JP2001022621A - Multidimensional database management system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To attain the high speed aggregate calculation and slice processing of multidimensional data by providing a data dividing method paying attention to plural dimensions in a multidimensional database management system for storing data while being divided into plural files. SOLUTION: As for data read by input data scan processing of a node 201 of this system, a coordinate value is allocated to an effective cell by converting processing, the information of a coordinate, where the effective cell exists, is generated and the generated information of the coordinate, where the effective cell exists, is transmitted to data merging processing of an input data analytic processing node 202 by sending processing. In data merging processing, the number of effective cells on each dimension is adjusted so as to be equally distributed as a result dividing the width of division with the number of effective cells on each dimension, and data are divided into plural blocks. The block is allocated so that the number of effective cells in each storage area can be almost equal. Cells in a close distance are to be entered into the same file as much as possible.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a database management system, and more particularly to a database management system for performing high-speed aggregation calculation of multi-dimensional data, slicing processing, and data storage processing in multi-dimensional data division processing. About the system.

[0002]

2. Description of the Related Art Generally, a relational database management system allocates data for parallel processing by:
Hash partitioning and range partitioning are used. As this type of technology, “DeWitt, D., et al., 'The Gamma Dat
abase Machine Project ', IEEETransactions on Knowled
ge and Data Engineering, vol.2, no.1, pp.44-63,199
Those described in "0" are known.

To process a multidimensional database in parallel,
You need to split the data. As a technique for dividing data, a method called hash division or range division is known. The division method based on hash division or range division divides data by focusing on one dimension, and does not consider dividing data of a plurality of dimensions. For this reason, if multidimensional data is divided by a division method based on hash division or range division, dimensions that are not taken into account appear, and a large overhead is required when performing an aggregation calculation or slicing processing on the dimensions. Highly likely to occur.

Further, as a data storage method suitable for multidimensional data retrieval, a method of linearly arranging data in the order of nesting of dimensional coordinates, a method of linearly arranging data in accordance with a memory arrangement method of array data of a computer language, and a method of storing effective values Is assumed to be divided into a direct product of a subspace of a coarsely distributed part and a subspace of a densely distributed part, a storage area is allocated only to a non-empty subspace, and further, the pointer array is changed to the part described above. There is known a method applied to space to point the storage area from a pointer in an array (for example, US Pat. No. 5,359,724).

However, as described above, simply arranging data in the nested order of dimensions has the advantage that page addressing for determining the page storing the data can be processed by calculating multidimensional coordinates of the data.
There is a problem that compression corresponding to the density of valid data generated due to data distribution is not effective, and the physical arrangement distance between adjacent data in the data space is largely biased depending on the dimensional direction, that is, clustering is biased. . In addition, the method described in the above-mentioned U.S. Patent Specification has a problem that the identification regarding the density of the data distribution must be clear.

[0006]

In a database management system according to the prior art, parallelization is one of the techniques for increasing the speed. This type of database management system needs to divide the data in order to process the multidimensional database in parallel, but if the method of dividing the multidimensional data by hash division or range division is used, as described above,
Since a dimension that is not taken into account appears, there is a problem that a large overhead is likely to occur when performing an aggregation calculation or a slicing process on the dimension.

[0007] An object of the present invention is to solve the above-mentioned problems of the prior art, realize nearly uniform data distribution for each storage area for storing data in a multidimensional database, and perform aggregation calculation, slice processing, and the like. In order to balance the load, and to increase the speed of parallel processing and improve the calculation speed by increasing the possibility that data at close distances in each dimension exist in the same storage area even in aggregate calculation etc. It is an object of the present invention to provide a multidimensional database management system capable of performing the above.

[0008]

According to the present invention, there is provided a multi-dimensional database management system for storing and managing data identified by a combination of members associated with a plurality of dimensions. And one or more processing nodes for assigning a series of coordinate values to each of the members of each of the plurality of dimensions, and defining a set of coordinate values of the dimension members as data included in the set. , And divides each dimension according to the number of storage areas of the multidimensional database so that the number of effective cells is equal.
This is achieved by creating one or more blocks and dividing the multidimensional database such that these blocks are arranged in the data storage area.

