JPH0844686A - Data management system - Google Patents

Abstract

PURPOSE:To improve the transportability of a program by providing one data managing computer to manage shared data and copying the shared data from the managing computer when accessing those data. CONSTITUTION:This system is composed of processors 1-4, communication medium 5 for connecting the processors 1-4, and inter-CPU shared memory 6 to be used for exchanging data for tasks between the processors 1 and 2. Then, the centralized management of shared data in the entire distributed system is performed by a name table 13 and a name managing means 12 of one data managing computer (processor 1) but at the time of accessing the shared data at the other computers (processors 2-4), the name table 13 and data contents are copied by the data managing computer 1 in the case of starting the system so that the shared data can be accessed not through any network when executing the task of the present computer. The name table 13 stores an attribute such as the name of a shared data area in the entire system, and the name managing means 12 manages the shared data to be accessed by a program corresponding to the name while using the name table 13 or the like.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a distributed system in which a plurality of computers are connected on a network, and to a data management system shared by programs of a plurality of computers on the distributed system.

[0002]

2. Description of the Related Art Conventionally, NFS (Network File System) has been used as a method for sharing data between computers in a distributed system.
em) and distributed DB (Data Base). This kind of management method requires inserting a procedure such as Open for the file or DB to be accessed in the program when accessing the data,
In the writing process, data is read / written via the network to the computer that manages the data, so the network load and the load of the computer that manages the data greatly affect the execution performance of the remote program. . Especially, in order to share real-time data of plant control requiring real-time property between computers, access performance becomes a problem.

In addition, the range in which the memory on the main memory is shared between programs and the data is transferred is sh in UNIX.
With ared Memory and Memory Mapped File, it is effective only for the program in the computer, and it is impossible to access it across computers.

In Japanese Unexamined Patent Publication No. 2-136954, for shared memory access, two networks are connected to each computer, one is used for address information and read command information at the time of shared memory access, and the second one is used. By using it for reading and writing the contents of actual data, the performance at the time of accessing the shared memory is improved.

[0005]

[Problems to be Solved by the Invention]

(1) In recent years, due to the speeding up of hardware and networks, the high recording density of memories, etc., and the cost reduction of hardware for single performance, high-performance computers are connected via a network, and the functions and loads of the entire system are increased. There is a proliferation of organic and flexible dispersion systems that disperse water.

Also in the plant control system, there are many plans for migrating an existing system to a distributed system environment. The feature of the distributed system is that the performance of the entire system can be easily improved by newly connecting the computer to the existing system as the system scale expands. In recent years, it has become easy to connect not only a system for controlling a plant but also a field as an information system by connecting a personal computer or a workstation. In particular, it is considered that the openness of PCs and WSs will advance, the ease of connection to the system and the spread of easy-to-use word processing software and spreadsheet type software will further advance the computerization of the system.

Further, in such a distributed system, since a computer can be easily incorporated into the distributed system, the system is remodeled more frequently than the conventional system, so that an efficient development environment is required. However, in the plant control system, since the system is normally controlled by operating for 24 hours, the system cannot be stopped frequently every time the system is remodeled. Therefore, an arbitrary time period is set and the system is remodeled during the period. . However, due to the large scale of the system,
It will be difficult to stop all the systems at once and carry out additional modifications. In particular, when migrating or porting software, it must be incorporated in the system in consideration of synchronization with programs and data that are being executed online in the existing system.

By the way, when the plant control program structure is analyzed, a plurality of programs are activated by real-time events, and common read / write among the programs is performed.
Each plant is controlled by storing and referring to the plant data in a writable memory. That is, the plant control is composed of events, data, and programs.

Considering the characteristics of the plant control program from the above program structure, when the distributed system is remodeled, the events that rise asynchronously from the plant and the programs that run on the newly added computer are stored in hardware. Since it depends on the system, there is little influence on the system, but it is difficult to transfer or switch the real-time data to the newly connected computer because the control data is directly related to the real-time control of the existing system. , Significantly hinders system scalability and migration.

Therefore, the present invention improves the portability of programs by making real-time data transparently accessible by all computers in the system.

(2) Further, the shared memory in the distributed system includes a common memory between CPUs directly connected to a computer found in a multi-computer and a transfer memory for transferring the memory of its own computer to the memory of another computer through a network. . Since these are provided by each company's unique interface and are not uniform, it is difficult to port the program to other companies' computers, and the data address etc. is changed, so it is necessary to recompile the program.

According to the present invention, portability is improved by allowing different heterogeneous shared memories to handle the same data without changing the program.

(3) According to the present invention, in accessing the shared data, the existence of the shared data to the data management computer and the processing of the program such as data read / write are performed by the data management computer via the network. The shared data access performance of the real-time control program is improved by performing it on each computer instead of accessing it.

That is, the shared data entity is centrally managed by one computer, but the data access for the task is to close it within the own computer.

(4) In a distributed system, not only products from one manufacturer but also products from multiple manufacturers will be incorporated in the system in the future, and computers with completely different architectures will be mixed in one system. Therefore, when a plurality of programs as described above access the same data, it becomes impossible to access the same address due to a difference in data alignment. For example, some computers access the memory only at 2-byte boundaries, some can only access memory at 4-byte boundaries, and some have address boundaries determined by the data access length determined by a hardware instruction ( Natural boundary).

In the distributed system as described above, when the same data is accessed from individual computers, it cannot be accessed unless the procedure is suitable for each computer architecture. Even if it can be accessed, it will destroy the data.

In the present invention, data can be shared between different architectures by changing the data accessible by the program of each computer into a structure so that data can be shared between computers having different architectures. To be realized.

(5) In the distributed system, computers having heterogeneous shared memories often coexist, and it has been difficult to share these data. According to the present invention, the expandability of the system is improved by making it possible to share the contents of the heterogeneous shared memory in the distributed system even by a computer having no shared memory.

[0019]

In the present invention, the above (1),
In order to solve the problem of (2), it is characterized by having the following means.

(1) One data management computer for managing all shared data of computers in the distributed system is provided in the system, and the attributes of the shared data such as the name of the shared data in the distributed system, the allocation memory type, and the size are centralized. A name table that is stored in memory and a name management unit that manages the name table are provided.

