JP2000339140A - Database retrieval device and program recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize a high speed processing by deciding the number of divisions on a memory resource used for a previous sorting processing from the processing content of the previous sorting processing and the processing content of a later sorting processing using the processing result. SOLUTION: A decision means 11 decides the number of divisions on a memory resource in a previous sorting processing so that the CPU use quantity of the previous sorting processing is reduced when the previous sorting processing is a CPU neck and a later sorting processing using the processing result of the sorting processing is an I/O neck at the time of analyzing a data base retrieval language. When the previous sorting processing is the I/O neck and the later sorting processing using the processing result of the sorting processing is the CPU neck, the number of divisions on the memory resource in the previous sorting processing is decided so that the CPU use quantity of the later sorting processing is reduced. The use of I/O and CPU can be balanced and a high speed processing can be executed.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a database search device for processing a database search language and a program recording medium for storing a program used to realize the device, and more particularly to a database search device for realizing high-speed processing. And a program recording medium in which a program used for realizing the device is stored.

In the processing of an SQL statement, a result is extracted through a plurality of steps using data reading from a secondary storage, data processing on a memory, and saving intermediate results to a secondary storage. Will be. In order to speed up data processing using a database, it is necessary to build a technology for reducing the time required for processing the SQL statement.

[0003]

2. Description of the Related Art In the processing of an SQL statement, the result is obtained through a plurality of steps using reading of data (record) from a secondary storage, processing of data on a memory, and saving of intermediate results to a secondary storage. Is extracted.

For example, the SQL sentence shown in FIG.
The data of the columns C1 and C2 are read from the table T1, and are arranged and output according to the order of the data of the column C1.

[0005] The processing procedure of this SQL sentence is shown in FIG.
As shown in (1), first, in the first step, the data of Table T1 is read from the database, PUT for sorting, and then, in the next step, GET the data from the sort and return it to the application. become.

The SQL sentence shown in FIG. 22A reads data (eg, name and address) of columns C1 and C2 from table T1, and reads data (eg, telephone number) of column C2 from table T2. The loaded table T
In the data of the columns C1 and C2, the read table T
2 and the data of column C2 are combined (T1.C1 = T2.C
1 (e.g., those having the same name) are combined, and arranged and output according to the order of the data in column C1 of table T1.

The procedure for processing this SQL statement is shown in FIG.
As shown in (1), first, in the first step, the data of table T1 is read from the database and PUT is performed for sort 1, and in the next step, the data of table T2 is read from the database and PUT is performed for sort 2. In the process, GET and JOIN the data from sorts 1 and 2 and PUT to sort 3,
In the next process, a four-stage process of GETting the data from the sort 3 and returning it to the application is performed.

The SQL statement shown in FIG. 23A reads data (for example, branch name and sales) in columns C1 and C2 from table T1, and performs grooving with the data in C2 (for example, branch name). The total of the data (for example, sales) of C1 is obtained, and defined as a table DT1,
The data (for example, the branch name and the total sales) of the columns C1 and C2 are read from the table DT1, and the data (for example, the address) of the column C2 is read from the table T2, and the data of the columns C1 and C2 of the read table DT1 are added to the data. Combine the read data of column C2 of table T2 (DT1.C
2 = T2. C1 (for example, the same branch name) is combined and output.

The processing procedure of this SQL statement is shown in FIG.
As shown in (1), first, in the first step, the data of Table T1 is read from the database and put to the work, and in the next step, the data is GET from the work, the data is aggregated, and the result is put to the sort 1 and put to the sort 1. In the next step, the data in Table T2 is read from the database and put into Sort 2
In the next step, get data from sorts 1 and 2 and JO
This involves a four-step process of returning the application to the application.

As described above, in the processing of an SQL statement, the result of reading the data from the secondary storage, processing the data on the memory, and saving the intermediate result to the secondary storage are processed through a plurality of steps. Extraction is performed, but in each of these steps, the data (record) processed in the previous step is carried over to the next step. Therefore, SQ
In the case where the processing of the L sentence includes a sorting process, there is a characteristic that the whole process must be input to the sorting process before the next process using the result can be performed.

In the prior art, considering this characteristic, SQ
In the case where the processing of the L sentence includes a sorting process, as shown in FIG. 24, a configuration is adopted in which data is sorted while an index is set in all areas of the memory used for the sorting process. While reading the data sorted using the index, processing was performed to enter the next step using the result.

[0012]

However, according to such a conventional technique, there is a problem that the maximum processing performance cannot be obtained.