The object is to divide a dimension having a large number of members for each dimension by a large number of divisions and divide a dimension having a small number of members by a small number of divisions, and to store a multidimensional database. When the number of areas is large, division is performed so that the number of divided blocks is large, and when the number of storage areas is small, division is performed such that the number of divided blocks is small.

Further, the object is to assign a coordinate value on each dimension to each of the divided blocks,
Each block is uniquely specified by using a set of coordinate values of a block on a dimension as block coordinates of each block, and one or a plurality of sets of blocks are arranged on a storage area, and distributed to each storage area. By calculating the total number of cells and arranging, on each storage area, a combination of blocks in which the difference between the maximum value and the minimum value of the total value among the storage areas is the minimum, This is achieved by arranging blocks in the same division range on each dimension of the block in different storage areas.

More specifically, the present invention analyzes data to be stored in a database management system that stores data identified by a combination of members of a plurality of dimensions and processes an inquiry request regarding the data. At this time, coordinate values are assigned to the input data according to the members of each dimension, the coordinates where valid cells are present are stored, and information on the coordinates where valid cells are present is stored. The information on the coordinates where the effective cells are obtained obtained in this way is essential for analyzing the data described below.

In analyzing the data, the number of effective cells for each member is calculated from the information on the coordinates where the effective cells of each dimension exist, and the number of effective cells of each member is added to make the number of effective cells uniform. Each dimension is divided so as to approach each other, one or a plurality of blocks are created on a multidimensional database, and the result of the division is stored. As a result, a block as a storage unit for evenly distributing data to the storage area is obtained. This block is required for the combination calculation described below.

The present invention has a processing node that includes one or more CPUs and one or more parallel readable storage devices, where the processing nodes are identified by a combination of multiple dimension members. When analyzing the target data, create sets as many as the number of storage areas such as files by combining the blocks created as a result of the division,
Calculate the sum of the number of valid cells for each set by changing the combination of blocks, change the combination so that the number of valid cells included in each set approaches evenly, and make a trial calculation. The uniformity is ensured within the range, and the result of the calculation of this combination is stored. As a result, valid cells can be evenly distributed to the storage areas, and load distribution during parallel processing can be performed.

The processing node, when storing data to be stored identified by a combination of a plurality of dimension members in a storage device, stores a set obtained by using the result of the above-described combination calculation in a file or the like. Is stored in a different storage device in which the storage area exists. Thereby, data can be actually arranged in the storage area.

[0015]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of a database management system according to the present invention will be described below in detail with reference to the drawings.

FIG. 1 is a diagram for explaining an example of multi-dimensional data having three dimensions of goods, districts, and sales. FIG. 2 is a block diagram showing a configuration of a data analysis process of a database management system according to an embodiment of the present invention. FIG. 3 is a diagram for explaining the distribution of the number of effective cells when the multidimensional data shown in FIG. 1 is divided by focusing only on dimension A. FIG. 4 is a diagram showing the multidimensional data shown in FIG. FIG. 5 is a diagram illustrating an example of division, FIG. 5 is a diagram illustrating allocation of coordinates to each block of the divided multidimensional data shown in FIG. 4, and FIG. 6 is a data storage process of a database management system according to an embodiment of the present invention. FIG. 7 is a flowchart for explaining a processing operation of a data analysis process, FIG. 8 is a flowchart for explaining a processing operation of a data storage process, and FIG. 9 is a data analysis process. Diagram illustrating a configuration example of a table for converting the members to coordinate values available in computer, FIG. 1
0 is a diagram illustrating an example of a table for expressing coordinates where valid cells are present in data analysis processing, FIG. 11 is a block diagram illustrating a configuration of dictionary definition function processing of a database management system according to an embodiment of the present invention, FIG. 12 is a diagram illustrating an example of the dimension configuration information. 2 and 6, 201
Denotes an input data read processing node, 202 denotes an input data analysis processing node, 601 denotes an input data read / distribution processing node, and 602 denotes a multidimensional data storage processing node.

