JPH0713824A - Data management method - Google Patents

Abstract

PURPOSE:To improve the processing efficiency of data having link structure by using a virtual storage system. CONSTITUTION:Data stored in a main memory 4 are represented in a virtual storage space 20a and data stored in an external memory 8 are represented in a virtual storage space 20b and a data file space 22. Data transfer between the main memory 4 and external memory 8 is performed in page units between the virtual storage space 20a and virtual storage space 20b. Data transfer between the virtual storage space 20a in the external memory 8 and the data file space 22 is performed in units of structure consisting of plural data having close link structure.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to data having a link structure such as list data according to the processing of a CPU (Central Processing Unit) in a computer system which performs storage management using a virtual storage system such as a paging system. The present invention relates to a data management method capable of efficiently processing data on a memory system.

[0002]

2. Description of the Related Art In the fields of artificial intelligence, mathematical expression processing, natural language processing, and the like, data having a link structure in which a link relationship between data is expressed using pointers is stored in a memory and is processed by a CPU. And dynamically change these link structures. Generally, many computer systems have a memory system in which a main memory such as a semiconductor memory that can be accessed at high speed and an external memory such as a hard disk that can be increased in capacity are hierarchically connected. In this computer system, data stored at address intervals close to each other in the memory address space are often accessed from the CPU at short time intervals, and the address space of the memory is assumed to be local to the data access by the CPU. That is, a virtual storage system such as a paging system is used in which a virtual storage space is divided into a plurality of pages, and data is transferred between a main memory and an external memory in units of this page for storage management.

However, when the CPU processes link-structured data, the data access by the CPU generally does not have the locality in the address space as described above, and has a direct or indirect link relationship. Data is accessed by the CPU at short time intervals.
Therefore, when the data having the list structure is processed without considering the link structure of the data by using the computer system which performs the storage management in the unit of page as described above, FIG.
As shown in Fig. 2, data having direct or indirect link relations are frequently stored in different pages, and a paging process between the main memory and the external memory frequently occurs, and There is a problem that the processing efficiency of the system is reduced. In order to solve these problems, for example, Japanese Patent Application No.
The No. 83227 handles a structure, which is a set of a plurality of data having a dense link relationship, as a unit for transferring between a main memory and an external memory, and efficiently manages a series of data having a dense link relationship. A data management method for storing in a memory is disclosed. According to this data management method, a series of data accessed at short time intervals in the processing of the CPU can be efficiently stored in the main memory, and the processing efficiency of the computer can be improved.

[0004]

As described above, if the data management method of Japanese Patent Application No. 5-83227 is used, the processing efficiency of the computer can be improved. However, this data management method is mainly structured not on pages. Since a unit of data is transferred between the memory and the external memory, the virtual storage method such as the paging method used in a general computer system cannot be used as it is.

The present invention has been made in view of the above-mentioned problems of the prior art, and an object of the present invention is to improve the processing efficiency of data having a link structure by directly using the virtual storage system used in the conventional computer.

[0006]

In order to solve the above-mentioned problems of the prior art and to achieve the above-mentioned object, the data management method of the present invention provides a virtual storage space for the first storage means and the second storage. And the data file space is represented in the second storage means. Then, according to the processing of the arithmetic control means, the data transfer between the first storage means and the second storage means is performed in the normal computer system for the data expressed in the virtual storage space. Similar to the data storage management, the page created based on the address in the virtual storage space is used as a unit. Further, regarding the data expressed in the data file space, the data file space and the first storage means are set in units of a structure composed of the data and data directly or indirectly linked to the data. To and from the represented virtual memory space

Further, in the data management method of the present invention, preferably, of the data stored in the first storage means and the second storage means, unused data is stored in the virtual storage space. Garbage collection is performed for each of the data existing in 1) in units of data, and for the data existing in the data file space, in units of the structure, by the data transfer process and the time division method.

In the data management method of the present invention, more preferably, in the garbage collection, an incremental data collecting method using a reference counter is used.
Garbage collection and mark-and-sweep garbage collection of mark-tracing used cells and sweeping unused cells are performed, and incremental garbage collection is performed prior to the mark-and-sweep garbage collection.

Further, the data management method of the present invention is more preferably such that the virtual storage space is divided into at least two small virtual storage spaces, and the data transfer processing and the data transfer processing for the data existing in one of the small virtual storage spaces are performed. When free access is made to the small virtual memory space after the access from the operation control means, the data being accessed from the operation control means and the data directly or indirectly linked to the data. Existing data is copied to the other divided small virtual memory space, and the data existing in the copied small virtual memory space is accessed from the data transfer process and the arithmetic control means.

[0010]

In the data management method of the present invention, the data having the link structure is stored in the first storage means and the second storage means, and the first storage means and the second storage means are processed according to the processing of the arithmetic control means. Data transfer in page units is performed with the storage means. This data transfer is performed by transferring a page including the data expressed in the virtual storage space. Further, in the case of transferring the data expressed in the data file space, that is, in the case of transferring the data stored in the second storage means and expressed in the data file space to the first storage means, , Transferring the structure containing the data from the data file space to the virtual storage space represented in the first storage means.

