JPH05204728A - Intelligent fetch system for data base retrieval - Google Patents

Abstract

PURPOSE:To shorten response time by decreasing the number of times for commanding retrieval in the case of retrieving a data base. CONSTITUTION:A retrieving command from a data base user 1 is analyzed by a retrieving command analyzing function 2 and a physical storage position, where an objective record exists, is decided by an input block decider 2. It is discriminated whether the decided block position is the same as the storage position calculated last time or not and as the result, the rate of reading as serial blocks in the past is registered on an integral coefficient/input history table 12. Then, the blocks having high possibility to be serially read are transferred to a main storage device 10 as the serial blocks according to one retrieving command. Therefore, when extracting plural blocks from the data base to the main storage device 10, the number of times for instructing inputs can be reduced.

Description

Detailed Description of the Invention

[0001]

[Industrial application] Searching for a database
The present invention relates to an intelligent fetch method of database search that can search a database at a higher speed.

[0002]

2. Description of the Related Art Conventionally, when the desired data is read from the database, only the relevant data is read from the magnetic disk, or the syntax of one data fetch command is used in advance. All data (or partially collected data) is read out. In the former case, an input operation is always involved, and in the latter case, a certain number of plural data are read out by one input operation.

Incidentally, this type of technique is disclosed in, for example, JP-A-57-48136, JP-A-58-3032, and JP-A-58-56145.

[0004]

SUMMARY OF THE INVENTION In the above conventional technique,
It has been difficult to efficiently read out a data group to be searched by a data fetching command that is independent in terms of time.

An object of the present invention is to use the past input history to determine the physical layout of data that should be input by the data fetch command that will be received after the read by the current data fetch command. When predicting physical continuity and reading a plurality of data in a single input operation in advance, it is determined what the number of data at that time is to be. By returning to the original single input when the level becomes lower than a certain level, the reading of the data group to be searched by the data fetch command which is independent in terms of time is performed.
By collectively reading in advance, the number of input operations is reduced, and the response time required for each data fetch command is shortened.

[0006]

SUMMARY OF THE INVENTION The above object is the physical arrangement of data to be read by a data fetch command which will be received after the current data fetch command is triggered. Is an acceleration that predicts the physical continuity from the past input history and reads out a plurality of data in a single input operation in advance, and decides the number of data at that time. This is achieved by providing a function and a deceleration function that restores the original single input when the physical continuity falls below a certain level.

[0007]

The physical layout of the data read several times in the past on the magnetic disk device is stored, and the data that will be read from now on and the data that will be needed in the future will be stored. By anticipating and reading in advance, the number of input operations can be reduced.

[0008]

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

First, terms used in this embodiment will be described.

Block: A minimum transfer unit from the magnetic disk to the memory, which can be input by one input operation. It also has the function of inputting multiple blocks with one input operation. A plurality of records are stored in the block. It is faster if a plurality of data read by one input operation are physically continuous on the disk and if they are not continuous. This is because the arm mechanical operation of the disk drive and the rotational waiting time of the disk are the smallest.

Order Alignment: A state in which blocks in which a record to be input by a search command issued by a user is located are stored side by side on the magnetic disk device from the lower level to the upper level without any disturbance.

Quasi-order arrangement: A state in which the arrangement is almost close to the order arrangement.

Reverse Alignment: A state in which blocks in which a record to be input by a search command issued by a user is located are stored side by side on the magnetic disk device from the upper level to the lower level without being completely disturbed.

Quasi-reverse alignment: A state in which the alignment is almost nearly reverse.

Non-alignment: A state in which blocks in which a record to be input by a search command issued by a user is located are uniformly distributed and randomly stored on the magnetic disk device.

Alignment function:
It is an arithmetic function to quantify and has a certain range. The maximum value is ordered, the minimum value is reversed, and the intermediate value is unaligned. The value representing the quasi-alignment is determined individually.

Alignment coefficient: A value obtained by an alignment function.

Threshold: The range to which the multiple block input is applied is from quasi-ordered to forward-ordered. If the alignment coefficient is a certain value or more, multiple blocks are input. If it is less than that, single block input is performed. This specific value is called a threshold value and is a boundary point of quasi-ordered alignment.