As an example for explaining the embodiment of the present invention, an example of a multidimensional database as shown in FIG. 1 will be described. Now, it is assumed that the multidimensional database M includes dimensions A and B as shown in FIG.
Dimension A is a product-specific dimension, and dimension B is a district-specific dimension. The black dots in FIG. 1 represent valid cells. The sales value is assumed as the value of the cell. That is, it is assumed that the database M stores sales data for each product and each district.

A database management system for managing a multidimensional database according to an embodiment of the present invention internally has two processing phases of data analysis processing and data storage processing. Then, as shown in FIG. 2, the database management system according to the embodiment of the present invention includes an input data read processing node 201 responsible for processing for reading input data, and a processing for collecting and merging read data in one place. And a processing block for performing a data analysis process with an input data analysis processing node 202 for analyzing the collected data.

The processing nodes 201 and 202 described above
Is configured to include one or a plurality of CPUs and one or a plurality of parallel readable storage devices. Also,
Processing nodes 601 and 602 described later have the same configuration.

In FIG. 2, one or more input data read processing nodes 201 responsible for processing of reading input data exist in the system.
There is one node 202 in the system that performs a process of collecting and merging the data read in step 1 and merging the data and analyzing the collected data. The nodes 201 and 202 that perform reading and data merging may be configured as the same node.

The processing node 201 includes an input data scanning process, a conversion process, and a sending process. The input data scanning process reads input data from a file or the like, and the conversion process performs a process of assigning coordinate values to valid cells to data obtained by the input data scanning process. At this time, it is necessary that a coordinate value corresponds to each member for each dimension in advance. However, in general, the dimension members are often a natural language, a code number added to a product, or the like. Therefore, it is necessary to convert a code number added to a natural language or a product into an ID that can be used as a coordinate value on each dimension by a computer. The conversion process is a process of converting a member to a coordinate ID on the dimension of input data. I do.

Information for performing the process of converting a member into a coordinate ID on the dimension of the input data described above is stored and provided in a member name coordinate ID table as shown in FIG. This member name coordinate ID table is created by the dictionary definition function before the data analysis processing. The dictionary definition function will be described later. The conversion process is a process of converting a member on the dimension of the input data into a coordinate ID. In the example of the table shown in FIG. 9, for example, if a member on the product dimension called “stabilizer” is used, a coordinate ID is used. Is converted to “0003”. Then, the direct product of the coordinate values of each dimension member is obtained, the coordinates of the multidimensional space on the multidimensional database are allocated to each cell, and the coordinates where the valid cell exists are stored.

As a method of storing the coordinates of the effective cell,
There is a method by creating a bitmap. The bit map associates one bit on the storage device with a cell, and indicates whether the corresponding cell is a valid cell or not a valid cell by turning on or off the bit. This bitmap has a table structure in which each axis is nested on a computer. FIG. 10 shows an example of this bitmap. The example of the bitmap shown in FIG. 10 is composed of the product axis and the area axis shown in FIG. On the calculator, on the bitmap
Since on and off are represented by 1 bit, the bit map has data several orders of magnitude smaller than the input data.

In the sending process, the information of the coordinates where the valid cell exists in the bit map created on the storage device by the conversion process as described above is sent to the data merging process of the node 202. Since the information of the coordinates where the valid cell exists is smaller in size than the original input data, the load on the network and the computer main body at the time of transmission can be reduced.

The node 202 is composed of a data merging process and a data analysis process, and receives information on coordinates at which valid cells exist from the processing node 201. Then, in the data merging process, the received information on the coordinates where the valid cells are present is merged, the information on the coordinates where the valid cells are present for all the input data is created on a storage device, and stored as merge information. In the data merging process, the number of valid cells is counted while performing the merging process. The merge information created as a result is formally the same as the information on the coordinates where the valid cells exist in FIG. In the data merging process, the merge information created as described above and the total number of valid cells are passed to the data analyzing process.