In the data management method of the present invention, since data transfer between the virtual storage space and the data file space is performed while reflecting the link structure of the data, a series of directly or indirectly linked data is stored. Since the data is collectively transferred between the first storage means and the second storage means, it is possible to reduce the frequency of data transfer performed between the first storage means and the second storage means. As a result, the processing efficiency of the computer system can be improved.

[0012]

EXAMPLE A first example will be described. FIG. 1 is a block diagram of a computing system using the data management method of this embodiment. As shown in FIG. 1, in the computer system of this embodiment, the CP as an arithmetic control means is connected via the internal bus 14.
A U (Central Processing Unit) 2 and a main memory 4 as a first storage means are connected. Also, C
The PU 2 is connected to the external memory 8 as the second storage means via the internal bus 12, the input / output circuit 6 and the external bus 10. The CPU 2 performs arithmetic processing according to a program incorporated in advance, for example. Main memory 4
Is a bipolar semiconductor memory that can be accessed at high speed, and holds list-structured data as link-structured data that is frequently used in the CPU 2, as will be described later.

The external memory 8 is, for example, a hard disk drive having a storage capacity larger than that of the main memory 4. For example, data of a list structure processed by the CPU 2 that is not stored in the main memory 4 is stored in the external memory 8. Remember. The CPU 2 inputs necessary data from the main memory 4 via the internal bus 12 when performing arithmetic processing. At this time, if the required data does not exist in the main memory 4, the data including the required data stored in the external memory 8 is used as a unit of a page or structure described later, and the external bus 10, the input / output circuit 6 and the It is transferred to the main memory 4 via the internal bus 12.

Hereinafter, the main memory 4 and the external memory 8
The memory management in will be described. FIG. 2 is a diagram for explaining storage management in the computer system of this embodiment. The virtual address of the data accessed by the CPU 2 is represented in the virtual storage space 20. Of the virtual memory space 20, the virtual memory space 20a is allocated on the main memory 4, and the virtual memory space 20b is allocated on the external memory 8. On the other hand, the address of data that is accessed from the CPU 2 and whose address is not expressed in the virtual storage space 20 is expressed in the data file space 22. The data file space 22 is allocated on the external memory 8.

In the storage management in the computer system of this embodiment, in the virtual storage space 20, the address space expressed in the virtual storage space is divided into pages of a predetermined size, similarly to the paging method in a normal computer system. , The virtual storage space 20a with this page as a unit
And data transfer between the virtual storage space 20b and the virtual storage space 20b.

On the other hand, between the virtual storage space 20 and the data file space 22, the applicant of the present invention filed Japanese Patent Application No. 5-83227.
Data management using the data management method of the link structure disclosed in No.

First, a storage management method of the virtual storage space in the computer system of this embodiment will be described. The virtual storage space is represented using a virtual address composed of a page number 30a and an in-page address 30b as shown in FIG. Access to the main memory 4 by the CPU 2 is performed using this virtual address. When the main memory 4 is accessed by the CPU 2, storage management is performed based on the page allocation table 32 as shown in FIG.

The page allocation table 32 has an identification field 32a and a page address field 32b corresponding to all the pages created in the virtual storage space as described above. The identification field 32a indicates whether or not the corresponding page exists in the main memory 4, and in FIG. 3, when the corresponding page exists in the main memory 4, it is “R”, and when it does not exist, “P”. Is shown.

The page address field 32b indicates the head address PN1 of the corresponding page in the main memory 4 when the corresponding page exists in the main memory 4. On the other hand, the page address field 32b indicates the start address PN2 of the corresponding page in the external memory 8 when the corresponding page does not exist in the main memory 4. The page allocation table 32 is expressed using, for example, a hash table.

When the main memory 4 is accessed from the CPU 2 using the virtual address 30 shown in FIG. 3, the page allocation table 32 described above is referred to, and the identification field corresponding to the page indicated by the page number 30a of the virtual address 30. 32a is identified. As a result of this identification, when the identification field 32a indicates "R", that is, the page number 30
When the page indicated by a exists in the main memory 4, the corresponding page address field 32b is set to the upper bit and the in-page address 30b of the virtual address 30 is set.
The main memory address 34 is created by using as the lower bit. Then, the CPU 2 accesses the created main memory address 34.

On the other hand, as a result of the above identification, the identification field 3
When 2b indicates "P", that is, when the page indicated by the page number 30a exists in the external memory 8, the page indicated by the corresponding page address field 32b is transferred from the external memory 8 to the main memory 4 and identified. The field 32a becomes "R", the main memory address 34 is created, and the created main memory address 34 is accessed from the CPU 2 as in the case described above.

