JP2923917B2 - Data access method for large-scale knowledge database system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Summary] The present invention relates to a data access method capable of efficiently executing data conversion between data on a main storage and data on a file. In a knowledge database system that consists of a computer and external storage means and handles large-scale data using an object-oriented language, with the aim of increasing the conversion efficiency between the two in almost the same format, the main storage and the external storage means Object management means for managing all objects, data management means for managing objects stored in the external storage means in the form of a relational database, and data indicated by an absolute address and a pointer on a main storage. Stored in a continuous area in a structure to be inserted using an offset on the external storage means. And data conversion means for mutually converting the data between the object management means and the data management means.

[Industrial applications]

The present invention relates to a knowledge base system for handling large-scale data, and more particularly, to a data access method capable of efficiently executing data conversion between data on a main memory and data on a film.

[Conventional technology]

With the technological progress and spread of electronic computers, the amount of software production has been increasing at an accelerating rate. Due to the lack of a software development population, software production cannot keep up with demand, and there is an urgent need to improve productivity.

The use of object-oriented languages is being advanced as a solution for that. Object-oriented languages are COBOL and F
It has the feature that programs can be described more naturally than conventional structured programming languages such as ORTRAN and BASIC, and is also suitable for prototyping and software reuse.

The basic unit of programming in an object-oriented language is an object, and the object is composed of data and a procedure for the data. By sending a message to an object, the object itself executes a procedure corresponding to the message, thereby realizing a process desired by the user. An object has a class and an instance. An instance is a specific value or data, whereas a class is a set of objects having similar properties. It is also possible to define an upper / lower relationship between classes.

FIG. 12 is an explanatory diagram of a class. There is a class called a doctor, and classes of doctors in each department such as a dentist, a physician, and a surgeon can be defined as lower classes. The doctor class of each department is a set in which each doctor of the department is an instance.

Instances belonging to the same class inherit the attributes described in that class and have them in common. For example, in the figure, the instance of the physician class is information of each physician such as physician 1, physician 2, etc., and the attributes (name, license number, patient name) described in the physician class are used by the physician. Each instance belonging to the medical class has in common. That is, the name attribute of the instance physician 1 is Sato, the license number attribute is 83, the patient name attribute is patient A, patient B, patient C, the name attribute of the instance physician 2 is Suzuki, and the license number attribute is
103, the patient name attributes are patient W, patient X, patient Y, and patient Z.

[Problems to be solved by the invention]

As described above, an object-oriented language that uses classes and instances as a basic unit of programming is being used to improve software productivity.However, in a conventional object-oriented language, for example, data conversion speed is slow and a large amount of data is However, there is a problem that it is not possible to efficiently handle data, and it is desired to solve this problem in order to improve the productivity of software by using an object-oriented language.

In the conventional object-oriented system, object processing is performed only on the main memory. Recently, management of large-scale knowledge has been studied, but in those studies, retrieval of databases other than the own system has become a problem, and the search results have been converted to the data expression form of the own system.

In such a case, double and triple copying is required for data conversion, and there is a problem that the cost of copying and data conversion increases.

Furthermore, some conventional systems execute a database language for another database and convert the search results into a format that can be easily handled on the main memory of the own system.
However, because of the unknown or the data of variable length attribute even in the case may take in advance how much of the area, there was a need to copy from to check the area. Therefore, when there are a plurality of variable length attributes, it is necessary to execute the copying a plurality of times, and there is also a problem that the conversion cost increases.

SUMMARY OF THE INVENTION It is an object of the present invention to improve the conversion efficiency between a main storage and a file in substantially the same format.

[Means for solving the problem]

FIG. 1 is a block diagram showing the principle of the present invention. Referring to FIG. 1, the large-scale knowledge database system includes an electronic computer 1 and an external storage unit 2, and handles large-scale data using an object-oriented language. The electronic computer 1 includes an object management unit 3, a data management unit 4, and a data conversion unit 5, and the data management unit 4
Is connected to the external storage means 2.

The object management means 3 manages all objects in the main storage and the external storage means 2, and manages classes and instances, manages programs, and the like.
The data management means 4 manages the objects stored on the external storage means 2, for example, a disk, in the form of a relational database. The data conversion unit 5 stores data pointed to by an absolute address and a pointer on the main memory when storing an object generated on the main memory by a command from a user, for example, on the external storage unit 2, for example, a disk. The data is converted into a format to be stored in the continuous area in a structure in which an offset is used on the means 2. Conversely, the offset is converted to an absolute address in order to copy the knowledge data stored in the external storage means 2 to the main storage.