That is, as shown in FIG. 25A, the previous sorting process (the sorting process and the process running at the same time) uses more I / O than the CPU,
On the other hand, if the later sorting process (the sorting process and the process running at the same time) uses more CPU than I / O, the processing time is reduced due to the poor use balance between I / O and CPU. There is a problem that it takes a long time.

As shown in FIG. 25 (b), the previous sorting process (the sorting process and the process running simultaneously) uses more CPU than the I / O and performs the later sorting process. (Steps of the sorting process and the process running at the same time) use more I / O than the CPU, the processing time is long because the balance between the I / O and the CPU is poor. There is a problem.

The present invention has been made in view of such circumstances, and provides a new database search device for realizing high-speed processing and a new program recording medium for storing a program used for realizing the device. To provide.

[0016]

FIG. 1 shows the principle configuration of the present invention.

In FIG. 1, reference numeral 1 denotes a database search device equipped with the present invention, and a database search language (SQL sentence).
2 is a definition file provided in the database search device 1, which manages definition information (column attributes, table storage structure, number of records, data distribution, etc.) necessary for creating a processing procedure. Reference numeral 3 denotes a database for managing records to be searched by the database search device 1.

The database search device 1 of the present invention comprises a processing procedure creating means 10, a determining means 11, and an executing means 12
The definition file 2 included in the database search device 1 of the present invention includes a table means 20 for realizing the present invention.

The processing procedure creating means 10 creates a processing procedure by analyzing a database search language and determining a schedule for accessing from which table.

The deciding means 11 is developed in the processing procedure creating means 10 to analyze the database search language.
The number of divisions of the memory resource to be used for the previous sort process is determined from the process contents of the previous sort process and the process contents of the sort process after using the processing result of the sort process.

The execution means 12 processes the database search language in accordance with the processing procedure created by the processing procedure creation means 10. At this time, the execution means 12 divides the memory resources according to the number of divisions determined by the determination means 11, and The sort processing specified by the database search language is executed in units of memory resources.

When the SQL control means 13 receives the database search language issued by the application, it creates a processing procedure by calling the processing procedure creation means 10 and calls the execution means 12 by designating the processing procedure. Process.

The table means 20 stores the memory resources in accordance with a matrix structure in which a search key is a search key that may be a processing content of the first sort processing and a table that may be a processing content of the second sort processing. Manages the number of divisions.

Here, the database search device 1 of the present invention
The functions of the program are realized by a program. This program is stored in a floppy disk or the like, or is stored in a disk such as a server. The present invention is realized by operating in.

In the database search apparatus 1 of the present invention configured as described above, when analyzing the database search language, the determination means 11 determines that the previous sort processing is a CPU bottleneck and the processing result of the sort processing is If the sort processing after using is an I / O bottleneck, the number of memory resource divisions in the previous sort processing is determined so that the CPU usage in the previous sort processing is reduced, If the previous sort process is an I / O bottleneck and the later sort process using the result of the sort process is a CPU bottleneck, the CPU usage of the later sort process is reduced. Thus, the number of divisions of the memory resource in the previous sorting process is determined.

Here, according to a matrix structure using search keys that may be the processing contents of the first sort processing and those that may be the processing contents of the second sort processing, such a form is used. When the table means 20 for managing the number of divisions of the memory resource is provided, the determining means 11 looks up the table means 20 so that
The number of divisions of the memory resource in the above sorting process is determined.

In response to this determination processing, the execution means 12
The memory resource is divided according to the number of divisions determined by the determining means 11, and a sorting process specified by a database search language is executed in units of the divided memory resources.

As shown in FIG. 2, a process A is applied to a database search language to sort data (records) by PU.
Then, it is assumed that the description describes that a process of performing a process B by GETting data (records) from sorting after that is performed.

Here, as shown in the figure, when the division number of the memory resource is small, the CPU usage of the process A is large, and when the division number of the memory resource is large, the CPU usage of the process A is small. Let it be smaller. On the other hand, it is assumed that when the number of divisions of the memory resource is small, the CPU usage of the process B is small, and when the number of divisions of the memory resource is large, the usage of the CPU of the process B is large.

For example, when the process A is a process of reading data (records) from a database, if the number of divisions of the memory resource is small, a large index must be provided and the number of searches increases, thereby increasing the number of searches. Use amount becomes large. Conversely, when the number of memory resource divisions is large, it is sufficient to set a small index.
The PU usage is reduced.

