing system that generates a histogram, and corresponds to the configuration of the above-described histogram generation apparatus. The histogram generation system 200 includes a division candidate point determination unit (hereinafter simply referred to as a determination unit) 210, a change value calculation unit 220, a division candidate point selection unit (hereinafter simply referred to as a selection unit) 230, and a histogram generation unit 240. Etc. Each of these processing units included in the histogram generation system 200 is also realized by executing a program stored in the memory by the CPU.

  The determination unit 210 acquires data (target data) for which a histogram is to be generated from the data storage unit 100, and determines a plurality of division candidate points based on the target data. Here, the division candidate point means a candidate point that can be a boundary between bins of a generated histogram.

  The change value calculation unit 220 calculates a value indicating the ease of division for each division candidate point determined by the determination unit 210. In the present embodiment, a change value of the frequency of the target data at each division candidate point is calculated as a value indicating the ease of division. This change value indicates how rapidly the frequency of the target data changes before and after the division candidate point. Since the histogram shows the density distribution in rectangular (bar graph) bins, in order to approximate the histogram to the actual density distribution, many dividing points are placed in areas where the frequency changes suddenly (assigning many bins), and the change is made It is preferable to place a small number of dividing points or not to divide the dividing points in a small portion. Thereby, this embodiment regards the easiness of being divided as a change value of the frequency at the dividing point.

  In the calculation of the change value, the change value calculation unit 220 may calculate data in a data accuracy range centered on each division candidate point. However, when the target data is small and there is missing data, if the change value is calculated only in the vicinity of the division candidate point, the change value becomes large due to the missing data, resulting in a result different from the original density distribution. there is a possibility. Therefore, in the present embodiment, the change value calculation unit 220 calculates the change value using the parameter specified by the parameter specification unit 300. This parameter includes, for example, a value indicating a data range as a target in the calculation of the change value. The parameters used in the change value calculation will be described in detail in the operation example section.

  The selection unit 230 samples the plurality of division candidate points determined by the determination unit 210 based on the change value of the frequency of the target data at each division candidate point calculated by the change value calculation unit 220. In this sampling, division candidate points with large frequency changes are emphasized and selected. If the number of sampling points with a large frequency change is emphasized and the number of sampling is smaller than the number of division candidate points determined by the determination unit 210, the number of samplings is set to a number specified by the parameter specification unit 300 described later. It may be set, or may be determined by a random number whose upper limit is the designated number.

  Specific sampling methods include the following examples, for example. The first example is a method of acquiring a change value threshold value from the parameter designating unit 300 and selecting a division candidate point having a change value equal to or greater than the threshold value. The second example is a method of further sampling at random from among candidate division points having a change value equal to or greater than this threshold value. The third example is a method of intensively selecting a division candidate point having a large frequency change with a value proportional to the change value. This third example will be described in detail in the section of the operation example.

  By this sampling, the selection unit 230 selects a number of points smaller than the number of division candidate points determined by the determination unit 210 as final division candidate points.

  The histogram generation unit 240 generates a histogram using the sampled division candidate points. At this time, the histogram generation unit 240 may select a division point to be finally used to create a histogram from the division candidate points selected by the selection unit 230 (sampled division candidate points). Alternatively, the division candidate point may be used as a division point as it is. FIG. 4 is a diagram illustrating an example of a histogram. For example, the histogram generation unit 240 generates a histogram as shown in FIG. 4 regarding the price of the product table of FIG. Various known methods as described above are used for this histogram generation method. Since the present embodiment does not limit the histogram generation method itself, description thereof is omitted here.

  The parameter specifying unit 300 specifies various parameters used in the histogram generation system 200. For example, parameters used by the change value calculation unit 220, the upper limit number of samples used by the selection unit 230, parameters used by the histogram generation unit 240, and the like are designated. The parameters used in the histogram generation unit 240 are determined according to the histogram generation method.