As an example for explaining the embodiment of the present invention, the example of the multi-dimensional database already described with reference to FIG. 1 will be used. Decide which members to divide for each dimension. Prior to this, it has to be determined for which dimension the division is to be performed. The dimension at which the division is to be performed is specified by an input parameter or the like at the start of the data analysis process by the user of the system. In the following, first, for simplicity of explanation, 1
The division regarding only the dimension will be described.

Now, it is assumed that the user of the system designates that the division is performed with respect to the dimension A of the database shown in FIG. In this case, the system divides the total number of valid cells by the number of divisions determined by a method to be described later, and calculates an average value by which the number of valid cells included in a block that is a group of cells created as a result of the division becomes equal. Calculate the number of cells. However, in general, since a plurality of valid cells exist on a member, there is not always a member that completely matches the average number of cells. Therefore, the system finds the member that approaches the average number of cells by adding the number of effective cells on the member. That is, the system determines the number of valid cells on each member by referring to information on the coordinates where valid cells exist, and from the nesting order of the dimensions of the information on the coordinates where valid cells exist and the number of members in each dimension. Calculate by calculating the bit position corresponding to the member and summing the number of bits that are on.

FIG. 3 shows the distribution of the number of effective cells when the multidimensional data shown in FIG. 1 is divided by focusing only on dimension A. In FIG. 3, the system adds the number of valid cells from the product a0 and searches for a member that comes closest to the above-described average cell number. In the example shown in FIG. 3, the product a1 is the member closest to this average cell number. The system repeats the same operation for members after the product a1, and searches for products a2 and a3 as a result of the calculation. The number of valid cells for each member is
It can be obtained from the information on the coordinates where the valid cells exist. The information of the coordinates where the valid cells are present has a nested structure in each dimension, and since the members of each dimension are arranged according to a certain rule, the cell of each member of each dimension is valid or invalid. The bit indicating whether or not the address can be obtained by calculation. This allows the system to count all valid bits.

In the example shown in FIG. 3, the number of divisions is four, but the number of divisions in each dimension is calculated by the system according to the following equations 1 and 2, or the system user can be specified. Shall be. However, the number of storage areas in Equation 1 is the sum of the storage areas on all nodes. Equation 2
Is applied when division is performed on a plurality of dimensions.
However, in general, there are a plurality of combinations of the number of divisions that satisfy Expressions 1 and 2. The system calculates by exhausting these combinations and selects one of the results. Which calculation result is selected will be described later. 0 ≤ number of divisions in each dimension ≤ number of storage areas ... (Equation 1) Ratio of the number of divisions in each dimension = ratio of logarithm of the number of members in each dimension ... (Equation 2)

Next, the multidimensional database shown in FIG.
A case where division is performed on two dimensions will be described.
When the user of the system requests division for dimension A and dimension B, division is performed for dimension B in the same manner as described above. In this case, the average number of cells is calculated for dimension B as well as dimension A, and the addition of the number of valid cells is repeated from area b0 to obtain a division point and perform division. B1 and b2 shown in FIG. 4 represent these division points in the multidimensional database shown in FIG. FIG. 4 also shows the dimension A division points a2 and a3 described above.

When a schema of a multidimensional database is defined, if a schema for aggregate calculation has been defined, the members of each dimension of the currently stored database are included in any member of the result of aggregation. The system can know in advance. Since this function is realized by existing multidimensional database products, detailed description will be omitted. The system determines, from the currently stored multidimensional database and the correspondence relationship between each member of each dimension in the schema used for the aggregation calculation, the end portion to be aggregated into the same member on one layer by the aggregation calculation. By using the end portion that is aggregated into the same member on this one layer as a division point, the aggregation process can be executed in the same processing node, so that a high-speed operation can be realized.