At this time, in the process B for performing the process on the data sorted by the process A, when the number of divisions of the memory resource is large, a set of data sorted by a small index must be merge-sorted. CPU usage increases. Conversely, when the number of divisions of the memory resource is small, it is sufficient to merge-sort a set of data sorted by a large index, so that the CPU usage is reduced.

The number of divisions of the memory resource and C of processes A and B
In consideration of the relationship established with the PU usage, as shown in the upper part of FIG. 3, the previous sorting process (the sorting process and the process A running at the same time) is an I / O bottleneck. If the subsequent sorting process (the sorting process and the process B running at the same time) is a CPU bottleneck, the determining means 11 sets a small value as the number of memory resource divisions in the preceding sorting process. By deciding, Figure 3
As shown in the lower part, the CPU usage in the sorting process after the CPU bottleneck is reduced, thereby
The overall processing time is reduced.

On the other hand, as shown in the upper part of FIG. 4, the previous sorting process (the sorting process and the process A running at the same time) is a CPU bottleneck, and the later sorting process (the sorting process and the If the process B (running at the same time) is an I / O bottleneck, the determining means 11
By determining a large value as the number of divisions of the memory resource in the previous sort processing, the CP shown in the lower part of FIG.
The amount of CPU used in the sorting process, which is a U-neck, is reduced, thereby shortening the overall processing time.

The shortening of the processing time will be specifically described.
If you do, "A I / O used by process A for "IO"
Quantity, "SP I / O using sort PUT for "IO"
Quantity, "B “IO” is the I / O amount used by process B, “SG”
"IO" is the amount of I / O used by sort GET, "A CP
"U" is the amount of CPU used by process A, "SP" CPU "
CPU amount used by sort PUT, "B CPU
Amount of CPU used by the processing B, "SG Sort CPU G
CPU amount used by ET, "MAX IO ”in unit time
I / O amount that can be processed per unit, "MAX CPU ”
If the CPU amount can be processed per unit time,
The input time of the sorting process and the
The output time is as shown in FIG. 5 (a) / FIG. 6 (a).
You.

As shown in the upper part of FIG. 3, when the previous sorting process is an I / O bottleneck and the later sorting process is a CPU bottleneck, the input time and the output The time is as shown in FIG. 5B, and the total processing time is the sum of the times. At this time, by adjusting the number of divisions of the memory resource in the previous sort processing, as shown in the lower part of FIG. "CPU" is moved to the sort PUT, thereby shortening and improving the output time as shown in FIG. 5 (c).

On the other hand, as shown in the upper part of FIG. 4, when the previous sorting process is a CPU bottleneck and the later sorting process is an I / O bottleneck, this input time And the output time are as shown in FIG.
The total processing time is this sum. At this time, by adjusting the number of divisions of the memory resources in the previous sort processing,
As shown in the lower part of FIG. 4, a part of the CPU amount at the time of the sort PUT “Y” "CPU" is moved to the sort GET, thereby shortening and improving the input time as shown in FIG. 6 (c).

As described above, in the database search apparatus 1 of the present invention, when adopting the configuration for processing the database search language, the processing contents of the previous sorting processing and the processing results after using the processing results of the sorting processing are used. The number of memory resources to be used for the preceding sort processing is determined from the sort processing contents, and the memory resources are divided accordingly, and the sort processing specified in the database search language in units of the divided memory resources By executing the above, high-speed processing is realized by balancing the use of the I / O and the CPU.

[0038]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail according to embodiments.

FIG. 7 shows an embodiment of the database search apparatus 1 provided with the present invention. In the figure, the same components as those described in FIG. 1 are denoted by the same reference numerals.

The database search device 1 of the present invention according to this embodiment includes a processing procedure control table creation program 100,
An execution program 101, a plurality of search processing components 102 called from the execution program 101, and an SQL control program 103 are provided.
Includes a division number determination table 200 to implement the present invention.

Here, the database search device 1 of the present invention
Will be installed via a floppy disk or a line.

As described with reference to FIG. 1, in the database search apparatus 1 of the present invention, the sort processing specified by the SQL statement is a CPU bottleneck, and the sort processing after using the processing result of the sort processing is performed. Is an I / O bottleneck, the number of memory divisions in the previous sort processing is determined so that the CPU usage in the previous sort processing is reduced. Is the I / O bottleneck, and the later sort process using the processing result of the sort process is the CPU bottleneck, so that the CPU usage of the later sort process is reduced. The number of divisions of the memory in the previous sort processing is determined, the memory is divided according to the number of divisions, and the sort processing is executed, thereby reducing the overall processing time.

To implement the present invention, a new definition file 2
Is used for determining the number of divisions of the memory.