  For example, the parameter specifying unit 300 may read a desired parameter from a user setting file that is set to be adjustable, or may acquire a parameter input through a user interface (not shown). This embodiment does not limit the specific parameter designation method of the parameter designation unit 300.

  The histogram storage unit 400 stores the histogram generated by the histogram generation system 200. FIG. 5 is a diagram illustrating an example of a histogram storage table. For example, the histogram storage unit 400 stores a histogram as shown in FIG. 4 in a table format as in the example of FIG. Each record in the histogram storage table of FIG. 5 corresponds to each bin of the histogram, stores the left division point in the start value column, stores the right division point in the end value column, and sets the number of data in the range. Store each in the occurrence number column. Note that this embodiment does not limit the histogram storage method.

  The data inquiry unit 500 receives a data query (SQL (Structured Query Language)) from the user, extracts data that matches the query from the data storage unit 100, and returns the extracted data to the user. The data inquiry unit 500 optimizes the query when extracting data that matches the query. Specifically, the data inquiry unit 500 generates several execution plans from the SQL issued by the user, and calculates execution costs while referring to the histogram for each execution plan. The data inquiry unit 500 selects and executes the execution plan having the lowest calculated execution cost.

[Operation example]
Hereinafter, an operation example of the DB system 10 in the first embodiment will be described.
First, an example of the operation of the DB system 10 when generating a histogram will be described with reference to FIG. FIG. 6 is a flowchart illustrating an operation example when generating a histogram of the DB system 10 according to the first embodiment.

  The parameter specifying unit 300 acquires various parameters used in the histogram generation system 200 and sends the acquired various parameters to the histogram generation system 200 (S100). These various parameters are read from a setting file, for example. The parameter specifying unit 300 may use a predetermined parameter value determined in advance.

The determination unit 210 acquires target data for which a histogram is to be generated from the data storage unit 100 (S110). For example, the determination unit 210 obtains the following data X when creating a histogram for the price data in the product table of FIG.
X = {6980, 8600, 1980, 100,000, 1980, ...}

  Next, the determination unit 210 determines a plurality of division candidate points from the data obtained in the above process (S110) (S120). A plurality of methods can be considered as the method for determining the division candidate points. Here, examples of two types of determination methods are given.

The first determination method is a method in which division candidate points are arranged before and after all values that can be taken by the target data. For example, the set of candidate division points determined by the first method can be represented by C1 below.
C1 = {x_min + e / 2 + t × e | t = 0, 1,. . . , (X_max−x_min) / e−1}

  Here, x_min indicates the minimum value of the target data, x_max indicates the maximum value of the target data, and e indicates the accuracy of the target data. In the example of price data in the product table in FIG. 3, assuming that the accuracy of the target data is set to 10, each division candidate point is set in units of 10 yen.

The second determination method is a method of arranging division candidate points before and after all data points of the target data. For example, the set of candidate division points determined by the second method can be represented by the following C2.
C2 = {x_j−e / 2 | x_jεX} ∪ {x_j + e / 2 | x_jεX}

  Here, X represents a set of target data, x_j represents each target data, and e represents the accuracy of the target data.

  According to the first determination method, the set C1 of candidate division points can be automatically determined if the maximum value and the minimum value are acquired from the target data. On the other hand, according to the second determination method, the set C2 of candidate division points can be determined by providing candidate division points before and after all the target data. Note that the present embodiment is not limited to such two types of division candidate point determination methods.

  Next, the change value calculation unit 220 calculates a change value for each of the plurality of candidate division points determined in the process (S120) (S130). FIG. 7 is a diagram conceptually illustrating an example of a method for calculating a change value of a division candidate point. In the example of FIG. 7, two parameters ε and dx specified by the parameter specifying unit 300 are used in calculating the change value. The parameter dx indicates a data range for obtaining the change value. When the value of the parameter dx is set to data accuracy, the change value is calculated only for the data range adjacent to the division candidate point, and when it is set to be larger than the data accuracy, the division candidate point is centered. A change value reflecting a wider range of data trends is calculated.