The system calculates the number of divisions in each dimension based on the total number of valid cells obtained by the above-described merging process. In this calculation, the above-described calculation formulas (1) and (2) are used, and a combination of a plurality of division numbers is obtained. the system,
From the combinations of the numbers of divisions, an attempt is made to adopt the blocks in the order of the smallest number of blocks created by the division. The combination of the number of divisions to be adopted is estimated based on the case of division for each dimension to be divided, and the maximum and minimum number of effective cells in the section on the dimension created by the division on each dimension It is determined by determining whether the difference from the value is equal to or less than a certain value.

If the difference between the maximum value and the minimum value of the number of valid cells in the above-mentioned dimension does not fall below a certain value for all combinations, the combination closest to the certain value is adopted. The constant value used for this determination is set as a default value in the system, but can be specified by the user. In the case of the example shown in FIG.
From a1, from a1 to a2, from a2 to a3, calculate the number of valid cells for the four sections after a3, find the difference between the maximum and minimum values, this difference is less than the system default value for the total number of valid cells or If the value is equal to or less than the value specified by the user, a combination of the division numbers is adopted.

At the node 202 shown in FIG. 2, the system calculates the number of effective cells of one or a plurality of blocks created by the division based on the result obtained by the calculation of the number of divisions described above. Is allocated to each storage area. As a procedure for allocating, a method of allocating coordinate values on dimensions to divided blocks on each dimension is used. As a result, as shown in FIG. 5, it is possible to allocate coordinates in a multidimensional space where each block can be uniquely identified. As a result, each block includes (α1, β1), (α1, β2), (α1, β3),
It is identified by coordinates (α2, β1),..., (α4, β3), (α4, β4). These blocks are allocated to each storage area by round robin.

The method of implementing round robin also has a plurality of patterns, but the system calculates all the patterns and selects a combination that makes the distribution of the effective cells most uniform. Then, when performing the allocation, the system avoids concentrating blocks having the same coordinates for each dimension on one storage area. For example, in the case of the example shown in FIG.
It is avoided that the blocks identified by the coordinates of (1, β1), (α1, β2) and (α1, β3) are all allocated to the same storage area. If (α1, β1), (α1,
When the blocks identified by the coordinates of (β2) and (α1, β3) are distributed to the same storage area, when the user requests the sales amount of the entire area related to the product a1, that is, the member on the product α1 When the slice processing related to
This is because the load is concentrated on the processing nodes storing the blocks (α1, β1), (α1, β2), and (α1, β3).
Thereby, parallel processing performance can be improved.

By the data distribution analysis process at the node 202 shown in FIG. 2 described above, the data in the block-divided database is temporarily stored in a file or the like.
After being read, distributed, and merged, it is stored in a plurality of multidimensional data storage areas and used.

Once the data in the block-divided database stored in a file or the like is read,
FIG. 6 shows the configuration of data storage processing for storing data in a plurality of multidimensional data storage areas after merging, and this will be described below.

As shown in FIG. 6, the data storage process includes an input data read / distribution processing node 601 having an input data scan process and a distribution process in charge of a process of reading input data, and receives data from the node 601. A multi-dimensional data storage processing node 602 performs data merging processing and data storage processing.
One or a plurality of data read distribution processing nodes 601 exist. Further, the data read / distribution processing node 601 and the multidimensional data storage processing node described above may be configured as the same node.

The input data scanning process in the data read distribution node 601 reads input data from a file or the like. The data obtained by this input data scanning process is assigned to the cells and passed to the data distribution process. In the data distribution processing, the cell value and the coordinates are sent to the multidimensional data storage processing node 602 in which the storage area determined by the above-described data analysis processing exists.

The node 602 receives the cell value and the coordinate value from the data distribution processing. The merging process in the node 602 checks cell duplication and the like, and makes the duplication an error.
A determination is made such as adding the value of the cell or validating the larger value of the cell. Whether or not to make these determinations is an option that can be specified by the user. The data storage process stores the divided multidimensional data in the multidimensional data storage area in this manner.

The data analysis processing shown in FIG. 2 and the data storage processing shown in FIG. 6 have been described above. FIGS. 7 and 8 show the flow of these processing operations.
And these flows will be described below.

First, the processing operation in the data analysis processing shown in FIG. 2 will be described with reference to FIG.