(2) The program is provided with address resolution request means for solving the address of the shared data at the time of execution.

(3) In accessing shared data from a program, indirect addressing access via an index table is used in order to easily change the mapping to different types of shared memory.

(4) Data allocation means for allocating shared data that can be shared between computers is provided for a computer that does not have a memory that can be shared between multiple computers.

In order to solve the above-mentioned problem (3),
It is characterized by having the following means.

(5) When accessing shared data that is not assigned to the own computer, the first access can be copied from the data management computer, and the second and subsequent accesses can be resolved within the own computer without going through the network. Name management means and name check request means are provided.

(6) A name table copy means for copying the name table from the data management computer when the system is started up is provided.

(7) To access the shared data of the program, a data allocating means for making a memory access in the own computer and a data matching means for matching the shared data contents are provided.

In order to solve the above problem (4),
It is characterized by having the following solution.

(8) A memory characteristic storage table storing the characteristics (alignment or byte order) of the shared memory of each computer is provided.

(9) An area for storing the data type to be accessed is provided in the name table.

In order to solve the above-mentioned problem (5),
It is characterized by having the following solution.

(10) A shared data allocation memory type for storing the type of shared memory for each computer is provided in the memory characteristic storage table.

(11) The data allocating means is provided with a process for determining to which shared memory data is allocated according to the type of shared data allocation memory.

[0034]

(1) Since the shared data on the heterogeneous shared memory can be accessed by one name without changing the program, the portability of the program is improved.

(2) Since it is sufficient to register the shared data with one computer, the maintenance of the shared data of the entire system is easy.

(3) The existence of shared data can be checked without passing through the network, and the performance at the time of accessing the shared data is improved.

(4) Since data can be shared between computers having different architectures, the connectivity with computers of other companies is improved.

(5) Even computers that do not have the same type of shared memory can share data on different types of shared memory with a single name, so connectivity with other manufacturers and computers with different architectures is improved. It will be easier and system scalability will be improved.

[0039]

【Example】

<Embodiment 1> FIG. 1 shows a system configuration example of a distributed system according to an embodiment of the present invention.

The present distributed system includes processing devices 1, 2, 3,
Communication medium 5 for connecting 4 and 4 processing devices 5, processing device 1
Used by the task between the processor and the processing unit 2 to transfer data
CPU shared memory 6, transfer memory used by tasks between processing devices 1 and 3 for data transfer, local shared memory 8 used by tasks within one processing device for data transfer, and program Shared data entity areas 6-1, 7-1, 7-3, 8-1, 8-2, 8-3,
An index table 9 for storing the start address of 8-8, a task 10 for controlling the plant by sharing the data between the processing devices, and an address solving means for performing a process for obtaining the start address of the shared data area used in the tax. 11, a name table 13 that stores attributes such as the name and data size of the shared data area of the entire system, a memory characteristic storage table 14 that stores the characteristics of the shared memory,
Name management means for managing shared data accessed by a program by name using the name table and the memory characteristic storage table, and a data entity area for the CPU based on management information of the name table in the name management means.
Data allocating means 15 for allocating to the inter-shared memory, the transfer memory, and the local shared memory, and storing the start address of the data in the index table,
A computer attribute storage table 16 for storing the attributes of each processing device in the distributed system, a name check request transmission means 17 for confirming to the processing device 1 whether shared data accessed by a task exists in the distributed system, A name check request receiving means 18 for receiving a name check request from the name check request transmitting means and searching for a designated shared data record in a name table in the processing device 1, between the CPU and a processing device having no shared memory or the transfer memory. In the case of sharing data in the processing device, it comprises a data matching device 19 for matching the contents of the local shared memory in the processing device, and a data registration device 20 for registering the data shared by the tasks in the distributed system in the processing device 1. It In this distributed system, the computer that centrally manages the shared data of the entire system is the processing device 1. Hereinafter, the processing device 1 will be referred to as a data management computer.

Hereinafter, this embodiment will be described by dividing it into the following configurations.

1. Basic table configuration and link processing in shared data link (1) Name table configuration (2) Index table configuration (3) Task shared data link processing 2. Shared data address resolution procedure and data registration procedure in data management computer (1) Data registration processing procedure (2) Task address resolution means 3. Address resolution procedure via network (1) Address resolution means of shared memory between CPUs (2) Data management method of local shared memory (3) Content matching processing of local shared memory 1. Basic table structure and link processing in shared data link (1) Name table structure Fig. 10 shows the name table structure for managing shared data. The elements stored in the name table are the shared data name 1301, the shared data byte size 1302, the data type 1303 for accessing the shared data, and the allocation memory type 1304 indicating the type of the shared memory in which the shared data is allocated. An address storage area 1305 for storing the start address of the record in the index table 90 corresponding to the name of the shared data, and a write flag 13-6 indicating that the task has written to the shared data. The allocation memory type is a transfer memory in which the memory contents in the CPU shared memory (GM) used for data transfer in a multi-computer system and the hardware in the network system are reflected in the memory in another computer at a constant cycle. There is a memory (CM) and a local shared memory (LM) that can share tasks in one computer.

(2) Structure of Index Table FIG. 9 shows the structure of the index table. In the index table, one record 901, 902, 903 is assigned for each name unit of the shared data in the name table, and the shared data entity (local shared memory 904, transfer memory 905, CP secured by the data allocating means 15).
The head address of the shared data allocated on the U-shared memory 906 is stored.

By using this index table method, it becomes possible to change the allocation memory of shared data without recompiling the task.

(3) Link processing of task to shared data FIG. 8 shows a coding example 80 in which the task links shared data in C language. In order to indirectly address access to the shared data, the variable type of the shared data is declared as a pointer type variable 801. In this embodiment, a datal variable is declared as a character type pointer declaration. Also, the index table 9 is stored in the variable datal.
Start address storage record area 901 of datal in 0
By calling the function DATA_ADDR_SET802 that stores the address of, the shared data with the name datal can be accessed.

