JPH07319839A - Distributed shared memory managing method and network computer system - Google Patents

Abstract

PURPOSE:To provide a distributed shared memory managing method and a computer system capable of preventing useless data from being transferred to improve the use efficiency of network resources. CONSTITUTION:Data in a shared data area 210 are classified into two types in accordance with sizes at the time of initializing and transfer control systems corresponding to respective types are applied. Concretely, the whole data are transferred altogether (211) when the size is small. When the size is large, only a page having a difference between shared data stored in a sending original computer and a sent destination computer is transferred (212). Since the amt. of transfer data and transfer frequency are reduced by using a transfer control system corresponding to each size of shared data, network resources can be effectively utilized.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention, in a network computer system composed of a plurality of computer systems connected by a network, is executing data on physically distributed memory on another computer. Involved in distributed shared memory technology for sharing between multiple applications of
In particular, the present invention relates to a distributed shared memory management method and a network computer system for effectively using network resources and, as a result, improving access performance to the distributed shared memory.

[0002]

2. Description of the Related Art In a network computer system, the memory is physically distributed, so that the data stored in the memory of each computer must be consistent in order to appear logically as one data to the application. The technology that guarantees is necessary. This technology is called distributed shared memory management technology.

As a prior art example of the distributed shared memory, there is a method in which a unit of shared data to be accessed is specified in an application and the shared data is transferred for each unit.
aum et al. `` Parallel Programming Using Shared Obj
ects and Broadcasting '', COMPUTER, IEEE, August 19
92, and the system described on pages 10-19.

In the AS Tanenbaum et al. System, a special instruction is provided for each shared data in order to perform a write operation on the shared data. This special write operation
It involves the process of broadcasting the contents to be written and updating the copy of the shared data existing in another computer. However, even if a plurality of computers issue a broadcast, one computer performs the broadcast ordering process to guarantee that the arrival order of broadcast data is the same for all computers.

However, in the system of AS Tanenbaum and others, useless transfer occurs when a single computer performs multiple write operations before reading or writing shared data by another computer. There is a problem.

There is also a method of designating a section for guaranteeing consistency when accessing shared data. For example, the Munin system is a system that realizes this method. The Munin system is described in JB Carter et al. "Implement
and Performance of Munin ", Proceedings of the 13th Operating System Principles Symposium (the 13th A
CM Symposium on Operationg Systems Principles), 1
October 991, pp. 152-164. In Munin,
When a write or read operation is performed on the shared data storage unit, an instruction that requests a consistency guarantee before the operation and an instruction that cancels the consistency guarantee after the operation are described in the application. This instruction enables efficient data transfer between computers for guaranteeing consistency. The instructions for guaranteeing consistency and the timing of data transfer between computers are described below.

The Munin system uses a paging mechanism in virtual memory to process pages in transfer units. Specifically, the computer that has issued the request for guaranteeing the consistency of certain shared data holds the latest page including the shared data in the computer. Here, the computer having the latest shared data is defined as the owner of the shared data. Until the owner releases the consistency guarantee, the owner keeps the page and writes to the shared data,
The change is not immediately reflected on other computers. After the consistency guarantee is released, only the changed page of the shared data is transferred. By doing this,
In the Munin system, even if the write operation for shared data is performed multiple times in the consistency guarantee section, the data transfer to other computers can be suppressed only once after the consistency guarantee section ends.

[0008]

[Problems to be Solved by the Invention] AS Tanenbaum mentioned above
In other systems, data is always transferred between computers in response to a write operation, so there is a problem that useless transfer occurs when writing the same data continuously.

Although the Munin system solves the above-mentioned problems as described above, the following problems, which are not present in the AS Tanenbaum et al. System, occur. That is, the transfer process performed to guarantee the consistency of shared data is in page units. Specifically, the changed shared data is transferred for each page to which the shared data is allocated. For this reason, when sharing data that is smaller than the page size, there is a problem that even the extra data is transferred.

As described above, according to the conventional method, in the data transfer between computers performed to guarantee the consistency of shared data, the number of transfers or the amount of data per one time is wasted. was there.

It is an object of the present invention to provide a distributed shared memory management method and a computer system which prevent wasteful data transfer and improve the utilization efficiency of network resources.

[0012]

According to a distributed shared memory management method of the present invention, a plurality of computers each having an independent address space are connected by a network, and shared among a plurality of computers on a memory in each computer. A distributed shared memory management method in a network computer system having a data area, wherein when a certain computer updates the shared data in the shared data area, the updated shared data is transferred from the computer to another computer. Depending on the size of the shared data, the transfer may be performed by switching between batch transfer for transferring the shared data in a batch or partial transfer for transferring a predetermined storage capacity unit including a changed part to another computer. Then, when a batch transfer is received, the transferred shared data is replaced with the corresponding shared data in its own shared data area, When receiving the partial transfer is obtained by the replaced shared data portion corresponding to only a portion of the shared data the transfer.