As described above, the data expressed in the virtual storage space 20 is subjected to paging type storage management between the main memory 4 and the external memory 6 as in a normal computer system.

Next, a storage management method between the virtual storage space 20 and the data file space 22 in the computer system of this embodiment will be described. The data being processed by the CPU 2 exists in the virtual memory space 20a of the main memory 4, but when trying to obtain new data by tracing a pointer from the data being processed, the new data is stored in the external memory. 8 in the data file space 22. At this time, in the computer system of this embodiment, the structure including the new data is transferred from the data file space 22 onto the main memory 4 by using the data management method disclosed in Japanese Patent Application No. 5-83227, that is, the structure. In. Since the structure including the new data exists in the external memory 8, the structure-structured structure exists in the main memory 4 when the structure-in ends.

When the storage management is performed in this way, when the data is structured into the virtual storage space 20b from the data file space 22, a series of data having a dense link structure is transferred as a bundle, so that the external memory 8 is used. The frequency of data transfer between the main memory 4 and the main memory 4 can be reduced.

The storage management method between the virtual storage space and the data file space 22 will be described below with reference to specific examples. First, list-structured data existing in the virtual storage spaces 20a and 20b and the data file space 22 will be described. FIG. 4 is a diagram for explaining the data stored in the virtual storage space 20 and the data file space 22. The virtual storage space 20 is composed of the virtual storage spaces 20a and 20b shown in FIG. In FIG. 4, mm1 to mm2, mm4 to mm8, mm of the virtual storage space 20
10, mm11, ss1 to ss3 of structure 1 and ss of structure 2 of the data file space 22
Reference numerals 1 and ss2 respectively represent individual data cells forming the list structure. The data cell includes a pointer part for linking (associating) another related data cell in addition to the data part to be stored. At this time, a plurality of pointers may be stored in the data cell. In FIG. 4, the pointer is also stored in the data section of each data cell. Arrows in FIG. 4 represent links between data cells, and dashed arrows represent links to data cells not shown.

Further, mm3 and m of the virtual storage space 20
m9 indicates an indirect pointer. This indirect pointer points the data cell of the virtual storage space 20 to the data file space 22.
Is a pointer for linking to the data cell of. Further, in the data file space 22, a list-structured data cell is stored in units of a structure composed of a plurality of data cells having a dense link structure. This structure has a virtual storage space 20 as described later.
The data transfer between the data file space 22 and the data file space 22, that is, a unit of structure in / structure out.

In the list-structured data shown in FIG. 4, links between data cells in the virtual storage space 20 (for example, links from mm1 to mm2), and
A link from a data cell in the data file space 22 to a data cell in the virtual storage space 20 (for example, a link from ss2 to mm4 in structure 2).
Is expressed by using an address in the virtual storage space 20 as a pointer. Further, the link from the data cell in the virtual storage space 20 to the data cell in the data file space 22 (for example, the link from mm3 to ss1 of the structure 1) is an indirect link in the virtual storage space 20 described above. It is expressed using a pointer (mm3 in this case).

Further, links between data cells in the data file space 22 (for example, s
The link from s1 to ss2) is expressed by using an address in the structure as a pointer. As a result, the data length for address expression in the structure can be shortened. In this way, since the representation of the pointer is different between the virtual storage space 20 and the data file space 22, data transfer is performed between the virtual storage space 20 and the data file space 22 in units of structure, as will be described later. In this case, the overhead of converting the representation of the pointer occurs, but as described later, the external memory 6
By sufficiently reducing the frequency of data transfer between the main memory 4 and the main memory 4, the processing efficiency of the entire computer system can be improved.

A link between the data cell in the virtual storage space 20 and the data cell in the data file space 22 described above,
And, the links between the data cells on different structures of the data file space 22 are the following three types of reference tables, that is, a structure reference table (SRT) and an XRT.
(External Lookup Table) and Reverse Lookup Table (IRT). The SRT, XRT, IRT are stored in a predetermined area on the main memory 4, for example.

FIG. 5 is a diagram for explaining the SRT. As shown in FIG. 5, the SRT, for each structure (str.id) of the data file space 22, has a reference count (refcounter) to the structure, a pointer to the structure (str.pointer), and a pointer to the external reference table ( XRT.pointer)
The data is in a table format and has a reference to a hash method or the like. Reference counter (refcount
ter) indicates the number of times a data cell existing in the structure is referenced by another structure or a data cell in the virtual storage space 20, and is used in the garbage collection process described later. For example, str
If there are N references to the data on 1 (Structure 1), the reference counter for str1 of the SRT is "N". Pointer to structure (str.p
(pointer) indicates the address of the structure in the data file space 22. For example, the pointer to str1 is "STR1". The pointer to the external reference table indicates, for example, an address to the external reference table stored in the main memory 4. For example, the pointer to the external reference table of srt1 is "XRT1".