2. Shared Data Address Resolution Procedure and Data Registration Procedure in Data Management Computer FIG. 3 shows shared data address resolution means and data registration procedure in the data management computer. FIG. 3 shows data registration means 20 and data allocation means 1 in the data management method.
5, name management means 12, address resolution means 11, name table 13, task 10, terminal 2 for registering shared data
2, the relationship between the local shared memory 8 to which shared data is assigned, the transfer memory 7, and the inter-CPU shared memory 6 is shown.

The following describes (1) shared data registration procedure and (2) shared data address resolution means.

(1) Procedure for registering shared data The registration of shared data is performed by the data registration means 20 from the terminal 22.
However, the name, size, layout memory type, and data type of the shared data are imported 301, the name of the imported shared data,
Name, size, allocation memory type, and data type management means 12
302 to send. The name managing means compares and checks 303 whether the received shared data name already exists with the record name in the name table. If it exists, an error report is sent to the data registering means, and if it does not exist, it is sent to the data allocating means 15. Send a request for allocating shared data to memory together with size of shared data and allocation memory type 30
4. Upon receiving the allocation request, the data allocating means 15 allocates a shared data area to the shared memories 8, 7, and 6 corresponding to the arrangement memory type 309, stores the start address of the area in the index table, and stores the record in the index table. 305: Send the top address to the name management means 12. Upon receiving the head address of the record in the index table, the name management means 12 stores the name, size, allocation memory type, data type, and head address of the record in the index table in the name table 303, and reports the registration completion. Is transmitted to the data registration means 20 (306).

Next, the detailed processing procedure of the name management means 12 and the data allocation means 15 in the shared data registration processing will be described.

FIG. 16 shows a processing flow of the name management means. First, together with the shared data registration request from the data registration means 20, the shared data name, size, allocation memory type,
After the data type name management means receives 1601 and determines a processing request 1602, if it is a registration request, it is checked 1618 if the designated name exists in the name table 13, and if there is, a double definition error report is issued. 1616 is transmitted to the data registration management means, and if not present, a data allocation request is transmitted 1619 to the data allocation means.

Next, FIG. 17 shows a processing flow of the data allocating means which has received the data allocating request from the name managing means.

The data size and the arrangement memory type are received 1701 from the name management means 12, and each shared data memory to be secured is received according to the received arrangement memory type (local shared memory 8, transfer memory 7, CPU shared memory 6). Search free memory of 1703, 1704, 1705,
1706, 1707, 17 for calculating assigned addresses
08. In the local shared memory, if the page of the allocation address that is to be reserved does not exist, a new page is reserved 1709, 1710. Since the transfer memory and the shared memory between CPUs are fixedly allocated in the logical space, it is not necessary to secure a new page, but only the boundary check of the memory is performed.

If the initial value of the shared data exists 1712, the initial data is stored in the shared data area 1713 from the disk. Since only the data management computer (processor 1) that centrally manages shared data loads initial data, it is executed only when the computer attribute in the computer attribute storage table 16 is M: master. The computer attribute storage table shown in FIG. 14 indicates whether or not the computers in the distributed system are centrally managing shared data. In the case of M16001, a master computer for data management and other Calculator is S1
It becomes 6002 (slave).

Next, the empty record in the index table is searched and the start address of the shared data entity is stored 1714. Finally, the head address of the record corresponding to the shared data entity is transmitted to the name management means 12 1
715.

By receiving 1620 the head address of the record in the index table transmitted from the data allocating means, the name managing means judges that the shared data area is secured, and the shared data name, size, and allocation memory are stored in the name table. The head address of the shared data of the type / data type index table is set as one record and stored 1621, and a completion report is transmitted to the data registration means.

Registration processing of shared data is realized by the above procedure.

Regarding the shared data, when the computer is started up, the registered shared data is assigned to the shared memory according to the attributes of the name table.
After 20 systems have been started by the data registration means, it is necessary to allocate them to the shared memory, and all of these processes are performed by the name management means 12.

(2) Shared Data Address Resolution Procedure The shared data address resolution means of the task 10 will be described below with reference to FIG.

First, when executing a task, the address solving means 1
From 1 (task 10 links functions by coding as shown in FIG. 8), the name of shared data accessed by the task is transmitted to name management means 12. Upon receiving the shared data name, the name management means checks the existence of shared data having the same name from the name table, and if it exists, the address resolution request is made to the head address of the record in the index table corresponding to the shared data. 308 to the means. By storing the head address of the record in the index table in a pointer type variable as shown in FIG. 8, shared data can be accessed from tasks.

Next, detailed processing flows of the address resolution means 11 and the name management means 12 are shown in FIGS.

The address resolution processing 1500 called from the task 10 fetches 1501 the name of the shared data passed as an argument of the function and sends 1502 a name search request to the name management means 12 together with the shared data name.

The name management means which has received the name search request,
It is checked whether or not the shared data name for which the search request is made exists in the name table, and if it does not exist, an error report of undefined name is transmitted to the address resolution processing 1614. If it is not a data management computer here, 1604 (computer attribute is slave) sends a shared data name search request to the data management computer (computer attribute is master) 1
604, 1605 and checks the name 1606, 1
607. The name check process is 3. See section.

If the shared data name exists, the contents of the head address storage area 1305 in the name table are sent to the address resolution processing 1613.

Next, in the address resolution processing in which the start address of the record in the shared data index is received 1503, the address is stored in the variable of the shared data and the process returns to the task 1504.

3. Address resolution procedure via network Procedure for address resolution via network (1) CP
The address resolution means of the U shared memory, (2) the address resolution procedure of the local shared memory, and (3) the matching processing of the contents of the local shared memory will be described below. It should be noted that the transfer memory has the same procedure as the shared memory between CPUs, and thus the description thereof is omitted.

(1) Address resolution procedure of shared memory between CPUs CPUs among a plurality of processing devices like a multi-computer.
FIG. 4 shows an address resolution procedure of shared data when data is transferred by the inter-shared memory 6. In a system in which the processing device 1 and the processing device 2 are connected by the communication medium 5 and the CPU shared memory is used for sharing data between the processing device 1 and the processing device 2, the task 10-2 in the processing device 2 is performed.
A procedure for accessing shared data on the shared memory between CPUs from the following is shown below. In such a system, it has conventionally been performed by mutually defining the shared data area of the same address from the processing devices, so that the registration work takes a double labor.