[Action]

In the present invention, the data conversion, that is, the mapping efficiency is improved by making the data on the main memory and the data on the file almost the same format. On the file, the variable length portion is managed by offset so that data is stored in a continuous area for speeding up access. On the main memory, data is managed by absolute addresses and pointers to speed up access. That is, on the object management means 3, data is managed by absolute addresses and pointers, and on the data management means 4, data is managed by offsets and stored in continuous areas.
Therefore, when data in a file is copied to the main memory in response to a data search request, an area is collectively secured using an absolute address and a pointer, and data copy and conversion of an offset to an address are performed. Will be. Further, in response to a request to store the object in the data external storage means 2, the conversion of the absolute address into an offset and the copying of the data on the main storage to a file are executed.

As described above, according to the present invention, it is possible to minimize the number of times of copying even for data including a plurality of variable length attributes.

〔Example〕

FIG. 2 is a block diagram showing the overall configuration of a large-scale knowledge database system using the access method of the present invention. In FIG. 1, the system includes an electronic computer 1 such as a large computer, an external storage device 6 such as a magnetic disk device and an optical disk device, and a terminal device 7. The computer 1 mainly includes a CPU 8, and the CPU 8 is connected to the main memory 9, the input control unit 10, and the external storage control unit 11. The terminal device 7 is operated by a user to input commands and data such as creation, storage, search, and execution of a procedure to the computer 1, and receive object information and processing results from the computer 1 on a display or the like. Output.
The external storage device 6 stores a large number of objects and forms a knowledge database.

The input / output control unit 10 is connected to the terminal device 7, the CPU 8, and the main memory 9, and serves as an input / output interface for connecting the computer 1 to input / output devices such as the terminal device 7. The external storage control unit 11 is connected to the external storage device 6, the CPU 8, and the main storage 9 and controls input and output of knowledge data between the main storage 9 and the external storage device 6.

On the other hand, a main memory 9 in the electronic computer 1 stores a knowledge base system (Jasmine) 12 serving as a core of a knowledge database, and a CPU 8 executes various processes on the knowledge base in accordance with the program. Further, a part of the knowledge data 13 is stored in the main memory 9, but in the present invention, conversion between the knowledge data 13 and the data stored in the external storage device 6 becomes a problem.

FIG. 3 is a configuration diagram of the knowledge base system (Jasmine) 12. The system is modularized, and the main control mechanism 14, analysis mechanism 15, optimization mechanism 16, code generation mechanism 1
7. Consists of an interpreter 18, an object management mechanism 19, and a data management mechanism 20. Next, the role of each module will be briefly described.

The knowledge base system of this embodiment provides a user with an object-oriented programming language and an interactive processing system. The main control mechanism 14 controls the execution of these object-oriented programs and the interactive processing system. In addition, the analysis mechanism 15 analyzes the syntax of the programming described in the object-oriented programming language and generates an intermediate code, and the conversational processing system analyzes the object operation sentence inputted by the conversational processing and converts the sentence into an executable form. I do. The interpreter 18 executes the object operation statement converted into an executable form.

The intermediate code generated by the analysis unit 15 is sent to the optimization unit 16 for optimization. That is, the optimization mechanism
16 performs optimization such as speeding-up by, for example, join order conversion, creates operation blocks such as a selection operation and a joint operation as intermediate information, and transfers the intermediate blocks to the code generation mechanism 17. The code generator 17 generates a C source program from the intermediate code optimized by the optimizer 16. The generated C source program is compiled by a C compiler stored in the main memory 9 of the electronic computer 1 in the main memory 9 separately from the knowledge of the base system, and linked to an execution library of the knowledge base system to form an executable file. And the debugger does this.

On the other hand, the object management mechanism 19 executes the management of the object. That is, it manages classes and instances, manages inheritance of inheriting attributes and procedures between classes and instances or between classes having an upper / lower relationship, and manages programs.

The data management mechanism 20 manages, for example, an extended relational database, can manage non-normalized tables, and has high-speed access methods such as index search and hash joint. Furthermore, it is a high-speed and general-purpose database system engine capable of giving a tuple, that is, a process for each row in a table of a relational database, by a compiled code called a function or a mapping function described later and executing the compiled code. Further, the object management mechanism 19 uses, for example, an active object space to
Interaction with is reduced.

The data access method of the present invention is an object management mechanism.
It is executed as a program inside 19 and the data management mechanism 29.

Before describing the data conversion algorithm in the present invention, FIG. 4 shows a specific example of data stored on the main storage and the database (file) in the present invention. In the figure, it is shown as the contents of the data that the favorite actresses of Suzuki Fumio are Sagara Haruko and Minamino Yoko, and that the friends with whom Sagara Haruko is dating are Saito Yuki and Asaka Yui. In the main memory of (a), the pointer after Fumio points to the absolute addresses of Haruko and Yoko, and the pointer after Haruko points to the absolute addresses of Yuki and Yui.