(1) Input data read processing node 20
1 reads input data by an input data scanning process, assigns coordinate values to valid cells by a conversion process, creates information on coordinates where valid cells exist, and sends coordinates by a sending process to the coordinates where the created valid cells exist. Is transmitted to the data merge processing of the input data analysis processing node 202 (steps 701 to 704).

(2) In the data merging process of the input data analyzing node 202, the received information of the coordinates where the valid cells exist is merged, and the result is passed to the data analyzing process (step 705).

(3) In the data analysis processing, the number of effective cells of each member of each dimension is calculated, and the ratio of the number of divisions of each axis is calculated based on the number of members of each axis of the database.
All combinations of the number of divisions of each axis are calculated (steps 706 to 708).

(4) Next, coordinates are assigned to each block, and the number of effective cells in the block is calculated (step 7).
09, 710).

(5) Blocks are extracted in one dimension direction according to the dimension order in the combination of dimensions of the multidimensional database, and each block is allocated to the data storage area by round robin. This processing is performed by first looping only in one dimension A in the order of nesting, and moving without moving in the other dimensions B,... Directions. The process is repeated by repeating the process of incrementing only by one, looping again in the dimension A direction, and repeating the process for the transition to the dimension B (step 711).

(6) The number of valid cells in the allocated block is added to calculate the number of valid cells in each data storage area. As a result, although not shown, a table of combinations of dimensions of the multidimensional database and the number of data stored in each data storage area of the plurality of data storage areas is created (step 712).

(7) The minimum value is subtracted from the maximum value of the number of valid cells in each storage area to determine the width of the variation in the number of valid cells, and the result is compared with the result of the previous loop processing. Remember. It is checked whether or not all the round robin combinations have been completed. If not, the processing from step 711 is repeated, and the calculation based on the next round robin combination is executed (step 71).
3, 714).

(8) After the calculation by all round robin combinations is completed by repeating the processing of steps 711 to 714, similarly to the processing of step 713, the number of effective cells in each storage area is calculated. After selecting and storing the one with the smallest variation, the processing from step 708 is repeatedly executed for the next combination of the number of divisions for each axis. As a result, a round-robin combination that minimizes the variation with respect to one of the combinations of the number of divisions of each axis obtained in step 708 is obtained, and the minimum result of the comparison for each loop process from step 709 is obtained. Is selected and stored (step 715).

(9) When the processing for the combination of the number of divisions of each axis is completed, the result obtained in step 714
That is, the number of divisions of each axis obtained as a result of the calculation and the distribution method with the smallest variation of effective cells are stored in the dictionary (step 716).

Next, the processing operation in the data storage processing shown in FIG. 6 will be described with reference to FIG.

(1) Input data read / distribution processing node 601 reads input data by input data scan processing, assigns coordinates to cells, and transfers the cells to the data distribution processing. In the distribution processing, the cell value and the coordinates are sent to the data merging processing of the multidimensional data storage processing node 602 in which the storage area determined by the data analysis processing exists (step 8).
01-803).

(2) Multidimensional data storage processing node 602
Receives cell values and coordinate values from the data distribution process.
The merging process in the node 602 checks cell duplication and the like, and makes a judgment such as making the duplication an error, adding a cell value, and validating a larger cell value. Do. Whether or not to make these determinations is an option that can be specified by the user. The data storage process stores the divided multidimensional data in the multidimensional data storage area (steps 804 and 80).
5).

According to the above-described processing according to the embodiment of the present invention, it is possible to realize nearly equal data distribution for each of the storage areas for storing data in the multidimensional database. For this reason, load distribution can be achieved by aggregation calculation and slicing processing, etc. Also, in aggregation calculation and the like, data at close distances in each dimension often exist in the same storage area, so that parallel processing is not performed. The speed can be increased, and the calculation speed can be improved.

Next, with reference to FIG. 11, a description will be given of the configuration of a physical block of a dictionary definition function used in an embodiment of the present invention.