The task 10-2 is the task of 2. By the address resolution requesting means 11-2 in the same manner as the procedure described in the section,
The shared data name is sent 401 together with the address resolution request to the name management means 20-2, and the name management means which has received the address resolution request, as shown in FIG.
Check whether the shared data name exists in the name management table in 1603, 408. 1613, 409 for transmitting the contents of the head address storage area 1305 in the name table to the address resolution requesting means in the same manner as the procedure described in the section.

If the data name does not exist, the computer attribute is checked 1604, and if it is other than the data management computer (the computer attribute is the master), whether the shared data exists in the data management computer (processor 1). Confirm 1
605, 1606. Shared data existence confirmation processing 1605 and 1606 to the data management computer will be described separately. If it does not exist in the data management computer 1607, it sends 1614 a name undefined error report to the address solving means 11-2.

If the shared data exists in the data management computer, the shared data allocation memory type is checked 1608 according to the allocation memory type received from the name check request transmission means 17-2, and the shared memory between CPUs (also the transfer memory). If it is the same), an empty record in the index table is searched and the start address of the shared data entity in the name check response packet is stored 1624 in the empty record. Then, the shared data size 3001, the data type 3002, the allocation memory type 3003, the shared data name 3003 in the received name check response packet 3000, the shared data name, and the start address of the record of the index table are added to the name table of the own processing device 161.
2. In the case of the shared memory between CPUs (similarly to the transfer memory), the data allocation processing is unnecessary again because it is already mapped in the processing device 1, and only a record is added in the name table for task address resolution. Good
406. This makes it possible to access the same address as the address of the shared data on the shared memory between CPUs as seen from the data management computer.

Next, the contents (head address of the shared data index table) 407 of the head address storage area 1305 in the name table are set to the address resolution request means 1.
1613 is transmitted to 1-2. Address resolution request means 11
-2 stores the address of the shared data in the in-task variable and returns to the task 1504, as in the above item 2.

Next, regarding the processing procedure for checking the existence of shared data with respect to the data management computer (processing device 1), the name management procedure flow of FIG. 16, the name check request transmission means flow of FIG. 18, and the name check of FIG. This will be described using the request receiving means flow. The name check request transmitting means receives 1801 the name 402 transmitted from the name managing means 20-2, and transmits a name check request packet 403, 1802 to the name check request receiving means of the data management computer via the network. The structure of the name check request packet is shown in FIG. The structure 2900 of the name check request packet is composed of the data name 2901 of the shared data. Next, the name check request receiving means 18 which received 1901 the name check request packet,
The name table in the data management computer is searched 1902 for the same record as the received shared data name, and a name check response packet is transmitted to the name check request transmission means 17-2 1903. The structure of the name check response packet is shown in FIG. The name check response packet includes a size 3001 of shared data, a data type 3002, an arrangement memory type 3003, and a start address 3004 of the shared data entity.

Next, the name check request transmitting means 17-2, which has received 1803 the name check response packet, displays the contents of the name check response packet in the name managing means 20-2.
1804 to send.

In this way, the task resolves the address of the shared data via the network.

By the address resolution procedure, after the second access, the address resolution processing can be performed without going through the network, so that the access performance of the shared data can be improved.

(2) Data management method of local shared memory FIG. 5 shows a data management method when data is shared between a plurality of processors using the local shared memory. The processing device 1 and the processing device 4 are connected by the communication medium 5, and the CPU
In a system that does not have a shared memory supported by hardware such as an inter-shared memory or a transfer memory, from the task 10-2 in the processing device 2 to the local shared memory 8-1 of the data management computer (processing device 1). The procedure for accessing the shared data of is shown below using FIG.

The shared data address resolution procedure is described in (1) above.
The procedure is substantially the same as that of the above-mentioned item, but the processing after 1606 when the name management means receives the name check response packet from the name check request transmission means is partially different. Only different parts will be described below. If the arrangement memory type is the local shared memory 1608, the size of the shared data and the arrangement memory type are transmitted to the data allocating means 15-4 of the own processing device 2 together with the shared data allocation request 501 1609.
The data allocating means 15-4 allocates a shared data area to the local shared memory 8-4 in the own processing device 2, secures an area for the shared data in the allocation request in the index table 9, and shares the shared data in the area. The head address of the entity is stored, and the head address of the record for the shared data in the index table is stored in the name management means 12-
Send to 4. The name management means 12-4, which has received 1610 the start address of the shared data record of the index table, uses the data in order to match the content of the shared data with the content of the shared data of the data management computer (processor 1). After transmitting 1611 a matching request to the matching means 19-4, the name, size, data type, and allocation memory type of the shared data transmitted from the data management computer are registered in the name table, and further, in the own processing device 4. The head address of the record in the index table that stores the head address of the shared data entity allocated as described above is stored in the head address storage area.

After that, by allocating the shared data and the name table to the shared memory in the own processing device, the communication medium 5
Since the shared data can be accessed from the task without going through, the access performance to the shared data is improved. Moreover, the contents of the local shared memory can be accessed from other computers, and the program can be easily ported. (3) Matching process of contents of local shared memory Next, a procedure of matching the contents of shared data allocated to the local shared memory of the data management computer (processor 1) and the local shared memory of processor 2 will be described. .

The shared memory between CPUs and the transfer memory have the same memory space on the main memory of a plurality of computers by hardware, and the hardware realizes the matching of the data contents, so that the data matching processing is unnecessary. However, with regard to the local shared memory, since the main memories of different computers are virtually made to appear as the same memory, data matching processing is required. The contents of this matching process will be described below.

The matching processing of shared data is performed by the matching means 19-1 and 19-4 shown in FIG. The matching request is issued when the shared data is newly allocated to the local shared memory (1611 in FIG. 16) or when the content of the shared data is changed in a processing device other than the data management computer (processing device 1). . The processing procedure when the contents of the shared data are changed will be described below.