FIG. 8 shows an embodiment of the division number determination table 200.

The division number determination table 20 according to this embodiment
0 is a matrix having search keys of a process that can be a process (process A) that runs simultaneously with the preceding sort process and a process that can be a process (process B) that runs simultaneously with the subsequent sort process. According to the structure, a configuration is employed in which the number of divisions of the memory in the previous sort processing is managed.

For example, a process A running at the same time as the previous sorting process is a JOIN process, and a process B running at the same time as the sorting process after using the processing result of the sorting process is GROUP.
In the case of the BY totaling process, it manages that the number of divisions of the memory in the previous sorting process becomes “50”.

Next, the relationship between the number of divisions of the memory used for the sorting process and the number of comparisons required for storing / retrieving records will be described.

In the sort processing at the time of PUT, the data is sorted and stored in the memory on the memory. By having an index, a record can be searched in a binary tree. At this time, the number of comparisons required to store one record is log ₂ (the number of records already stored).

Therefore, if the number of records is N,
If the memory is not divided, the number of comparisons needed to store all _{^{records, Σ N log 2 x ≒ ∫}} N logxdx = (1 / log2) × (NlogN-N) = Nlog 2 N according to the equation that -N / log2, it becomes _{N × (log 2 N-1} / log2).

As can be seen from this equation, the larger the record storage area, the higher the record search cost and the larger the CPU amount required at the time of PUT.

Therefore, it is conceivable that an upper limit is set for the size of the record storage area, and when the upper limit is exceeded, the search cost is not increased by using a new record storage area. As shown in FIG. 9A, it is assumed that k record storage areas are prepared by dividing the memory into k pieces. The number of records stored in one record storage area is “N / k”.

Therefore, the number of comparisons required to store all records in one record storage area is, as can be seen from the above formula, (N / k) × [log ₂ (N / k) − (1 / log
2)], the number of comparisons required to store all records in all k record storage areas is k times the number N × [log ₂ (N / k) − (1 / log ₂ ) )].

On the other hand, if the memory is not divided, it is not necessary to perform the merge sort at the time of GET, but if the memory is divided, as shown in FIG. 9B, it is necessary to perform the merge sort at the time of GET. is there.

In this case, when there are k record storage areas, the number of comparisons required to retrieve one record is “log ₂ k”, so that it is necessary to retrieve all records. The number of comparisons is N × log ₂ k.

As described above, when the memory is divided into k pieces, the CPU cost required at the time of PUT is N × [log ₂ (N / k) − (1 / log 2)] × (key comparison cost) And the CPU cost required at the time of GET is N × log ₂ k × (key comparison cost).

The sum of these two CPU costs does not depend on k, as can be seen from the equation: N × [log ₂ N− (1 / log 2)] × (key comparison cost). Therefore, even if the memory used for the sorting process is divided, the entire C
PU costs do not increase.

However, when the memory used for the sorting process is divided, as shown in FIG. 10, when the division number k is large, the CPU cost at the time of PUT is reduced and the CPU cost at the time of GET is increased. On the other hand, when the division number k is small, the CPU cost at the time of PUT increases and the CPU cost at the time of GET decreases.

From now on, when the sort processing of the PUT is an I / O bottleneck and the GET sort processing using the processing result of the PUT sort is a CPU bottleneck, the memory used for the PUT sort processing is By reducing the number k of divisions, the overall processing time can be shortened as shown in FIG.
When a PU neck is used and a GET sort process using the processing result of the PUT sort process is an I / O neck, the number k of divisions of the memory used for the PUT sort process is increased. As shown in FIG. 4, the overall processing time can be reduced.

Focusing on this point, the database search device 1 of the present invention realizes high-speed processing specified by an SQL statement.

FIG. 11 shows an example of the SQL for realizing the present invention.
One embodiment of the processing flow executed by the control program 103, FIG. 12 shows one embodiment of the processing flow executed by the processing procedure control table creation program 100 to realize the present invention, and FIG. 13 realizes the present invention. An embodiment of the processing flow executed by the execution program 101 for this purpose is shown in FIG. Next, the present invention will be described in detail according to these processing flows.

When the application issues an SQL statement, the SQL control program 103 first receives the SQL statement issued by the application in step 1 as shown in the processing flow of FIG. Subsequently, in step 2, the SQL statement is specified, and the processing procedure control table creation program 100 is called to create a processing procedure control table.

Subsequently, in step 3, the execution program 101 is called by designating the created processing procedure control table, and the processing result of the SQL statement is received from the execution program 101. Subsequently, in step 4, the processing result is returned to the application, and the processing ends.