An external reference table (XRT) is created for each structure of the data file space 22, and data cells on the main memory 4 to which data cells on the structure refer (link) and other data files on the data file space 22 are referenced. It has information about the data cells on the structure. This information is, for example, as shown in FIG. 5, a pointer (Ptr.id) of the data cell of the link source, an area in which the data cell of the link destination is stored (for example, the main memory 4,
Information (dstar) that identifies structures 1, 2, etc.
ea) and the pointer (d) of the data cell of the link destination
st. id).

For example, the data cell ss3 on the structure 1 of the data file space 22 shown in FIG. 4 is linked to the data cell mm6 on the virtual storage space 20. As information regarding this link, XRT1 is shown in FIG.
, "Ss3" as "ptr.id",
“Main” (indicating the virtual storage space 20) as “dstrea” and “mm” as “dst.id”
6 ”.

Similarly, ss2 on structure 2
5 to X4 on the virtual memory space 20 and information on the link from ss2 to ss2 on the structure 1 in XRT2 as shown in FIG.

Further, the external reference table also manages the dereference (referenced) of the data cell on the structure from the data cell on the other structure. For example, as shown in FIG. 4, ss2 on str1 is linked from ss2 on str2, and this link corresponds to the data cell linked from another structure on the data file space 22 in XR1 shown in FIG. Pointer (dst.id),
It is managed by information (ptrarea) that specifies the structure in which the data cell of the link source is stored, and a pointer (ptr.id) of the data cell of the link source.

One reverse reference table (IRT) is used for the entire system, and manages the referencing from the structure existing in the data file space 22 of the data cell existing in the virtual memory space 20. As shown in FIG. 5, the IRT specifies the data cell in the virtual storage space 20 referred to by the structure existing in the data file space 22, dst. i
d and XRT.D indicating a pointer to the structure that is the link destination (reference destination). and a pointer. In FIG. 4, mm6 on the virtual storage space 20 is str1 and s.
It is referred from tr2, and as shown in FIG.
XRT1 and XR indicating pointers to and str2
Has T2.

As described above, in the data management method of the present embodiment, since the links between the data cells are managed by using the reference table such as SRT, XRT, IRT, etc., the virtual memory space 20 and the data file space 22 are linked. If the information about the link is changed in the reference table when the data is transferred in the unit of structure in, the link of the link is actually accessed without actually accessing the data cell in the data file space 22 requiring a long access time. You can make changes.

Next, a data management method for transferring the list-structured data stored in the data file space 22 to the virtual storage space 20 will be described. In the arithmetic processing of the CPU 2, when necessary data is not stored in the virtual storage space 20 or when the virtual storage space 20 has a sufficient free area, the necessary data is stored from the data file space 22 to the virtual storage in structure units. Transfer to space 20.

For example, when the data of the link structure is stored in the virtual storage space 20 and the data file space 22 as shown in FIG. 4, all the data cells stored in the structure 1 or the structure 2 are stored. All data cells stored in the virtual memory space 20 are transferred to the virtual memory space 20. For example, the data file space 22 shown in FIG.
When all the data cells stored in the structure 1 of the above are transferred to the virtual storage space 20, the data link structure in the virtual storage space 20 and the data file space 22 is changed as shown in FIG. In FIG. 6, ss1, ss2, and ss3 on the structure 2 shown in FIG.
0 are stored as mm1 'and mm2'mm3', respectively. At this time, in FIG. 4, indirect pointer mm3
The link from mm2 to ss1 that has been realized through is changed to a direct link from mm2 to mm1 ′ as shown in FIG. When the transfer process ends, the data cell of the structure 1 exists in the virtual memory space 20a represented in the main memory 4 of the virtual memory space 20.

FIG. 7 shows SRT, XRT, IR after transfer.
It is a figure for demonstrating T. After the transfer, since the structure 1 does not exist in the data file space 22, the information corresponding to the structure 1 is not described in the SRT, and neither XRT1 nor IRT1 exists. Further, in the XRT 2 shown in FIG. 7, s on the structure 1 before transfer
Since s2 becomes mm2 ′ in the virtual storage space 20 after the transfer, dst. mm2 'is described as id.

Further, when it becomes necessary to create a predetermined free area in the virtual storage space 20, a series of data having a link relationship is transferred from the virtual storage space 20 to the data file space 22, That is, structure out. Therefore, when data is transferred from the virtual storage space 20 to the data file space 22, an address in the virtual storage space 20 in which no data exists is worm-eating. Data file space 22 to virtual storage space 2
When data is transferred to 0, the transferred data is stored so as to fill the above-mentioned worm eating state of the virtual storage space 20. Further, in order to improve the worm-eating state of the data existing in the virtual storage space 20, a process of dividing the virtual storage space 20 into two and using it is performed as described later.

As described above, according to the data management method of this embodiment, the data having the link structure required for the processing of the CPU 2 can be efficiently stored in the main memory 4, and the processing efficiency of the computer system can be improved. Can be improved. Further, the data management system of this embodiment can be realized by using the storage management of the paging system used in a normal computer system and is excellent in versatility.