The dictionary definition function 1101 receives, as input, dimension configuration information storing dimension names and member names passed from a system user, and assigns a coordinate value on a computer to each member for each dimension of the multidimensional database to be created. This is a function of allocating a coordinate ID that can be used as, and creating a member name coordinate ID table on the storage device. The dimension configuration information passed from the system user is, for example, as shown in FIG. 12, where the dimension names include “product” and “district”, and the member names include the product name and the district name. Are associated with each other. As already described, the dictionary definition function 11
In FIG. 9, which is an example of the member name coordinate ID table created by No. 01, 0001, 0002,.
.., 0100 coordinate IDs are assigned.

[0058]

As described above, according to the present invention, when data is stored in a multidimensional database, data distribution close to the storage area is realized, and load distribution is performed by aggregation calculation and slicing processing. In addition, even in an aggregate calculation or the like, data at close distances in each dimension often exists in the same storage area, so that the speed of parallel processing can be increased, and the calculation speed is improved. be able to.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating an example of multidimensional data having three dimensions of a product, a district, and sales.

FIG. 2 is a block diagram illustrating a configuration of a data analysis process of the database management system according to the embodiment of the present invention.

FIG. 3 is a diagram illustrating a distribution of the number of effective cells when the multidimensional data shown in FIG. 1 is divided by focusing only on dimension A.

FIG. 4 is a diagram illustrating an example in which the multidimensional data shown in FIG. 1 is divided with respect to dimensions A and B.

FIG. 5 is a diagram illustrating the allocation of coordinates to each block of the divided multidimensional data shown in FIG. 4;

FIG. 6 is a block diagram illustrating a configuration of a data storage process of the database management system according to the embodiment of the present invention.

FIG. 7 is a flowchart illustrating a processing operation of a data analysis process.

FIG. 8 is a flowchart illustrating a processing operation of a data storage process.

FIG. 9 is a diagram illustrating a configuration example of a table for converting members of each dimension into coordinate values usable by a computer in a data analysis process.

FIG. 10 is a diagram showing an example of a table for expressing coordinates where valid cells exist in data analysis processing.

FIG. 11 is a block diagram showing a configuration of a dictionary definition function process of the database management system according to one embodiment of the present invention.

FIG. 12 is a diagram illustrating an example of dimension configuration information.

[Explanation of symbols]

 201 input data read processing node 202 input data analysis processing node 601 input data read distribution node 602 multidimensional data storage processing node

Claims (5)

[Claims] 1. A multidimensional database management system for storing and managing data identified by a combination of members respectively associated with a plurality of dimensions, comprising one or a plurality of processing nodes for processing a database. The node assigns a series of coordinate values to each of the members of each of the plurality of dimensions, and sets a set of coordinate values of the dimension members as cell coordinates of data included in the set,
According to the number of storage areas of the multidimensional database, data is divided so that the number of effective cells in each dimension is equal, thereby creating one or more blocks containing effective cells, and dividing these blocks into data. A multidimensional database management system, wherein a multidimensional database is divided so as to be arranged in a storage area. 2. The multidimensional database management according to claim 1, wherein the dimension having a large number of members for each dimension is divided by a large number of divisions, and the dimension having a small number of members is divided by a small number of divisions. system. 3. When the number of storage areas of the multidimensional database is large, the data is divided so that the number of the divided blocks is large.
2. The multidimensional database management system according to claim 1, wherein when the number of storage areas is small, the division is performed so as to reduce the number of divided blocks. 4. A coordinate value on each dimension is assigned to each of the divided blocks, and each block is uniquely specified by using a set of coordinate values of the block on the dimension as block coordinates of each block, A set of one or a plurality of blocks is arranged in a storage area, the total number of valid cells distributed to each storage area is calculated, and the difference between the maximum value and the minimum value of the total value among the storage areas is calculated. 2. The multidimensional database management system according to claim 1, wherein a combination of blocks that minimizes is arranged in each storage area. 5. The multidimensional database management system according to claim 1, wherein blocks in the same division range on each dimension of the divided blocks are arranged in different storage areas.