In FIG. 7, the function f (x) indicates the frequency with which the value of [x−ε / 2, x + ε / 2] (open interval) appears in the data using the parameter ε. In this case, the change value calculation unit 220 calculates the change values P ^to (x) by the following formula.
P ^to (x) = | f (x1−dx / 2) −f (x1 + dx / 2) | / dx

  As a result, the calculated change value represents the slope of the arrow in FIG. FIG. 8 is a diagram conceptually illustrating a calculation example of the change value for the division candidate point. The parameters dx and ε may be set to data accuracy. Further, when the target data is small with respect to the data accuracy and there is data omission, the parameters dx and ε may be set to a value larger than the data accuracy.

Next, the selection unit 230 normalizes each change value so that the change values (P ^to (x)) at each division candidate point have a probability distribution. Assuming that a value obtained by normalizing the change values (P ^to (x)) at each division candidate point is P (x), the normalization can be expressed by the following expression, for example.
^{P (x) = P ~ (} x) / ΣP ~ (x)

  The selection unit 230 uses N of the plurality of division candidate points determined in the process (S120) as division candidate points based on the probability distribution (P (x)) of the change values normalized in this way. Select (S140). The sampling number N is acquired from the parameter specifying unit 300, for example. For example, the selection unit 230 selects the division candidate points with the sampling number N specified from the parameter specification unit 300 as an upper limit at random according to the probability distribution (P (x)).

  The histogram generation unit 240 generates a histogram using the candidate division points sampled in the above process (S140) (S150). As this histogram generation method, for example, a histogram generation method whose calculation amount depends on the number of division candidate points (for example, a method proposed in Non-Patent Document 2 and Non-Patent Document 3) is used. When parameters are required for each histogram generation method, various parameters are acquired from the parameter specifying unit 300.

  The histogram generation unit 240 stores the histogram generated in the above process (S150) in the histogram storage unit 400 (S160).

  Next, an operation example of the DB system 10 when processing a data query using the histogram generated as described above will be described.

The data inquiry unit 500 generates an SQL execution plan after parsing the inquiry SQL from the user. For example, when data as shown in FIGS. 2 and 3 is stored in the data storage unit 100, it is assumed that the following SQL query has been made for the data.
SELECT * FROM Purchasing T, Commodity T WHERE Date &Time> 08/10 AND Price> 5000 AND Purchasing T. Product ID = Product T. Product ID;

At this time, the data inquiry unit 500 generates, for example, two execution plans shown in FIGS. 9A and 9B. 9A and 9B are diagrams illustrating examples of execution plans. The two execution plans include a join between the two tables. The data inquiry unit 500 calculates the cost of each execution plan, for example, using the following equation as the cost when the two tables are nested loop joined.
P1 + Ceil (P1 / (M−1)) × P2

  Here, P1 indicates the number of pages in the outer loop table, P2 indicates the number of pages in the inner loop table, and M indicates the work buffer size. Ceil (n) is a function for obtaining the smallest integer value greater than or equal to the value specified for n. It is sufficient that a known method is used for such a cost calculation method (see Hiroyuki Kitagawa's database system, etc.), and in this embodiment, the calculation method is not limited, so the description is simplified here. .

  The data inquiry unit 500 calculates the number of pages (P1 and P2) of the outer loop table and the inner loop table from the selection rate of each table in order to calculate the above-described cost. The data inquiry unit 500 acquires the selection rate using the histogram generated as described above and stored in the histogram storage unit 400. This selection rate means the ratio of the number of extracted records to the total number of records.

  Here, when calculating the cost of the execution plan of FIG. 9A, a record extracted from the purchase table with “date and time> 08/10” is used for the outer loop of the join, and “price> 5000” from the product table is used for the inner loop. ”Is used. The data inquiry unit 500 obtains each selection rate using the histogram stored in the histogram storage unit 400 in order to obtain the number of pages.