The second embodiment will be described. The data management method in the computer system of the present embodiment is a time-division method in which the processing of the above-described data management method of the first embodiment and the processing of garbage collection, which is a work of collecting unused pointers, that is, garbage collection, are performed. Use and run in real time. That is, in the data management method of this embodiment, an interrupt occurs at a predetermined cycle, and the processing of the data management method described in the first embodiment and the processing of garbage collection are alternately performed. When an interrupt occurs and the process being executed is interrupted, the process state at the time of the interrupt occurrence is stored, and when the interrupt occurs next time, the process is restarted from the previously stored process state. Further, the garbage collection process includes a process of incremental garbage collection that is rich in immediacy, and a mark-and-sweep garbage collection process that can collect the garbage associated with the circular join and the overflow of the reference counter. The process of incremental garbage collection is prioritized over the process of mark and sweep garbage collection.

The garbage collection process will be described in detail below. The garbage collection is processed in the virtual memory space 20 in units of data cells, and in the data file space 22 in units of structures using a reference table. SRT (structure management table) is used to count and mark the reference of the structure in the data file space 22. The reference from the structure on the data file space 22 is traced using the XRT (external reference table). Therefore, it is not necessary to access the actual data cell stored in the external memory 8, and high-speed processing is possible. The data cell in the virtual storage space 20 collected by the garbage collection is used as a free cell. Free cells exist only in the virtual storage space 20. When the structure on the data file space 22 becomes garbage, the data file space 22 is released. In the garbage collection in the data management method of this embodiment, the structure on the data file space 22 can be treated as one object. In the following description, the data cell in the virtual storage space 20 and the structure in the data file space 22 are collectively expressed as an “object”.

As shown in FIG. 8, the garbage collection includes an incremental garbage collection (GC) and a mark and sweep garbage collection (GC). As the garbage collection in the data management system of this embodiment, an incremental GC based on a reference count or multiple reference marks is basically used. Incremental GC is suitable for real-time garbage collection because it has good immediacy from generation to collection of garbage and usually does not have a large amount of execution once. However, in the incremental GC, the data of the ring structure,
Objects whose reference counter overflows and objects that have multiple references cannot be collected. for that reason,
Execute Mark and Sweep GC less frequently.

In the list data storage configuration method, a method of counting the reference from the control stack of the list processing on the object or a delay type of not counting is possible. In FIG. 1, when the delayed incremental GC is performed, a zero reference table indicating an object that is no longer referred to is used, the zero reference table is managed, the reference from the stack or the like is counted at the time of garbage collection, and the garbage collection is executed.

In order to obtain real-time property, the time for collecting or marking one time after another is controlled. Further, the interval of the garbage collection execution or the ratio of the garbage collection execution time within the monitoring time is controlled so that the execution of the garbage collection is not temporarily concentrated in the whole process. That is, the incremental GC is executed in real time time division. Mark and sweep G
Since C does not have to be executed frequently, its execution interval is controlled to be longer. When the garbage collection that operates in a time-sharing manner is interrupted, the processing is restarted next, and therefore the intermediate state is retained. In this embodiment, since the two types of garbage collection, that is, the incremental GC and the mark and sweep GC, operate in parallel, the two may conflict with each other regarding the collection of objects, but they are held when the garbage collection is restarted. This conflict is resolved by checking the changes in the state during the process.

In storage management such as the paging method,
If there are no free cells during list processing, starvation occurs. However, in the data management method using the exact data storage configuration method of the present embodiment, a new free cell can be obtained by the structure out of the data as long as the secondary storage permits, so a starvation state in such a form does not occur. . In the list data storage configuration method, when the garbage that is not collected occupies the virtual storage space 20, it becomes starved. Therefore, the execution frequency and time of each garbage collection may be set so that the number of uncollected objects does not increase constantly, and the number of objects that cannot be collected and has an overflow of reference structure count or reference count.

In the incremental GC, among the objects that are no longer referred to, those that are in the process of collection are managed by the "object collection stack". When you restart garbage collection after it has been suspended, you can start collecting it from the objects on that stack. Garbage collection can be restarted by, for example, the appearance of new objects that are no longer referenced, the increase of uncollected objects, the increase of zero reference objects in the case of delay type, or the function that consumes free cells, for example, calling cons. . When the garbage collection is restarted, the uncollected objects on the stack may have already been collected by the mark and sweep GC, or may be used as new objects by the list processing. In this case, do not retrieve from that object. This can be judged by changing the value in the case of the reference counting method, or by specially marking uncollected objects before stacking them in the case of multiple references. However, in the data management method of this embodiment using the list data storage configuration method, the incremental GC with good immediacy is prioritized over the mark and sweep GC that may be executed less frequently. Therefore, the side effects on the object collection stack due to Mark and Sweep GC need not be considered.