Since it is the task to change the shared data, it is sufficient for the task itself to make a match after changing the data content. In other words, since the contents of shared data are related 1 byte 1 byte, after changing the contents of shared data,
Normally, matching processing is performed. FIG. 5 shows the configuration of shared data matching processing. The shared data matching processing is performed by the matching requesting means 21-4 called from the task 10-4, the name managing means 12-4 receiving the matching data name 503 from the matching requesting means, and the name managing means. The data matching unit 19-4 is configured to receive 504 the data name, the start address of the shared data entity, and the size of the shared data and perform the data matching process. The data matching means is composed of the data transmission process of FIG. 21 and the data reception process of FIG.

Details of various means in the data matching processing procedure are shown in the processing flows of FIG. 16, FIG. 20, FIG. 21, and FIG. The data matching requesting means transmits 2001 the shared data name to be matched to the name managing means 12-4. The name management means searches the name table with the shared data name received from the matching requesting means, and fetches the start address and size of the shared data entity and the arrangement memory type 1615,
If the fetched layout memory type is the shared memory between CPUs or the transfer memory, the matching process is not necessary. 1616: Only the local shared memory performs matching processing. Then, the name of the shared data, the start address of the shared data entity, and the size are transmitted 1617 to the data matching means. The data matching means receives 2101 the name of the shared data, the start address of the shared data entity, and the size from the name management means,
The content of the shared data for the size is read from the shared data entity start address 2102, and the shared data name 3101 and the shared data content 3102 shown in FIG. 31 are transmitted to the data management computer (processor 1) as a data matching request packet 3100 2103. Upon receiving the data matching request packet, the 2201 data matching means 19-1 fetches the head address of the record of the index table corresponding to the shared data name from the name table 13-1 2202, and the data content 3102 in the shared data area. 2203 to write. Then, a matching completion report is transmitted 2204 as a data matching response packet 3200 to the transmission processing of the data matching means shown in FIG. Next, 2104 data matching means 19 that has received the data matching response packet
In the transmission processing of -4, the coincidence completion report 505 is transmitted 2105 to the name management means 12-4. Finally, the name management means receives 1625 the matching completion report, and the matching processing ends. By performing the matching process in units of the size of the shared data as described above, the overhead of the matching process can be suppressed.

Further, the matching means of the data management computer is
2205, 2206 for making the changed contents of the shared data consistent with the computer using the local shared memory other than the above computer by using the data transmission process.

<Second Embodiment> Furthermore, in addition to the first embodiment, name table copy means 22-1 and 22-4 for copying the name table of the data management computer (processing device 1) into the self-processing device at the time of startup. It is added to the data management computer (processor 1) and other processors. In the first embodiment, when the shared data name accessed by the task does not exist in the name table in the own processing device, the existence check or the allocation of the shared data to the data management computer is performed. Access is time consuming because it accesses the data over the network. In a real-time system that controls plants, work such as new registration and deletion of data areas after online operation is rarely performed in consideration of adverse effects on online control. In this embodiment, in consideration of this, by copying the entire system startup name table from the data management computer and allocating the shared data area, the performance improvement at the time of the first shared data access from the task can be realized.

FIG. 6 shows a configuration in which a name table copy means is added. Name table copy means 23 of processing device 4
-4 is set to start the processing device 4 at the time of startup, and at startup, the name table in the data management computer is transferred to the name table copy means 23-1 of the data management computer (processing device 1). A request to copy to (name table copy request 601) is transmitted. The name table copy means of the data management computer reads 602 the contents of the name table and sends 603 all of the name table contents to the name table copy means of the processing device 4. The name table copying means 23-4 creates 604 a name table in the processing device 4, and in the first embodiment, the name managing means 1
2-4 has made a data allocation request to the data allocation means 15-4, but the name copy means 23-4 makes an allocation request 60 for all shared data registered in the name table.
5 is transmitted to the data allocating means 15-4, the head address of the record in the index table to which the shared data is allocated is received 606, and the head address of the record in the index table is stored in the head address area 13-5 of the name table. To do. Next, in order to match the shared data, a matching request 608 is transmitted to the data matching means 19-5 to match the data.

When the shared data is allocated to the shared memory between CPUs or the transfer memory, the allocation request is not transmitted to the data allocation means 15-4, and the same processing as the name management means of the first embodiment is performed. The empty record in the index table is searched for without the name table copying means, the start address of the shared data is stored in the empty record in the index table, and the empty address in the index table is stored in the start address storage area 13-5 of the name table. Stores the start address of the record.

However, since the name table of the shared data of the shared memory which is not mounted in the processing device is also copied, the shared memory mounting table 2400 is stored in the task 10-4 in order to prevent the task 10-4 from making an erroneous access. It is provided in each processing device, and in the memory allocation process of the shared data in the name table copying means 23-4, the record of the shared data name for which the unmounted instruction is given to the shared memory mounting check table 2400 is not registered.

FIG. 24 shows the structure of the shared memory mounting table. The shared memory mounting table 2400 is composed of a local shared memory mounting flag area 2401, an inter-CPU shared memory mounting flag area 2402, and a transfer memory mounting flag area 2403. When the value in the area is 0, it is not mounted. A value of 1 means implementation. <Third Embodiment> Data matching means 19-4 of the first and second embodiments
Performs the data matching by the matching request from the task 10-4. However, in the data matching method of the first and second embodiments, when a plurality of tasks simultaneously issue matching requests, the data matching is performed. Since the load is concentrated on the means and the matching process of the shared data contents is delayed, this embodiment describes a method of periodically performing the data matching process.

FIG. 7 shows a data matching means 24-having matching time tables 24-1 and 24-4 for setting the cycle time.
A data management system having a write flag 13-6 indicating the end of writing of shared data in the shared data record in the name table 1 and 24-4 is shown. The processing of the matching request means 21-4 of the task 10-4 is the same as that of the first and second embodiments, but the name management means 12-4 which receives the shared data name 701 as the matching request is the same as that of the first embodiment. , Unlike 2,
The write flag 13-6 of the shared data requested to be matched in the name table is turned on 702. At this time, if the layout memory type is the shared memory between CPUs or the transfer memory, the matching process is not necessary and the write flag is not set.