Next, an intelligent fetch method for realizing the present invention will be described.

The intelligent fetch method stores the physical layout of the data read several times in the past on the magnetic disk device, and reads the data that is actually needed from now on.
This is a method of predicting data that will be needed in the future as well as the data and reading it in advance.

In order to implement this intelligent fetch system, the specific means for the database user to issue a data search command to the DBMS and return the result will be described below.
This will be described with reference to FIG. FIG. 1 is a diagram showing a flow of control of internal equipment.

In response to a record search command from the database user 1, this record search command is analyzed by the search command analysis function 2 and the input block determiner 3 is based on the result.
Determines the physical storage position of the block where the target record is located. The physical storage position of the block is passed to the alignment function unit 4 to obtain the alignment coefficient. At this time, the current storage position is registered in the alignment coefficient / input history table 12. The number of blocks when the data base (block) 9 is input from the magnetic disk device 8 to the main storage device 10 by the input device 5 is adjusted by using the acceleration / deceleration device 6. The record 11 in the block thus obtained is judged by the record condition judging device 7, and the requested record is returned to the database user 1.

Next, the processing flow of the intelligent fetch method will be described with reference to FIG. A certain database
When the user 1 issues a record search command for the first time, n is initialized to 0 in step 101. (N is the number of consecutive blocks to be input and is calculated by the nth power of 2.) Step 101 is not executed in the second and subsequent record search commands. In step 102, the physical storage position of the block to be input from the magnetic disk device 8 is obtained. In step 103, it is determined whether or not the obtained block position is the same as the storage position similarly obtained in step 102 immediately before. That is, if the desired record is in the same block, the determination in step 103 is yes, and if it is in a different block, it is no.

Alignment coefficient / input history table 1 in step 104
The block storage position is registered in 2. Step 105
Then, the past series of registered block storage positions are applied to the alignment function unit 4 to obtain the alignment coefficient. When the alignment coefficient thus obtained is larger than a predetermined threshold value, a plurality of blocks are input. Otherwise, enter a single block instead of entering multiple blocks. It is called deceleration that what has been input in a plurality of blocks until now becomes no in the determination in step 106. The accelerating / decelerating device 6 executes this.

When accelerating, the number of continuous blocks to be input at step 110 and step 111 is determined.
The number of blocks is a power of 2 and is "1, 2, 4, 8, 1
6, 32 ,. ． I will proceed step by step. When a certain value is set as the upper limit and the value is reached, step 110 is not performed any more. When a certain specific value continues in this way, it is called constant speed.

When decelerating, in step 107, the alignment coefficient / input history table is initialized. In step 108, n-
By setting to 1, n becomes 0 in step 110, and a single input can be made.

Regardless of yes or no in the determination in step 106, it is determined in step 109 whether or not a plurality of block inputs have been already read in the main memory,
If yes, there is no need to enter. In this respect, the input operation to the disk device can be reduced and the response time can be shortened. On the other hand, if the answer is no, it is not in the main storage device and is input in step 112. At that time, 2 n blocks are input by one input operation.

As described above, when it is desired to input 2 n power blocks, 2 n power blocks can be input in one input operation rather than the total time for inputting a single block input to 2 n power input operations. The time is shortened by utilizing the hard characteristic of the disk drive device that the reading time is shorter.

At step 113, it is judged whether the record in the block read in the main storage device satisfies the retrieval condition. If yes, at step 114, the record is given to the database retrieval user. reply.

A record for evaluating such a series of processes.
Repeat as long as there is. Further, when a new search command is issued from the user, the process is executed from step 102.

Next, FIG. 3 shows a specific example of the intelligent fetch method of the present invention. FIG. 3 compares a series of search commands issued by a database user with the alignment coefficient obtained by using the alignment function at which occasion and from what block the input by the intelligent fetch method is performed. It is a figure. 21 is a series of block numbers determined by the input block determiner, 22 is an alignment coefficient axis, 23 is a continuous block number axis, 24 is a range from quasi-order alignment to normal alignment, 2
5 is a continuous block input range, 26 is a single block input range, 27 is a block number to be input in one input operation,
28 is a definition of the alignment function.