In the mark and sweep GC, an object in the middle of marking is managed by placing a pointer to the object in the "marking stack". During the processing of mark and sweep GC, the list processing
For example, the function rplacd allows the data to determine its necessity by reference count or multiple reference marks. When the data is unnecessary, it is collected by incremental GC. If the detached object is referenced from the stack and may be used later, put it on the marking stack for later marking. A data cell in the virtual storage space 20 created by the function cons or structure-in and a structure in the data file space 22 created by structure-out are marked at the time of creation. Two types of marks are prepared, switched at the beginning of the mark phase, and used alternately for each set of marks and sweep GC. The free cell is marked with the free cell to avoid collection. During the suspension of the mark and sweep GC, the data cell in the virtual memory space 20 managed on the marking stack can be made a free cell by the incremental GC or the structure out, or can be reused by the function cons or the structure in. There is a nature. Further, there is a possibility that the structure on the data file space 22 becomes garbage due to the incremental GC and is collected. For this reason, the object to be placed on the marking stack is marked to recognize the side effect, and the continuation of the processing is determined.

Management of the execution time of garbage collection in real time can be performed without burdening the list processing and the execution of garbage collection by using the interrupt function on the architecture having a timer or the like. Even if the execution time of garbage collection is controlled by the number of objects to be collected, it can be processed at high speed in the processor, so it does not cause a large overhead. To avoid placing multiple pointers to the same object on the marking stack, mark the object before placing it on the stack. Therefore, the marking operation for recognizing the side effect during the garbage collection interruption can be executed at the same time.

When the structure is rewritten by the list processing, the separated data can usually be recovered by the incremental GC. Therefore, the overhead of the marking accompanying the change of the data structure is larger than that when the mark and sweep GC are operated independently. It can be small. Further, since the mark and sweep GC are executed less frequently, this overhead is small in the entire processing. Marking of created objects and free cells can also be processed simultaneously in the processor when constructing its object data.
In the data management method of the present embodiment to which the list data storage configuration method is applied, the starvation state due to the shortage of free cells does not occur, but in order to improve the structure-in correspondence, the data management method of the present embodiment is also constant. Try to secure a quantity of free cells. Securing a free cell causes a decrease in the usable amount in the virtual storage space 20, and thus becomes an overhead in real time.

The data management system of this embodiment requires reference counts or multiple reference marks for the data cells in the virtual storage space 20 and the structure in the data file space 22 for structure management. The efficient realization of real-time garbage collection in the data management method using the list data storage configuration method depends largely on the incremental GC, but the incremental GC can be realized by using this referenced information. The real-time garbage collection in the data management method of this embodiment reduces the use efficiency of the virtual storage space 20, but there is almost no overhead of real-time realization. According to the data management method of the present embodiment, in addition to the effect described in the data management method of the first embodiment described above, by performing the garbage collection in real time, it is possible to efficiently remove the garbage. As a result, the data required for the processing of the CPU 2 can be efficiently stored in the main memory 4, and the processing efficiency of the computer system can be improved.

The third embodiment will be described. In the data management method of this embodiment, in addition to the data management method of the first embodiment described above, the virtual storage space 20 shown in FIG. 4 is divided into two, and the divided spaces are alternately used to create a list structure. Manage the data of. FIG. 9 is a diagram for explaining the data management method of this embodiment. As shown in FIG. 9A, the virtual memory space 20 is divided into a virtual memory space 24 and a virtual memory space 26. For example, the virtual memory space 24 is used in the process currently being executed by the CPU 2, and the virtual memory space 24 is used. 26 is not used. That is,
The virtual storage space 24 shown in FIG. 9A is used as the virtual storage space 20 shown in FIG. 4, and the storage management described in the first embodiment is performed between the virtual storage space 24 and the data file space 22.

When the CPU 2 is executing a process using the data existing in the virtual memory space 24 and there are no free cells in the virtual memory space 24, the CPU 2 changes the data cell in use to the data cell in use. Copies into virtual memory space 26 with one or more data cells associated with This copying ends, for example, when all the data cells that are directly or indirectly referenced from the data cell in which the CPU 2 is executing the processing are copied to the virtual storage space 26. For example, as shown in FIG. 9B, when the data cell 40 existing in the virtual storage space 24 shown in FIG. When the free cell runs out,
The data cell 40 and the data cells 42 and 44 directly and indirectly linked from the data cell 40 are copied to the virtual storage space 26, and the data cells 40a, 42a and 44a existing in the virtual storage space 26 Become. The data cells existing in the virtual memory space 24 are copied to the virtual memory space 26 so as to enhance the locality of addresses in the virtual memory space 26. That is, the data cell 40a,
42a and 44a exist at virtual addresses adjacent to each other in the virtual storage space 26.