  For example, it is assumed that the histogram shown in FIG. 10 is stored in the histogram storage unit 400 for the price data of the product table of FIG. FIG. 10 is a diagram illustrating a second example of the histogram storage table. At this time, the data inquiry unit 500 obtains the number of records extracted with “price> 5000” from the number of appearances (300 = 100 + 100 + 20 + 80) in the range where the start value is greater than 5000, and the total number of appearances of this number of appearances A ratio (60%) with respect to (500) is obtained as a selection rate in the case of extracting “price> 5000” from the product table.

  Similarly, the data inquiry unit 500 obtains the selection rate when extracting from the purchase table with “date and time> 08/10” based on the histogram stored in the histogram storage unit 400. Here, it is assumed that the selectivity is acquired as 10%. Further, assuming that the number of pages in the purchase table is 500, the number of pages in the product table is 30, and the work buffer size M is 5, the cost of each execution plan in FIGS. 9A and 9B is calculated as follows: .

The cost of the execution plan in FIG. 9A = 500 × 0.1 + Ceil (500 × 0.1 / 4) × 30 × 0.6 = 284
The cost of the execution plan in FIG. 9B = 30 × 0.6 + Ceil (30 × 0.6 / 4) × 500 × 0.1 = 268

  According to this example, the cost of the execution plan of FIG. 9B is lower than that of FIG. 9A. Therefore, the data inquiry unit 500 executes the execution plan of FIG. 9B and obtains a result. FIG. 11 is a diagram illustrating an example of records extracted as a result of the data inquiry process.

[Operation and Effect of First Embodiment]
In the DB system 10 in the first embodiment described above, the determination unit 210 determines a plurality of division candidate points for determining each section of the histogram from the data that is the target of histogram generation. Subsequently, the change value calculation unit 220 calculates the change value of the frequency of the target data at each division candidate point, and the selection unit 230 attaches importance to the division candidate points having a large frequency change based on each change value. A plurality of division candidate points are sampled so as to be selected. As a result, the histogram generation unit 240 generates a histogram of the target data using the division candidate points set to a number smaller than the number of division candidate points determined based on the target data by sampling.

  As described above, the selection unit 230 samples the division candidate points determined based on the target data at a predetermined number or less, so even if a histogram generation method in which the calculation amount depends on the number of division candidate points is adopted. A histogram can be generated at high speed.

  Furthermore, in the first embodiment, points unnecessary for division are omitted from a plurality of division candidate points determined based on the target data, and points important for division are selected as division candidate points. . As described above, it is preferable that a large number of division points for determining each section of the histogram be placed in a portion where the frequency changes rapidly. In the first embodiment, the importance of this division is evaluated by the change value.

  Therefore, according to the first embodiment, even if the division candidate points are reduced, since the points important for the division are selected with priority, it is possible to suppress a decrease in accuracy of the histogram.

[Second Embodiment]
In the first embodiment described above, the upper limit value of the number of samplings is specified in the parameter specifying unit 300, but the second embodiment is realized so that the histogram generation time can be specified. Hereinafter, the DB system 10 in the second embodiment will be described focusing on the contents different from the first embodiment, and the description of the same contents will be omitted as appropriate.

〔System configuration〕
FIG. 12 is a conceptual diagram illustrating a configuration example of the DB system 10 in the second embodiment. As shown in FIG. 12, the DB system 10 in the second embodiment further includes a processing time storage unit 600 in addition to the configuration of the first embodiment. The processing time storage unit 600 is realized on a memory such as a RAM (Random Access Memory), an HDD (Hard Disk Drive), or the like.

  The parameter specifying unit 300 sends the histogram generation desired time to the selection unit 230 in addition to the upper limit value of the sampling number of the division candidate points. The histogram generation desired time is a time length spent for generating a histogram and is a time length desired by a user or the like. The desired histogram generation time may be stored in a setting file together with the upper limit value of the number of samplings, or may be acquired via a user interface or the like.