On the other hand, in a database, that is, on a file, data is stored in a continuous area. First, the tuple length as a row of the entire data is stored, and then Fumio is stored. The next offset of Fumio indicates a variable length size as shown. This variable length size is the size of the inner table, here the size of the inner table to indicate that the actresses who like Fumio Suzuki are Sagara Haruko and Yoko Minamino, and the number of rows in the table, that is, the number of tuples, It becomes 2.

Length stored after the number of tuples (length)
Is the length of the area for storing data indicating that Sagara Haruko's data and friends with whom Sagara Haruko is dating are Saito Yuki and Asaka Yui, and data for Saito Yuki and Asaka Yui is determined by the offset stored next to Haruko. Is indicated.

The NIL (length) stored thereafter indicates that one tuple in the inner table has been completed, and then the data of Minamino Yoko is stored. The offset stored under Fumio's age “27” indicates Fumio's surname Suzuki, and this is all the data. The offset after Harco's age "20" indicates Haruko's surname Sagara, and the offset after Yoko's age "22" indicates Yoko's surname Minamino.

FIG. 5 shows an outline of the data conversion algorithm. FIG. 9A shows an outline of a data search algorithm. In step 21, a database, that is, a file is searched by the tuple identifier tid, and the search result is copied to the main storage in step 22. Then, in step 23, the offset is converted to an absolute address, and in step 24, registration in the active object table described later is performed. FIG. 5B shows an outline of an algorithm when an object is generated, that is, when an object formed on the main memory is stored in a file. First, in step 25, the absolute address is converted into an offset. At this time, if the data is not in the continuous area, the conversion to the offset is performed simultaneously with the rearrangement of the data into the continuous area. And step
At 26, the information of the continuous area is written to the file. The processing in the next step 27 is performed exceptionally as necessary, and in step 25, when the data is arranged in a continuous area, the offset is returned to the absolute address again to speed up access. It is. That is, when data on the main memory is copied as it is, there is no need to convert it to an offset, but since copying requires a large area and takes time, in step 25, it is assumed that the data is in a continuous area. Regardless of whether or not the conversion to the offset is performed.

FIG. 5 (c) shows an outline of the data update algorithm. In step 28, only the update attribute is converted.
In step 29, only the update attribute is updated. The image section (d) is an algorithm for deleting data,
At 30, data is erased using the tuple identifier tid.

FIG. 6 shows an embodiment of the active object table in step 24 of FIG. In the figure, from the patient class in the database file, the object identifier oi
A state in which data in which d is 3: 36: 1 is stored in the active object table on the main memory is shown.

FIG. 7 shows a structure of data conversion as a common data structure, that is, auxiliary information for mapping. FIG. 7A shows the data structure of the mapping auxiliary information mapinfo of the instance of the class, and shows the total size of the instance, the number of variable-length attributes including the pointer attribute, and information about the table in the database, for example, whether it is sequential or inner. It consists of a pointer to rb information and an array of mapping information for each attribute. FIG. 2B shows a cell (mapinfoc).
ell), the address of the attribute, a pointer to the actual data, an integer indicating the type of the attribute, ie, pointer or inner table or variable length data,
And in case of inner table, pointer to mapinfo is stored.

FIG. 8 shows the data structure of auxiliary information for storage in a database. Updating of data in the database by the data management mechanism 20 is performed by a field pointer array,
This information is indicated by fpainfo, and includes an integer indicating the number of variable-length data attributes and an array of database auxiliary information of each attribute as shown in FIG. FIG. 2B shows a cell (fp)
ainfocpll), which stores an integer indicating a field number, an integer indicating a relative address, a character indicating a type, and a character indicating whether or not it is multi-valued.

FIG. 9 is a detailed algorithm of step 25 in FIG. 5, that is, an algorithm for converting an absolute address into an offset. FIG. 7A shows the overall algorithm. In step 31, it is determined whether or not the variable length attribute has been updated. If not, in step 32, a trans offset list sub-unit for converting to an offset without copying data is used. (Tr
ansoffsetlistsub) is executed. If the variable-length attribute is updated, the variable-length attribute is copied in step 33, and the conversion from the absolute address to the offset is performed using the copy.
t sub).

FIG. 9 (b) shows a trans offset sub-algorithm, and the following processing is performed in step 34 by the number of variable length attributes in the mapinfo of FIG. 7 (a). That is, the relative position of the attribute is obtained in step 34a, and the
In 4b, the type of pointer, inner table or variable length data is checked.If it is a pointer, NULL is set as the offset value.If it is variable length data, the size is updated and the offset from the beginning is obtained. Registration is performed. If there is an inner table, a trans offset sub is called for each tuple.