When the copying is completed, the CPU 2
Performs processing using data existing in the virtual memory space 26, and unused data cells existing in the virtual memory space 26 are used as free cells. As mentioned above, C
Address locality of the data cell used in the process being executed by the PU2, from the virtual memory space 24 to the virtual memory space 26, together with the data cell directly or indirectly linked to the data cell. By performing the copying so as to obtain p.p.m., the frequency of paging between the main memory 4 and the external memory 8 can be reduced.

In the above-mentioned copying process, when the data cell directly or indirectly linked from the data cell used in the process being executed by the CPU 2 exists in the data file space 22 shown in FIG. 4, that is, , C
When there is a direct or indirect reference to a data cell in the data file space 22 from a data cell used in the process being executed by the PU 2, the marking process described in the second embodiment is performed in the data file space 22. It may be performed for the structure. By performing the marking process in this way, when the above-described copying process from the virtual storage space 24 to the virtual storage space 26 is completed, the sweep process described in the second embodiment is performed in the data file space 22. It can be executed in units of structures. The original data processing in the CPU 2 is continuously performed after the copying and sweep processing are completed.

As described above, according to the data management method of the present embodiment, the above-mentioned copy processing is performed in the virtual storage space 20 to reduce the frequency of paging, and the garbage collection described in the second embodiment can also be executed. .

Next, a method of operating the original processing in the CPU 2 and the above-mentioned copying processing in real time using the time division method will be described. In the data management method of the present embodiment, the original processing in the CPU 2, the copying processing in the virtual storage space 20 described above, and the mark and garbage collection processing in the data file space 22 are performed in real time using a time division method. To work with. At this time, in this embodiment, the copying process and the CPU in the storage space are performed.
It is disclosed to operate in real time using the time division method with the original processing in, for example, Henry G. Baker, Jr: L
ist Processing in Real Time on a Computer, Communic
By applying the data management method of ations of the ACM, vol.21, no.4, pp.280-294, 1978, the copy processing in the virtual memory space 20 and the original processing in the CPU 2 are performed in real time in a time-division manner. To work with.

For example, referring to FIG.
Is running a process using the data existing in the virtual memory space 24, if there are no free cells in the virtual memory space 24,
The copying process described above is operated in a time-division manner. When the copy process is performed by the time division method as described above, the copy process may be interrupted before the copy process is completed. However, the data cells copied to the virtual memory space 26 are the virtual memory space 26. Since the addresses are arranged in the order of the addresses above, when the copy process is started next time, the restart state of the copy process is determined by the copy position of the virtual storage space 26 when the copy process was interrupted last time. To be done.

When a data cell to be processed by the CPU 2 exists in the virtual memory space 24 after the copying process is started, the data cell is moved from the virtual memory space 24 to the virtual memory space 26, and the moved data is moved. The CPU 2 continues the process by using the cell. At this time, the addresses in the virtual storage spaces 24 and 26 for recognizing the interruption of the copying process and the restart of the copying process are converted according to the movement of the data cell. In the virtual memory space 26, the free cells are obtained from the direction opposite to the direction in which the free cells copied from the virtual memory space 24 to the virtual memory space 26 are arranged. That is, when copying a data cell existing in the virtual memory space 24 from a position with a large address number in the virtual memory space 26 to a position with a small address number,
A free cell exists in a position of the virtual memory space 26 having a small address number.

In the data management method of this embodiment, the CPU 2
In addition to performing the original processing in step 2 and the copy processing using the time division method, the data file space 22
The mark-and-sweep-garbage-collection process in is operated in real time using the time-sharing method.

As described above, according to the data management method of this embodiment, the locality of data at the address in the virtual memory space 20 can be enhanced, and paging performed between the main memory 4 and the external memory 8. The frequency of
The processing efficiency of the computer system is improved.

In the first to third embodiments described above, the data file space 22 can be expanded according to the increase in the data amount as long as the hardware resources allow. Further, if the data used in the data management method of the present embodiment occupies the entire virtual storage space, other application software and the like cannot operate. Therefore, generally, a predetermined number of pages in the entire virtual storage space are used. Is used as a virtual storage space used in the data management method of this embodiment. Therefore, the CPU
The number of pages to be allocated as a virtual storage space used in the data management method of the present embodiment may be increased or decreased among all the virtual storage spaces depending on the content of the processing in step 1 and the load state. For example, when it is desired to increase the number of pages of the virtual storage space used in the data management method of this embodiment, the area of the increased page can be used as a free cell. Further, when it is desired to reduce the number of pages in the virtual storage space used in the data management method of the present embodiment, the data stored in the page to be deleted is virtualized in units of structure using the link structure data management method described above. Move from storage space to data file space,
The page to be deleted is deleted from the virtual storage space.