Next, the data matching means will be described.
FIG. 23 shows a processing flow of the data matching means for matching the shared data in the cycle time.

First, the data matching means fetches 230 the cycle time in the matching time table 24 shown in FIG.
One. The cycle time is set in seconds. Next, the cycle time data matching process is set to sleep 2302, and Wakeup is performed.
After that, the records in the name table are sequentially read from the beginning, and the record whose write flag is ON is taken out 23
03. Next, the data content is read from the memory 2304.
Data Matching Means for Data Management Calculation Send 2305 the data matching request packet 3200 shown in FIG. 23, and receive the matching completion report from the matching means of the data management computer 23
06, the write flag of the record for which the matching processing is performed in the name table is turned off 2307. After this process is repeated for the number of records in the name table, the process goes to sleep again. In this way, by performing the data matching process asynchronously with the stack process, it is possible to minimize the influence of the matching process on the task execution performance.

<Embodiment 4> A distributed system is rarely composed of only computers having the same model and the same architecture, and computers having different architectures are often mixed to form a system. When sharing data in such a distributed system, it becomes impossible to share the data due to the data alignment on the memory and the difference in byte order, and the source code of the task can be coded according to the characteristics of the memory. It becomes necessary to recode. In the present embodiment, the portability of tasks is improved by implementing shared data allocation that eliminates such task recoding work.

In order to realize the above arrangement of shared data, the memory characteristic storage table shown in FIG. 11 is provided on each processing device, and the byte order type 1102 and the data align type 1103 are stored. The byte order is a sequence in which data is reversed due to a difference in computer architecture, and FIG. 12 shows a case of a longword (4 bytes) type. The types of byte order are:
There are a type of Big Indian 1201 and a type of Little Indian 1202. Big Indian 12
In 01, data bytes are arranged from the left side, and in Little Indian 1202, bytes are arranged from the right side. This simply copies the data from the Big Indian processor to the Little Indian processor,
From the task, it looks like a completely different value. This difference in byte order has an effect when copying the data contents in the shared data matching process.

Next, the data alignment will be described. Memory alignment includes byte alignment, word alignment, natural alignment 3 and long word alignment.

FIG. 25 shows charactor type data 2504, 2
505, word type data 2506, longword type data 2507,
An example of arranging them in the order of 2505 and arranging them on the memory of various aligns is shown. Here, a byte align 2501, a word align 2502, and a natural align 2503 are used for description.

In the byte-aligned memory, since it is possible to access the memory in units of 1 byte, 2504
Data of 2507, 2506, and 2505 can be arranged continuously, but in a word-aligned memory, access is only possible if the start address of each data is an even address, so a 1-byte empty area precedes the longword type data of 2507. Must be inserted to make an even address. In addition, natural alignment is a data type (word /
Data alignment is determined in relation to longwords, etc.). In other words, the character type is byte aligned, the word type is word aligned, and the long word type is long word aligned. When the data 2504, 2507, 2506, and 2505 are arranged in this order in 2503, the address of the longword type data 2507 becomes a 4-byte boundary, and there is a space of 3 bytes between the data 2504 and 2507.

For such data alignment, it is necessary to change the address calculation processing 1706 when determining the address in the data allocating means of FIG. Address calculation processing 170 relating to shared memory between CPUs and transfer memory
Nos. 7 and 1708 are memories irrelevant to the data alignment because the memory contents are mapped one-to-one in the hardware. Therefore, this embodiment cannot be applied.
Generally, since the shared memory between CPUs and the transfer memory are connected between computers having the same architecture, the data alignment is the same.

The following describes (1) additional processing corresponding to byte order and (2) additional processing corresponding to byte alignment.

(1) Additional processing corresponding to byte order In order to absorb the difference in byte order, in addition to changing the shared data writing processing 2203 in the reception processing of the data matching processing of FIG. 22 as shown in FIG. , FIG. 31
This is realized by adding the byte order type 3103 of the matching data transmission source to the data matching request packet 3100 shown in FIG. In the shared data writing process, the byte order type is read from the memory characteristic storage table 14 in the own processing unit 2701, and compared with the byte order type 3103 in the data matching request packet 3100 2702.
Then, if they match, the received matching request data is written 2704 as it is, and if they do not match, a process of inverting and writing the data type unit is performed.

As described above, the difference in byte order between the processing devices is absorbed.

(2) Additional processing corresponding to byte alignment In order to absorb the difference in byte alignment, the allocation address calculation processing 1706 of the data allocating means 15 shown in FIG. 17 is changed as shown in FIG. Management means 12
It is realized by adding a data type to the data allocation request (shared data size, allocation memory type) from the name table copying means 23.

First, the data alignment type is fetched from the memory characteristic storage table 14 in the self processing device 2601,
It is checked 2602 whether or not the data alignment type is natural alignment, and if it is natural alignment, the data type is assigned 2603 to a variable indicating an alignment value. Next, the start address of the empty memory searched in 1703, 1704, and 1705 of FIG. 17 is divided by the align value 2604, and if there is no remainder, shared data is allocated by the start address of the empty memory. If there is a remainder, assign the assigned address (assigned address / align value + 1)
Recalculation 2605 is performed with an align value, and shared data is allocated to the recalculated allocation address.

As described above, the difference in data alignment between processing instruments is absorbed.

<Fifth Embodiment> The first, second, third, and fourth embodiments are examples of allocating shared data to different shared memories, and an access method and a data management method for each shared data. The present embodiment is a data management system that treats different shared memories as one shared memory by making shared data on different shared memories accessible by one variable.

FIG. 2 shows a system configuration diagram of this embodiment.
The processing device 1 200-1, the processing device 2 200-2, the processing device 3 200-3, and the processing device 4 200-4 are connected by a communication medium 1 206-1, and the processing device 12
00-1 and processing device 3 200-3, processing device 4 200
-4 is a transfer memory 201-1, 201-3, 201-
4 has a shared memory among the four processing units,
0-2 has a local shared memory. The processing device 1200-1 is also connected to the processing devices 5200-5 and 6200-6 by a communication medium 2206-2. Processing device 5 200-5 and processing device 6
Reference numeral 200-6 is a distributed system having an inter-CPU shared memory 203-5. In this system, the data management computer has the phase value 1200-1.