FIG. 9 (c) shows the algorithm of the trans offset list sub. First, in step 35, the fapi shown in FIG.
Scanning is performed by the number of nfo variable length attributes, and the total size of the data is obtained. At this time, the address of the variable length data is registered. In the case of an inner table, a trans offset list sub is called for each tuple in the inner table. Next, in step 36, the area is secured and only the fixed length area is copied, and in step 37, the following processing is performed for the number of variable length attributes. That is, the relative position of the attribute data is obtained in step 37a, and the type is checked in 37b. If the type is pointer, NULL is set as the offset value, and if the data is variable length data, the data area is copied by that length.

FIG. 10 is an algorithm for converting the offset of step 23 or 27 in FIG. 5 into an absolute address. FIG. 7A shows the overall algorithm. First, at step 38, it is determined whether or not the variable length attribute is updated. If there is no update, it is necessary to convert the offset into an address for the variable-length data, so in step 40, the process of converting the offset of the continuous area into an absolute address is executed as trans address-s (transaddr-s). If there is an update, the fixed length area is copied and restored at step 39. FIG. 7B shows the algorithm of the trans address-s. In this processing, the following processing is performed by the number of variable length attributes in the mapinfo in FIG. That is, the relative position of the attribute is obtained in step 41a, and the type of the attribute is checked in step 41b.
In the case of variable length data, the offset is converted to an absolute address. In the case of an inner table, the trans address -s is called for each tuple, and the head and tail addresses are filled.

FIG. 11 shows an algorithm in the attribute update processing. This algorithm, updater fpa (update fp
In a), the following processing is performed in step 42 by the number of attributes of fpainfo as the database auxiliary information of FIG. 8A. That is, in step 42a, the type of the attribute is checked, and if the data is variable-length data, the absolute address is
Registered to fpa. In the case of the inner table, the update data fpa is called for each tuple. Then, in step 42b, the attribute is updated based on the fpa.

In the above-described embodiment, when the data on the filter is brought into the main memory, the offset is converted into the absolute address. However, the data on the file is converted without converting the offset into the address. It is also possible to use. In this case, the mapping cost can be further reduced, and the function for the data on the file can be applied to the data on the main memory. Slows down.

〔The invention's effect〕

As described in detail above, according to the present invention, it is possible to mutually convert data in the main storage and data in a file only by converting an absolute address and an offset,
The processing required to convert the data of an object from a file to the main storage or to write the data from the main storage to the file can be extremely reduced, which greatly contributes to speeding up the system.

[Brief description of the drawings]

FIG. 1 is a block diagram showing the principle of the present invention, FIG. 2 is a block diagram showing the overall configuration of a large-scale knowledge database system using the access method of the present invention, and FIG. 3 is a block showing the configuration of a knowledge base system (Jasmine). FIGS. 4 (a) and 4 (b) are diagrams showing specific examples of data stored in a main storage and a file, and FIGS. 5 (a) to 5 (d) are diagrams showing an outline of a data conversion algorithm. Fig. 6 is a diagram showing an embodiment of the active object table, Figs. 7 (a) and 7 (b) are diagrams showing the structure of mapping auxiliary information, and Figs. 9 (a) to 9 (c) show an embodiment of an algorithm for converting an absolute address into an offset, and FIGS. 10 (a) and 10 (b) show offsets. Algorithm to convert to absolute address Shows an embodiment, Figure 11 shows an embodiment of an attribute of the update algorithm figure FIG. 12 Class - an illustration of an instance. 1 ... Electronic computer, 6 ... External storage device, 8 ... CPU, 9 ... Main memory, 12 ... Knowledge base system (jasmine), 13 ... Knowledge data, 19 ... Object management mechanism, 20 ... Data management mechanism.

Continuation of front page (56) References JP-A-63-67638 (JP, A) JP-A-63-136147 (JP, A) JP-A-63-158643 (JP, A) JP-A-1-159750 (JP) , A) Fumio Suzuki, et al., "Implementation of Jasmine, an Object-Oriented Multimedia Knowledge Base", Information Processing Society of Japan Research Report, Information Processing Society of Japan, July 20, 1989, Vol. 89, NO. 63 (DBS-72), p. 57-P. 64 Fumio Suzuki, et al., "On the Realization of the Program Interface of the Object-Oriented Multimedia Knowledge Base Jasmine", Proc. Of the 37th Annual Conference of IPSJ, 1988, IPSJ, 1988 October, VOL. 37th, NO. 1, P. 405-P. 406 (58) Field surveyed (Int.Cl. ⁶ , DB name) G06F 12/00 G06F 15/40

Claims (1)

(57) [Claims] 1. A knowledge database system comprising an electronic computer (1) and external storage means (2) and handling large-scale data using an object-oriented language. Object management means (3) for managing objects; data management means (4) for managing objects stored in the external storage means (2) in the form of a relational database; absolute addresses and pointers on main storage And data stored in a continuous area in a structure to be inserted using an offset on the external storage means (2).
A data access method in a large-scale knowledge database system, comprising a data conversion means (5) for mutually converting between the object management means (3) and the data management means (4).