Thus, the SQL control program 10
3, the processing procedure control table creation program 100 first executes the SQL while referring to the definition information of the table stored in the definition file 20 in step 1 as shown in the processing flow of FIG. The SQL statement received from the control program 103 is parsed and semantically analyzed.

Subsequently, in step 2, the processing procedure defined by the SQL statement is determined in an optimal form using the analysis result and referring to the optimization information of the table stored in the definition file 20. That is, FIG. 21 (b), FIG. 22 (b), FIG.
The processing procedure as shown in FIG. 3B is determined.

Subsequently, in step 3, it is determined whether or not the determined processing procedure includes a sort process. When it is determined that the sort process is performed, the process proceeds to step 4 and runs simultaneously with the previous sort process. Processing (processing A) and processing that runs simultaneously with the sorting processing after using the processing result (processing B)
Extract the combination with

Subsequently, in step 5, the definition file 20
Reads the division number determination table 200 from
By using the combination of the processes extracted in (1) as a search key and indexing the division number table 200, the number of divisions of the memory in the previous sorting process is determined.

Subsequently, in step 6, a processing procedure control table describing the processing procedure determined in step 2 and the number of memory partitions determined in step 5 (determined in association with the sorting process) is created. On the other hand, when it is determined in step 3 that the processing procedure determined in step 2 has no sort processing, a processing procedure control table describing only the determined processing procedure is created.

Then, in the following step 7, the created processing procedure control table is passed to the SQL control program 103, and the processing is terminated.

As described above, when called from the SQL control program 103, the processing procedure control table creation program 100 determines the processing procedure defined by the SQL statement and associates it with the sort processing included in the processing procedure. Determine the number of memory partitions. Then, a processing procedure control table that describes them is created, and is processed so as to be passed to the SQL control program 103.

In the prior art, only the processing procedure defined by the SQL statement is determined, and this is determined by the SQL control program 103.
In contrast, the processing procedure control table creation program 100 that implements the present invention determines the number of memory divisions in association with the sort processing included in the processing procedure, and sends it to the SQL control program 103. It is processed to be passed.

As described above, this memory division number determination processing is performed by indexing the division number determination table 200 having a table structure as shown in FIG.

For example, when the SQL statement shown in FIG. 21A is received from the SQL control program 103, the process that runs simultaneously with the previous sort process (PUT side) is a record read process from the database, and the process result is read. Since the process running at the same time as the sort process (GET side) after use is return to the application, “30” is determined as the number of memory divisions in the previous sort process (PUT side).

This is because the process of reading records from the database uses a lot of I / O and hardly uses the CPU, whereas the process of returning to the application uses I / O.
Uses neither / O nor CPU. From now on, CPU
Are both equally open, so if anything,
The reason is that the CPU is used at the same time as I / O by reducing the number of memory divisions and bringing the use of the CPU to the PUT side.

When the SQL statement shown in FIG. 22A is received from the SQL control program 103, the process that runs simultaneously with the previous sort 1 and 2 processes (PUT side) is the process of reading records from the database. Since the process that runs simultaneously with the sort 3 process (GET side) after using the process result is the JOIN process, the sort 1 process (P
“1” is determined as the number of memory divisions in the UT side), and “1” is determined as the number of memory divisions in the previous sort 2 processing (PUT side).

This is because the process of reading records from a database uses a lot of I / O and uses little CPU, whereas the JOIN process uses a lot of CPU and uses I / O.
O is rarely used. From now on, I / O on PUT side
There is a high likelihood of becoming a bottleneck, and the GET side is likely to become a CPU bottleneck. The reason for this is to reduce the number of memory divisions and actively use the CPU to the PUT side. is there.

The process that runs simultaneously with the previous sort 3 process (PUT side) is the JOIN process, and the process that runs at the same time as the sort process (GET side) using the processing result is return to the application. , Sort 3 ahead
“70” as the memory division number in the processing (PUT side)
To determine.

This is because the JOIN process uses a lot of CPU and uses little I / O, whereas the return process to the application uses little I / O and CPU. From now on, the PUT side is likely to be a CPU bottleneck, while the GET side is not sure whether it will be an I / O bottleneck. There is a reason to take it to the GET side.

Also, from the SQL control program 103
When receiving the SQL statement shown in FIG.
GROU is a process that runs simultaneously with the sort 1 process (PUT side)
P This is the BY totaling process, and the source
The process that runs at the same time as the
Because of this, in the previous sort 1 processing (PUT side)
"50" is determined as the memory division number.