  FIG. 13 is a diagram illustrating an example of processing time data stored in the processing time storage unit 600. As illustrated in the example of FIG. 13, the processing time storage unit 600 stores the time spent for actually generating the histogram with respect to the sampling number of the division candidate points.

  The selection unit 230 determines the sampling number of the division candidate points based on the upper limit value of the sampling number acquired from the parameter specification unit 300, the histogram generation desired time, and the processing time data stored in the processing time storage unit 600. To do. Specifically, when processing time data of a predetermined number of records is stored in the processing time storage unit 600, the selection unit 230 determines the upper limit of the number of samplings from the processing time data extracted from the processing time storage unit 600. The number of samplings corresponding to the histogram generation desired time is determined within a range not exceeding the value.

  The histogram generation unit 240 measures the time required to generate the histogram, and stores the measured generation time and the number of samplings at that time in the processing time storage unit 600.

[Operation example]
FIG. 14 is a flowchart illustrating an operation example when generating a histogram of the DB system 10 according to the second embodiment.

  The parameter specifying unit 300 sends the desired histogram generation time to the histogram generation system 200 in addition to the various parameters (for example, the upper limit value of the number of samplings) used in the first embodiment (S101). Henceforth, each process of (S110), (S120), and (S130) is the same as that of 1st Embodiment (refer FIG. 6).

  The selection unit 230 calculates and determines the number of samplings before sampling the division candidate points (S141). Specifically, when data of a predetermined number of records or more is stored in the processing time storage unit 600, the selection unit 230 extracts records including the sampling number and processing time, and the extracted sampling number and processing Regression analysis is performed using multiple combinations of time. The selection unit 230 calculates a value corresponding to the desired histogram generation time from the obtained regression curve, and determines the calculated value as the number of samplings. If the processing time storage unit 600 does not have more data than the predetermined number of records, the sampling number may be determined by the same method as in the first embodiment.

  Thereafter, the selection unit 230 samples the division candidate points according to the determined sampling number (S140). The subsequent processes (S150) and (S160) are the same as in the first embodiment.

  When the histogram is stored in the histogram storage unit 400, the histogram generation unit 240 stores in the processing time storage unit 600 the processing time taken to generate the histogram and the number of division candidate points sampled at that time (S161).

[Operation and Effect of Second Embodiment]
Thus, in the second embodiment, when a histogram is generated by the histogram generation unit 240, the processing time and the number of samplings (number of division candidate points) are stored in the processing time storage unit 600. The sampling number, which is the number of candidate division points used for generating the histogram, corresponds to the desired histogram generation time desired by the user or the like based on the history data of the processing time and the sampling number stored in the processing time storage unit 600. The value to be determined.

  As described above, according to the second embodiment, for example, a plurality of histograms are generated in advance at each of a plurality of sampling numbers, and history data is stored in the processing time storage unit 600, so that a user or the like desires. A histogram can be generated at a time approximate to the generation time. This is because the sampling number corresponding to the desired time can be calculated by regression analysis of the history data stored in the processing time storage unit 600.

  The user can recognize the sense of time allowed as the time spent for generating the histogram, but it is difficult to recognize the number of samplings for generating the histogram at the allowable time. For this reason, it is often more useful for the user to specify the desired histogram generation time than to specify the sampling number. Therefore, according to the second embodiment, a user-friendly system can be provided.

[Modification]
In each of the above-described embodiments, the selection unit 230 determines a set of one division candidate point by sampling the plurality of division candidate points determined by the determination unit 210. However, the selection unit 230 may determine a set of a plurality of types of division candidate points. That is, the selection unit 230 is a set of sampling numbers determined corresponding to a range that does not exceed the upper limit value of the sampling number or the histogram generation desired time, and at least one division candidate point included in each set is different. A set of candidate division points is determined.