[0064]

As described above, according to the data management method of the present invention, the data management method of the link structure is performed by expanding the generally widely used virtual storage method to improve the processing efficiency of the computer system. Can be improved. That is, the virtual storage space and the data file space of the data management method of the present invention can be easily realized by using the virtual storage space and the data file space already realized on a general computer system as they are. As a result, artificial intelligence information processing, natural language processing,
By using a computer system to which the data management method of the present invention is applied, processing such as mathematical expression processing can be performed more efficiently than when a conventional computer system is used. Further, according to the data management method of the present invention,
The garbage collection process, which is indispensable for link structure data processing, can be efficiently performed with a lower access frequency to the second storage means than when a virtual storage system such as a paging system is used alone. Further, according to the data management method of the present invention, when the dynamic signal analysis and the diagnosis in the manufacturing process, or the processing requiring the real time such as the human interface is performed, the processing of the data of the link structure and the garbage collection thereof are performed. Can be operated in real time by using a time division method in which a simple process of suspending and resuming these processes is added. Further, according to the data management method of the present invention, the locality of data in the virtual storage space can be increased by dividing the virtual storage space into two parts, and the frequency of paging processing can be reduced. As a result, the processing efficiency of the computer can be improved. Further, according to the data management method of the present invention, by dynamically changing the size of the virtual storage space to be managed using the data management method of the present invention among all virtual storage spaces, The processing efficiency of the system can be improved.

[Brief description of drawings]

FIG. 1 is a configuration diagram of a computer system using a data management method according to a first embodiment.

FIG. 2 is a diagram for explaining a virtual storage space and a data file space that represent data stored in a main memory and an external memory according to the first embodiment.

FIG. 3 is a diagram for explaining a virtual address of a virtual storage space according to the first embodiment.

FIG. 4 is a diagram for explaining data expressed in a virtual storage space and a data file space according to the first embodiment.

5 is a diagram for explaining SRT, XRT, and IRT in the state of FIG.

FIG. 6 is a diagram for explaining the data represented in the virtual storage space and the data file space after the data in the data file space is transferred to the virtual storage space.

FIG. 7 is a diagram for explaining SRT, XRT, and IRT in the state of FIG.

FIG. 8 is a diagram for explaining garbage collection performed by the data management method according to the second embodiment.

FIG. 9 is a diagram for explaining a data copy process in a virtual storage space performed by the data management method according to the third embodiment.

FIG. 10 is a diagram for explaining a paging method in a conventional computer system.

[Explanation of symbols]

 2 ... CPU 4 ... Main memory 6 ... Input / output circuit 8 ... External memory 10 ... External bus 12 ... Internal bus 20, 20a, 20b ... Virtual storage space 22 ... Data file space 24, 26 ... Small virtual memory space 30 ... Virtual address 32 ... Page allocation table 32a ... Identification field 32b ... Page address field 34 ... Main memory address 40, 42 , 42a, 44, 44a ... Data cells

Claims (8)

[Claims] 1. A first storage means that can be accessed at high speed by an arithmetic control means, and a second storage having a large storage capacity capable of saving data stored in the first storage means. A data management method for managing data processed by the arithmetic and control unit and having a changed link structure in a computer system having a means and an arithmetic and control unit, the virtual storage space comprising the first storage unit and the second storage unit. And a data file space is expressed in the second storage means, and data transfer between the first storage means and the second storage means is performed according to the processing of the arithmetic control means. For the data expressed in the virtual memory space, the page created based on the address in the virtual memory space is used as a unit, and is expressed in the data file space. Data, between the data file space and the virtual storage space represented in the first storage unit, with the structure composed of the data and the data directly or indirectly linked to the data as a unit. A data management method characterized in that 2. Out of data stored in the first storage means and the second storage means, unused data is recovered in units of data for data existing in the virtual storage space. The data management method according to claim 1, wherein the data existing in the data file space is collected by the structure as a unit. 3. The data management method according to claim 2, wherein the processing of the data transfer and the garbage collection is performed in a time division manner. 4. The garbage collection includes an incremental garbage collection that collects garbage using a reference counter and a mark-and-sweep garbage collection that marks traces used cells and sweeps unused cells. The data management method according to claim 2, wherein the garbage collection is performed prior to the mark-and-sweep garbage collection. 5. The virtual storage space is divided into at least two small virtual storage spaces, and the data existing in one of the small virtual storage spaces is accessed by the data transfer processing and the arithmetic control means, and the small virtual storage space is accessed. When there is no free data in the storage space,
The data being accessed from the arithmetic control unit and the data directly or indirectly linked to the data are copied to the other divided small virtual memory space, and the copied small data is copied. 5. The data management method according to claim 1, wherein the data transfer processing and the access from the arithmetic control means are performed with respect to the data existing in the virtual storage space. 6. The data management method according to claim 5, wherein mark-and-sweep garbage collection is performed in the data file space. 7. The data management method according to claim 5, wherein the data transfer process, the copy process in the virtual storage space, and the garbage collection process are performed in a time division manner. 8. The data management method according to claim 1, wherein the number of pages created in the virtual storage space is changed according to the frequency of processing the data.