This is a GROUP BY counting process is also J
OIN processing also uses a lot of CPU and uses little I / O. From now on, there is a high possibility that both the PUT side and the GET side become CPU bottlenecks. Therefore, by using an average value as the number of divisions of the memory, the PUT side and the GET side can average the CPs.
There is a reason for using U.

The process that runs simultaneously with the previous sort 2 process (PUT side) is the process of reading records from the database, and the subsequent sort process (GET
Side) Since the process running at the same time is the JOIN process,
“1” is determined as the memory division number in the sort 2 processing (PUT side).

This is because the process of reading records from the database uses a lot of I / O and uses little CPU, whereas the JOIN process uses a lot of CPU and uses I / O.
O is rarely used. From now on, I / O on PUT side
There is a high likelihood of becoming a bottleneck, and the GET side is likely to become a CPU bottleneck. The reason for this is to reduce the number of memory divisions and actively use the CPU to the PUT side. is there.

A processing procedure control table is created by the processing procedure control table creation program 100 in accordance with the processing flow of FIG. 12, and is called by specifying the processing procedure control table by the SQL control program 103 in accordance with the processing flow of FIG. First, as shown in the processing flow of FIG. 13, the execution program 101 first receives a processing procedure control table from the SQL control program 103 in step 1.

Subsequently, in step 2, one of the processing procedures specified in the received processing procedure control table is selected in accordance with the order, and in step 3, whether or not all the processing procedures have been selected is determined. When it is determined that all of the processing procedures have been selected, the process proceeds to step 4 where the processing result obtained in the processing to be described hereinafter is expressed in SQL.
The processing is passed to the control program 103 and the processing ends.

On the other hand, when it is determined in step 3 that the selection of all the processing procedures has not been completed, that is, when it is determined in step 2 that the processing procedure has been selected, the process proceeds to step 5, and It is determined whether or not the selected processing procedure is a sort processing.

When it is determined by this determination that the process is a sort process, the process proceeds to step 6, where the number of memory divisions specified in the processing procedure control table is specified, and is used for the sort processing according to the specified number of memory divisions. Sorting is performed by calling the corresponding search processing component 102 while dividing the memory, and then the process returns to step 2.

On the other hand, when it is determined in step 5 that the selected processing procedure is not the sort processing, step 7
To execute the processing designated by calling the corresponding search processing component 102, and then execute step 2
Go back to

In this way, the execution program 101
Divides the memory used for the sorting process according to the number of memory divisions specified in the processing procedure control table created by the processing procedure control table creation program 100, and executes the sorting process using the divided memory. In this way, the overall processing time is reduced.

If this execution process is specifically described with reference to FIGS. 14 to 17, when PUT is performed to sort the records, as shown in FIG. While creating an index, records are stored in a record storage area (sort order is guaranteed by the index).

When the record storage area becomes full by this storage processing, the records stored in the record storage area are saved in the record save area as shown in FIG. At this time, the index is not required by replacing the records in the sort order.

Thereafter, as shown in FIG. 14C, the records are stored in the memory while the index is created again according to the size of the record storage area, and when the record storage area becomes full, FIG. As shown in (a),
Records stored in the record storage area are saved in a new record save area. By repeating this process, when the memory becomes full, FIG.
As shown in (b), records are saved in the secondary storage.

On the other hand, when a GET is performed on the sorted records, the records are read from the beginning in the saved unit and are merge-sorted. When the record storage area is not full, FIG. As shown, it is only necessary to read from the index.

When only the memory is used and the secondary storage is not used, the records stored in the record storage area / record save area developed on the memory are stored as shown in FIG. Merge sort will be performed. Furthermore, when secondary storage is used, FIG.
As shown in FIG. 7, merging and sorting are performed in a form including that. If an area to be read from the secondary storage is not left in the memory, the area stored in the record storage area is secured by saving the record to the secondary storage.

Thus, the execution program 101
Divides the memory used for the sorting process according to the number of memory divisions specified in the processing procedure control table created by the processing procedure control table creation program 100, and executes the sorting process using the divided memory. In this way, the overall processing time is reduced.

The division number determination table 200 shown in FIG. 8 employs a configuration in which the number of memory divisions is managed in accordance with the matrix structure. However, the qualitative management of CPU and I / O usage for each process is shown in FIG. A division number determination table 200 as shown in FIG. 18B for preparing a division number determination table 200 as shown in FIG. 18A and quantitatively managing the amounts of CPU and I / O used for each process is shown in FIG. It is also possible to adopt a method in which the processing procedure control table creation program 100 prepares and determines the number of memory divisions programmatically using this table data. Here, FIG.
When the number-of-divisions determination table 200 shown in FIG. 1 is prepared, it is possible to adopt a configuration in which the CPU and I / O usages are quantitatively managed using a function having variables such as the number of records and data distribution. is there.