  In this case, the selection unit 230 provides the histogram generation unit 240 with a set of the determined plural types of division candidate points. The histogram generation unit 240 generates a histogram for each set, and stores the histogram having the highest evaluation among the generated histograms in the histogram storage unit 400. In addition, since a well-known technique should just be utilized for the evaluation method of a histogram (for example, refer nonpatent literature 2 and nonpatent literature 3), description is abbreviate | omitted here.

  According to such a modification, it is possible to realize a high speed generation of the histogram and to stably suppress a decrease in the accuracy of the histogram.

  In the description of the above embodiment, a plurality of flowcharts are used, and a plurality of steps are described in order, but the order of the plurality of steps may be changed within a range that does not hinder the contents. it can. Further, the above-described embodiment and a plurality of modified examples can be combined within a range in which the contents do not conflict with each other.

10 Relational database system (DB system)
DESCRIPTION OF SYMBOLS 100 Data storage part 200 Histogram generation system 210 Division | segmentation candidate point determination part (determination part)
220 change value calculation unit 230 division candidate point selection unit (selection unit)
240 Histogram generation unit 300 Parameter specification unit 400 Histogram storage unit 500 Data inquiry unit 600 Processing time storage unit

Claims (7)

Determining means for determining a plurality of division candidate points to be candidates for division points for determining each section of the histogram based on target data for generating a histogram;
Change value calculating means for calculating a change value of the frequency of the target data at each of the division candidate points;
By sampling the plurality of division candidate points based on each change value calculated by the change value calculation unit, a smaller number of division candidate points than the number of division candidate points determined by the determination unit are selected. A selection means;
A histogram generation means for generating a histogram of the target data using the division candidate points selected by the selection means;
A histogram generating apparatus comprising: Parameter specifying means for sending a parameter indicating a data range to be calculated of the change value to the change value calculating means,
Further comprising
The change value calculation means calculates a change value of the frequency of the target data at each division candidate point in the data range indicated by the parameter acquired from the parameter specifying means,
The histogram generation apparatus according to claim 1. The parameter designation means sends an upper limit value of the number of samplings used by the selection means to the selection means,
The selection means selects a number of division candidate points that is less than the number of division candidate points determined by the determination means within a range that does not exceed the upper limit value acquired from the parameter designation means.
The histogram generating apparatus according to claim 2, wherein: Processing time storage means for storing a plurality of combinations of the sampling number for sampling by the selection means and the processing time of histogram generation by the histogram generation means using the division candidate points selected by the sampling number;
Further comprising
The parameter specifying means sends a histogram generation desired time to the selection means,
The selection unit calculates a sampling number corresponding to the histogram generation desired time sent from the parameter designating unit by analyzing a plurality of combinations stored in the processing time storage unit, and calculates the number of samplings calculated. Sampling so that candidate division points are selected,
The histogram generation apparatus according to claim 2 or 3, wherein The selection means selects a plurality of types of division candidate point sets each including a number of division candidate points smaller than the number of division candidate points,
The histogram generation means generates a plurality of histograms using each of the plurality of division candidate point sets selected by the selection means, and outputs any one of the generated plurality of histograms.
The histogram generation device according to claim 1, wherein Computer
Based on target data for generating a histogram, a plurality of division candidate points that are candidates for division points for determining each section of the histogram are determined;
Calculate a change value of the frequency of the target data at each of the division candidate points,
By sampling the plurality of division candidate points based on the calculated change values, a number of division candidate points smaller than the number of division candidate points is selected,
Generating a histogram of the target data using the selected candidate division points;
Histogram generation method including the above. Computer
Determining means for determining a plurality of division candidate points to be candidates for division points for determining each section of the histogram based on target data for generating a histogram;
Change value calculating means for calculating a change value of the frequency of the target data at each of the division candidate points;
By sampling the plurality of division candidate points based on each change value calculated by the change value calculation unit, a smaller number of division candidate points than the number of division candidate points determined by the determination unit are selected. A selection means;
A histogram generating means for generating a histogram of the target data using the division candidate points selected by the selecting means;
Histogram generation program characterized by functioning as

Images