In the above distributed system, when tasks share data, the processing devices 1, 3 having a transfer memory are provided.
Embodiments 1, 2, 3, since they are classified into a processor group 5 of 4 and a processor group 5 of 6 having a shared memory between CPUs and a processor 2 having a shared memory of locale.
In the shared data allocation method of No. 4, it is impossible for tasks to share data on the shared memory with each other under a single name. Therefore, in the present embodiment, in order to handle the areas on the shared memory different from each other as shared data having the same name, the memory characteristic storage tables 204-1 204-2, 204-2
3, 204-4, 204-5, 204-6 have shared data allocation memory types 205-1, 205-2, 205-
3, 205-4, 205-5, 205-6 are provided and the arrangement memory type 1 in the name table 13 shown in FIG.
3-4 is deleted, and the data allocation means flow 170 of FIG.
By changing the arrangement memory type check 1702 in 0 to the processing of checking the shared data arrangement memory type in the memory characteristic storage table, one shared data name is transferred in the processing device 1, the processing device 3, and the processing device 4. The processor 5 and the processor 6 access the memory as a shared memory between CPUs, and the processor 2
Then, it becomes possible to access as a local shared memory. FIG. 28 shows the structure of the memory characteristic storage table. In addition to the configuration of FIG. 11, in this embodiment, a shared data allocation memory type 2803 is added.

Since one shared data name from the above can be assigned to the shared memory unique to the processing unit and shared by the entire system, it is possible to share the data with the heterogeneous shared memory even from a processing unit that does not have the same type of shared memory. .

[0107]

As described above, according to the present invention, in a distributed system in which control is performed via a shared memory, one data management computer that centrally manages shared data of the entire system is provided, and other computers are provided. When a task accesses shared data by distributing only management information,
Since it can be accessed without going through the network, the task processing can be executed without being affected by the load of the network or the load of the data management computer. Therefore, the task execution performance is improved.

Further, according to the present invention, even when a computer having a different main memory architecture of the computer is added to the distributed system, the memory characteristic storage table is provided so that the difference in the main memory architecture can be absorbed and shared data can be accessed. This makes it easy to add different or different types of computers, and improves the system expandability.

Further, according to the present invention, in a distributed system in which computers having heterogeneous shared memories coexist, the heterogeneous shared memories can be accessed by one shared data name so that the task does not have a transfer memory or an inter-CPU shared memory. Since it can be accessed from the computer, it is easy to add / extend a computer of a different model and improve the system expandability.

The present invention uses the indirect addressing method to access the shared data, and since the change of the shared data does not affect the task object, it has the effect of improving the maintainability of the system.

[Brief description of drawings]

FIG. 1 is a diagram showing an outline of a data management system configuration in a distributed system.

FIG. 2 is a diagram showing a system configuration in which heterogeneous shared memories are connected.

FIG. 3 is a diagram showing a shared data address resolution procedure and a data registration procedure in a data management computer.

FIG. 4 is a diagram showing an address resolution procedure of a shared memory between CPUs.

FIG. 5 is a diagram showing a data management system of a local shared memory.

FIG. 6 is a diagram showing a name table matching process procedure at startup.

FIG. 7 is a diagram showing an asynchronous data matching process procedure.

FIG. 8 is a diagram showing a task processing outline matching processing procedure.

FIG. 9 is a diagram showing the configuration of an index table.

FIG. 10 is a diagram showing a name table configuration.

FIG. 11 is a diagram showing a memory characteristic storage table.

FIG. 12 is a diagram showing a byte order.

FIG. 13 is a diagram showing the configuration of a computer attribute storage table.

FIG. 14 is a diagram showing a matching time table configuration.

FIG. 15 is a diagram showing an address resolution processing flow.

FIG. 16 is a diagram showing a flow of name management means.

FIG. 17 is a diagram showing a data allocation means flow.

FIG. 18 is a diagram showing a flow of name check request transmission means.

FIG. 19 is a diagram showing a flow of name check request receiving means.

FIG. 20 is a diagram showing a flow of data matching request means.

FIG. 21: Data matching means flow (data transmission process)
FIG.

FIG. 22 is a flow chart of data matching means (data reception processing)
FIG.

FIG. 23 is a diagram showing a flow of data matching means.

FIG. 24 is a diagram showing a mounting check table configuration of a shared memory.

FIG. 25 is a diagram showing an example of data arrangement according to a difference in memory alignment.

FIG. 26 is a diagram showing an allocation address calculation processing flow.

FIG. 27 is a memory write processing flow depending on the byte order (shared data allocation depending on the memory characteristic)
FIG.

FIG. 28 is a diagram showing a memory characteristic storage table configuration.

FIG. 29 is a diagram showing a name check request packet.

FIG. 30 is a diagram showing a name check response packet.

FIG. 31 is a diagram showing the structure of a data matching request packet.

FIG. 32 is a view showing the structure of a data matching response packet.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Processor 1 (data management computer), 2 ... Processor 2, 3 ... Processor 3, 4 ... Processor 4, 5 ... Communication medium,
6 ... CPU shared memory, 7 ... Transfer memory, 8 ... Local shared memory, 9 ... Index table, 10 ... Task, 11 ... Address solving means, 12 ... Name management means, 1
3 ... Name table, 14 ... Memory characteristic storage table, 1
5 ... Data allocating means, 16 ... Computer attribute storage table,
Reference numeral 17 ... Name check request transmitting means, 18 ... Name check request receiving means, 19 ... Data matching means, 20 ... Data registration means, 21 ... Matching request means, 22 ... Terminal, 23 ...
Name table copy means, 24 ... Matching time table.