The blocks arranged on the magnetic disk device are arranged in the physical storage order of 1, 2, 3 ,. ． And consecutive block numbers are given. On the other hand, the block number to be input, which is determined by the search command issued by the database user, is written on the time axis.

First of all, the input of the first block numbers 1 and 2 is
According to the definition of the alignment function shown in FIG. 4, since the alignment coefficient is 0, it becomes a single block input. In block number 3, since the alignment coefficient is 1, it becomes a continuous block input. At this time, the number of blocks was 1 at the previous time, so here, the number of blocks is doubled to input 2 blocks.

Since the block number 4 has already been entered, it need not be entered here. Since the alignment coefficient is 1 in block number 5, it is further doubled to input 4 blocks. It is not necessary to perform the input operation up to the block number 8.

Block numbers 12, 14, 9, 11, 1
In Nos. 2, 13 and 14, since the alignment coefficient is less than the threshold value of 0.4, all input operations performed during this period are performed by single block input. At block number 15, the alignment coefficient becomes larger than the threshold value, so continuous block input is performed.
Since the number of blocks at this time has returned to 1 the previous time, it is doubled to 2 here. After that, when the block number is 17, 2
4 blocks are input by doubling, and when 21 is input, doubling is further performed and 8 blocks are input.

In this embodiment, 12 types of blocks are used to input 12 blocks.
The input operation is performed twice, and 27 types of blocks are actually input. In the conventional method, the input operation is performed 20 times and the input of 20 kinds of blocks is performed.

Although the database is a magnetic disk device in this embodiment, it may be a large-scale storage device such as an optical disk device.

As described above, conventionally, when the data to be taken out is not in the main memory, N times of input operation is always required to input N blocks. In the intelligent fetch method of No. 3, when the input is from a completely random position, the same is N times, but when the input is from an almost physically continuous position, N / M (M
Has the effect that it can be performed with the maximum number of blocks input at one time.

Further, if the number of N pieces is 3M and each of them is partially continuous, there is an effect that the number can be reduced to N. (3 + log M) / 3M times.

Further, in the case of repeating the pattern of M continuous and M random patterns, N. (M + 1 + log
There is an effect that it can be reduced to M) / 2M times.

[0041]

By the intelligent input, the data which is desired soon will be made to exist in the memory in advance, so that the subsequent retrieval command can be referred to the memory immediately without inputting it from the disk. Since the search result is shortened, the search result can be returned to the database user quickly and a better service can be provided.

[0042]

[Brief description of drawings]

FIG. 1 is a diagram showing a control flow of an internal device until a data base user issues a data search command to a DBMS and returns a result.

FIG. 2 is a flowchart showing a flow of processing for performing intelligent fetch.

FIG. 3 is a diagram comparing in what triggers and from what block from what, by an intelligent fetch method, an alignment coefficient obtained by using an alignment function.

FIG. 4 is a diagram showing a definition of an alignment function used in FIG. 3 and an alignment coefficient as a calculation result using the alignment function.

[Explanation of symbols]

2 ... Search command analysis function, 3 ... Input block determiner, 4 ...
Alignment function unit, 5 ... input unit, 6 ... acceleration / deceleration unit, 7 ... record condition determination unit.

Claims (3)

[Claims] 1. A first means for storing the physical storage position of the data stored in the data base as an input history when the data under a certain arbitrary condition is read out from the data base. Based on the input history stored in the first means, when the recently read data group is judged to be completely or to some extent physically consecutive storage positions, the data to be read thereafter is read. A second means for preliminarily reading a group of physically continuous data from the data base in a single read operation, expecting that the data group will also be physically continuous. An intelligent fetch method for database retrieval, which is characterized by shortening the time required for database retrieval. 2. The second means comprises a third means for changing the number of data read by one read operation by learning the continuity of the storage positions. An intelligent fetch method for database search. 3. The database retrieval according to claim 2, wherein said third means learns for each database user and each database schema. Intelligent fetch method.