When such a division number determination table 200 is prepared, the processing procedure control table creation program 10
0 extracts a combination of a process running at the same time as the previous sorting process and a process running at the same time as the sorting process using the processing result. For example, according to the process flow shown in FIGS. The number of memory divisions will be determined.

That is, the processing procedure control table creation program 100 extracts the combination of the processing that runs at the same time as the previous sorting processing and the processing that runs at the same time as the sorting processing after using the processing result. As shown in the processing flow, first, in step 1, by indexing the division number determination table 200, table data of a process running at the same time as the previous sorting process is obtained, and then in step 2
Then, by indexing the division number determination table 200, table data of a process that runs simultaneously with the subsequent sorting process is obtained.

Then, in step 3, it is determined whether or not there is a high possibility that the previous sort process is an I / O bottleneck and the later sort process is a CPU bottleneck according to the obtained table data. When it is determined that there is a high possibility, the process proceeds to step 4, where a small value is determined as the number of memory divisions in the previous sort processing, so that the CPU amount of the previous sort processing is increased and C
It is set to reduce the PU amount.

On the other hand, in step 3, the previous sort
If it is determined that there is no high possibility that the later sort process will be a CPU bottleneck due to the / O bottleneck, the process proceeds to step 5, where the previous sort process is a CPU bottleneck and the later sort process is an I / O bottleneck. When it is determined whether or not the possibility is high, and when it is determined that the possibility is high, the process proceeds to step 6, where a large value is determined as the number of memory divisions in the previous sort processing, whereby the previous sort is performed. Processing C
The setting is made so that the PU amount is reduced and the CPU amount for the later sorting process is increased.

On the other hand, in step 5, the previous sort
When it is determined that there is not a high possibility that the later sort process becomes an I / O neck due to the PU neck, the process proceeds to step 7 to determine whether or not the previous sort process uses a lot of I / O. When you decide to use a lot,
Proceeding to step 8, a smaller value is determined as the number of memory divisions in the previous sort processing, so that the CPU amount for the previous sort processing is set to be increased.

On the other hand, in step 7, the previous sorting
When it is determined that a large amount of / O is not used, the process proceeds to step 9, where it is determined whether or not the preceding sorting process uses a large amount of CPU. When it is determined that a large amount of / O is used, the process proceeds to step 10. By setting a large value as the number of memory divisions in the previous sort processing, the CPU amount for the previous sort processing is set to be reduced.

On the other hand, at step 9, the previous sorting
When it is determined that the PU is not used much, the process proceeds to step 11, and it is determined whether or not the subsequent sorting process uses a large amount of I / O. When it is determined that the PU is used a lot, the process proceeds to step 12. Then, by setting a larger value as the number of memory divisions in the previous sort process, the CPU amount for the later sort process is set to be increased.

On the other hand, if it is determined in step 11 that the subsequent sorting process does not use a lot of I / O, the process proceeds to step 13 to determine whether or not the subsequent sorting process uses a lot of CPU. When it is determined that the memory is to be used a lot, the process proceeds to step 14, where a smaller value is set as the number of memory divisions in the previous sort processing,
The setting is made so as to reduce the CPU amount for the later sorting process.

On the other hand, if it is determined in step 13 that the subsequent sorting process does not use a lot of CPU, the process proceeds to step 15 and an intermediate value is set as the number of memory divisions in the preceding sorting process.

In this way, the CPU and I
FIG. 18 (a) for qualitatively managing the amount of I / O used is prepared, and the CPU and I / O used for each process are quantitatively managed.
It is also possible to adopt a method in which a division number determination table 200 as shown in (b) is prepared, and the processing procedure control table creation program 100 determines the memory division number using this table data.

Although not described so far, the present invention is a system suitable for parallelization.

This is because, in the PUT phase, the whole record is divided into a plurality of parts and sorted, so that the structure itself is suitable for parallelization.

If records are evacuated in the full record storage area in the order in which they were sorted, no index is required for the evacuated records. This is because it is configured to be reusable.

As a result, in addition to the fact that the index area can be reduced, the saved records are sorted in the order in which they are sorted, so that they can be processed in the same format on secondary storage or in communication processing. is there. For example, in the merge sort process, it is only necessary to process the records in order from the first record, and the communication process and the process of reading from the secondary storage can be performed efficiently.