 ─────────────────────────────────────────────────── ─── Continuation of front page (72) Masanori Naito 5-2-1 Omika-cho, Hitachi-shi, Ibaraki Hitachi Process Computer Engineering Co., Ltd. (72) Hiroshi Kobayashi 5-chome, Omika-cho, Hitachi-shi, Ibaraki No. 1 Incorporated company Hitachi Ltd. Omika factory (72) Inventor Tomoaki Nakamura 52-1 Omika-cho, Hitachi city, Ibaraki prefecture Incorporated Hitachi Ltd. Omika factory (72) Inventor Mari Shigematsu Omika Hitachi city, Ibaraki prefecture 5-2-1, Machi, Ltd. Inside the Omika Plant, Hitachi, Ltd. (72) Inventor, Tamio Iizuka 5-2-1, Omika-cho, Oita, Hitachi, Ibaraki

Claims (13)

[Claims] 1. A distributed system comprising: a plurality of processing devices connected by a communication medium; and a program on the plurality of processing devices, comprising a shared memory for passing data, (1) the processing in the shared memory. Data allocation means for allocating a data area that can be shared by a plurality of programs on the device (2) Data registration means for registering shared data in the shared memory (3) Management of the data area allocated by the data allocation means by name (4) Name management means for managing the name table (5) Address access means for solving the address of the shared data at the time of execution of a program to access the shared data by name is provided. A data management system characterized in that 2. The data management system according to claim 1, further comprising an index table for making indirect addressing for shared data. 3. The data management system according to claim 1, further comprising one data management computer for centrally managing shared data of the entire system. 4. The data management system according to claim 3, wherein the processing device other than the data management computer includes a name check request transmission means for performing a presence check of the shared data in a name check request packet to the data management computer. The data management computer includes a name check request receiving unit that receives the name check request packet transmitted from the name check request transmitting unit and returns a size and a start address of the shared data entity. And a data management system. 5. The data management system according to claim 3, further comprising means for allocating the name table on a shared memory accessible by the name management means of a plurality of processing devices. system. 6. A distributed system in which a plurality of processing devices are connected by a communication medium and which does not have a memory that can be shared between the processing devices, unlike the distributed system according to claim 1, (1) A program in one processing device Data allocation means for allocating data to a memory that can be shared only between (2) Data registration means for registering shared data in the shared memory (3) For managing the data area allocated by the data allocation means by name Name table (4) Name management means for managing the name table (5) Data matching requesting that the program running on the processing device, after changing the shared data, match the contents of the shared data of the plurality of processing devices Requesting means (6) The contents of shared data on the shared memory on at least one or more processing devices are made to match by the matching request. Data management system, wherein the program is provided an address resolution means for resolving an address of the shared data at runtime to access by the data matching means (7) Name of the shared data. 7. The data management system according to claim 6, wherein the processing devices other than the data management computer copy all the name tables of the data management computer when the system is started up, and copy them in a shared memory of the processing device. And a name table copy means for requesting to allocate the same address as the shared data of the data management computer to the data allocation means in the processing device, wherein the data management computer is transmitted from the name table copy means. A data management system comprising: a means for receiving a name table copy request and transmitting a name table shared data name, a size, and a start address of a shared data entity. 8. The data management system according to claim 6, wherein the processing device has data matching means for reflecting the contents of the shared data, which has been periodically changed, on the shared data of at least one or more computers. A data management system characterized in that 9. The data management system according to claim 6, wherein the processing devices store the data alignment type of the shared memory of the self-processing device when the data alignments of the shared memories provided to each other are different. Memory characteristic storage table (2) The processing device has a data allocating means that adds a process of calculating an address of shared data to be allocated on the shared memory in accordance with the data alignment type in the memory characteristic storage table. Data management system. 10. The data management system according to claim 6, wherein: (1) a memory characteristic storage table that stores a byte order type of a shared memory of its own processing apparatus (2) the processing apparatus stores in the memory characteristic storage table Means for adding a byte order type to a transmission data packet to be matched according to the byte order type and making a matching request (3) The processing device that has received the transmission data performs self-processing with the byte order type in the transmission data. A data management system having data matching means for inverting and matching the contents of write data when the byte order types stored in the memory characteristic table of the device are different. 11. A program on a plurality of processing devices includes a processing device having a shared memory for passing data and a processing device not having the shared memory,
Furthermore, in a distributed system in which each processing device is connected by a communication medium, (1) in addition to the process of allocating a data area that can be shared by a plurality of programs on the processing device to the shared memory, Data allocating means having a process of allocating data to a memory that can be shared only between programs (2) Data registering means for registering shared data in the shared memory (3) Name of the data area allocated by the data allocating means (4) In addition to the name of the data area, the name table has a type of shared memory in which data is arranged. (5) Name management means for managing the name table (6) Data matching request means for requesting matching of contents of shared data of a plurality of processing devices after a shared data is changed by a program operating in the processing device (7) Due to the matching request, only for the data whose allocation memory type does not include the shared memory between processing devices,
Data matching means for matching the contents of shared data (8) Data accessing means for accessing the shared data by name, wherein a program is provided with address solving means for solving the address of the shared data Management system. 12. The data management system according to claim 11, wherein: (1) the data registration means for the processing device to change the allocation memory type. (2) The processing device changes the data name according to the allocation memory type. The data management system further comprises a data allocating means for changing the shared memory to which the data is allocated. 13. A program on a plurality of processing devices includes a processing device having a shared memory for passing data and a processing device not having the shared memory,
Furthermore, in a distributed system in which each processing device is connected by a communication medium, (1) a table that stores the memory type that can be used as a shared memory in its own processing device in all processing devices (2) shared according to the memory type Data allocation means for allocating data area (3) Data registration means for registering shared data in the shared memory (4) Name table for managing the data area allocated by the data allocation means by name (5) The name Name management means for managing the table (6) Data matching request means for requesting the contents of shared data of a plurality of processing devices to be matched by a program running on the processing device after changing the shared data (7) The matching In response to the request for defragmentation, the shared data is made to be the same for the processors having different shared memory types to which the data is allocated. Data management system, wherein the program is provided an address resolution means for resolving an address of the shared data at runtime to access at data matching means (8) Name of the shared data.