In the case of parallel processing, the processing executed by each processor to implement the present invention is basically the same as the processing executed by the above-described single-structure processor to implement the present invention. The only difference is that the merge-sorted data of each processor needs to be merge-sorted by all processors.

In the merge sort, if each processor cannot access the memory of another processor, a communication process is added. However, for the above-described reason, the saved record is communicated as it is. Since it is only necessary to go, this communication processing can be easily achieved.

[0111]

As described above, in the present invention, when adopting a configuration for processing a database search language,
The number of divisions of the memory resources used for the sort processing is determined from the contents of the previous sort processing and the contents of the sort processing after using the processing result of the sort processing, and the memory resources are divided accordingly. By executing the sort processing specified by the database search language in units of the divided memory resources, the use balance between the I / O and the CPU can be achieved, thereby realizing high-speed processing. become able to.

[Brief description of the drawings]

FIG. 1 is a principle configuration diagram of the present invention.

FIG. 2 is an explanatory diagram of the present invention.

FIG. 3 is an explanatory diagram of the present invention.

FIG. 4 is an explanatory diagram of the present invention.

FIG. 5 is an explanatory diagram of the present invention.

FIG. 6 is an explanatory diagram of the present invention.

FIG. 7 is an embodiment of the present invention.

FIG. 8 is an example of a division number determination table.

FIG. 9 is an explanatory diagram of the number of comparisons required for storing / retrieving a record.

FIG. 10 is an explanatory diagram of a relationship between a memory division number and a CPU cost.

FIG. 11 is a processing flow of an SQL control program.

FIG. 12 is a processing flow of a processing procedure control table creation program.

FIG. 13 is a processing flow of an execution program.

FIG. 14 is an explanatory diagram of the present invention.

FIG. 15 is an explanatory diagram of the present invention.

FIG. 16 is an explanatory diagram of the present invention.

FIG. 17 is an explanatory diagram of the present invention.

FIG. 18 is another example of the division number determination table.

FIG. 19 is a processing flow of a processing procedure control table creation program.

FIG. 20 is a processing flow of a processing procedure control table creation program.

FIG. 21 is an explanatory diagram of a processing procedure of an SQL sentence.

FIG. 22 is an explanatory diagram of a processing procedure of an SQL sentence.

FIG. 23 is an explanatory diagram of a processing procedure of an SQL sentence.

FIG. 24 is an explanatory diagram of a conventional technique.

FIG. 25 is an explanatory diagram of a conventional technique.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Database search apparatus 2 Definition file 3 Database 10 Processing procedure creation means 11 Decision means 12 Execution means 13 SQL control means 20 Table means

Claims (5)

[Claims] In a database search device that processes a database search language, when analyzing a database search language, the processing contents of a previous sort processing and the processing contents of a sort processing after using a processing result of the sort processing are described. Determining means for determining the number of divisions of memory resources to be used for the sort processing, and dividing the memory resources according to the number of divisions determined by the determining means, and specifying the divided memory resources as a unit in a database search language. Executing means for executing a sorting process to be performed. 2. The database search device according to claim 1, wherein the determination unit determines the number of divisions of the memory resource according to whether the two sort processes use a lot of CPU or I / O. A database search device characterized by the following. 3. The database search device according to claim 2, wherein the determining means determines whether the first sort process is a CPU bottleneck and the second sort process is an I / O bottleneck. The number of divisions of the memory resources is determined so that the CPU usage of the sort processing is reduced. If the previous sort processing is an I / O bottleneck and the later sort processing is a CPU bottleneck, A database search device characterized in that the number of divisions of memory resources is determined so that the amount of CPU used in later sorting processing is reduced. 4. The database search apparatus according to claim 1, wherein the database search apparatus has a possibility that the contents of the sort processing may be performed.
A table unit for managing the number of divisions of the memory resource in accordance with a matrix structure using a search key that may be the processing content of the later sorting process; A database search device characterized by determining the number of resource divisions. 5. A program recording medium for storing a program used for realizing a database search device for processing a database search language, wherein when analyzing the database search language, the processing contents of the above-mentioned sort processing, From the processing contents of the sort processing after using the processing result of the sort processing, a determination processing for determining the number of divisions of the memory resources to be used for the preceding sort processing, and dividing the memory resources according to the number of divisions determined by the above determination processing A program for causing a computer to execute an execution process of executing a sort process specified by a database search language in units of the divided memory resources.