For example, the predetermined storage capacity unit is a page, and at the time of the partial transfer, the page index representing the identifier of the page is transferred together with the data of the page to be transferred.

The network computer system according to the present invention is a network computer system in which a plurality of computers each having an independent address space are connected by a network, and a shared data area for sharing between a plurality of computers is provided in each computer memory. In this case, a plurality of control means for ensuring consistency of the shared data memory of each computer are provided, and a means for switching the control means according to the size of the shared data is provided.

More specifically, the network computer system according to the present invention more specifically connects a plurality of computers each having an independent address space through a network, and shares a shared data area on the memory in each computer among the plurality of computers. In a network computer system having, each of the plurality of computers has a shared data storage means for storing shared data used in its own computer, a shared data management table means, and a shared data duplication for storing the latest shared data. Consistency guarantee request means for requesting storage means and designation of consistency guarantee of shared data at an arbitrary point in the program, and consistency guarantee for canceling designation of the consistency guarantee at an arbitrary point in the program When the canceling means and the shared data in the shared data storage means are rewritten, the rewritten page is recorded. And change the page table means that,
The latest shared data is transferred from the shared data copy storage means to the shared data storage means in response to the request from the consistency guarantee requesting means, and the shared data in the shared data storage means is released in response to the cancellation of the consistency guarantee. And the control means for updating the corresponding shared data in the shared data copy storage means of the own computer with the shared data and transferring the shared data to the shared data copy storage means of another computer. The control means includes the shared data to be updated according to the size of the shared data to be updated in the shared data duplicate storage means in the own computer and transferred to the shared data duplicate storage means in another computer. Batch transfer for transferring the entire data or page-by-page transfer for transferring only changed pages by referring to the page table means Wherein the switching whether to perform selective.

In this network computer system, each computer is determined based on the identification number of the shared data used by itself and the data size of the shared data.
The shared data management table means for storing the data type indicating the batch transfer or the page unit transfer is provided, and the control means recognizes the data type of the shared data by referring to the shared data management table means when transferring the shared data. You may do it.

[0017]

In the present invention, the transfer control method for guaranteeing the consistency of the data shared by the applications existing on a plurality of computers is switched according to the data size of the shared data. Specifically, when the application specifies the data size and initializes the shared data, the shared data is classified into one of two types. If the data size does not meet a certain criterion, it is "batch transfer processing data", and if it satisfies the criterion, it is "page unit transfer processing data". When transferring each data, the former,
Transfer data size in batch. The latter divides one shared data page by page and transfers only the necessary pages to guarantee the consistency of the shared data.

Further, when the latest contents of the shared data held by the owner computer are reflected on another computer, it is detected that the application has accessed the shared data in order to specify the difference in the reflected data contents. Which part of the shared data has been accessed is recorded in the changed page table means. Specifically, when there is a write in the page-unit transfer processing data, the page that has been written is recorded using a page fault interrupt, and there is a difference between the duplicates of the shared data held by each computer. Manage.

Further, in order to specify a timing for guaranteeing consistency of shared data on a plurality of computers, two commands of (1) request for guarantee of consistency of shared data and (2) cancellation of guarantee of consistency of shared data. Is provided to allow arbitrary description in the program.

As a specific example, it is assumed that all the computers that have initialized the shared data have a copy of the shared data and the data contents thereof are the same and are in the latest state. First, the latest data held in the computer is passed to the application by the consistency guarantee request for shared data. Next, when the application reads and writes the shared data and then issues the shared data consistency guarantee release, the updated data is transferred to all other computers. Here, in the case of batch transfer processing data, the entire shared data is transferred in a batch, but in the case of page unit transfer processing data, only the written page is transferred.

As described above, all the computers hold the latest data when the processing of (2) is completed.

Since the shared data is classified into two types according to the size and is processed by the transfer control method suitable for each type, the data transfer required to guarantee the consistency of the distributed shared memory is performed. The amount can be suppressed.

For example, when the size is small, it is treated as batch transfer processing data, and only the size of the data is transferred. If the size is large, page-based transfer processing data is used, and only the pages that differ between the latest shared data held by the owner computer and the shared data of other computers are transferred. As described above, since it is possible to prevent useless transfer regardless of the size of the data size, it is possible to reduce the amount of data to be transferred and the number of times of transfer, and to improve the utilization efficiency of the network.

[0024]

1 is a block diagram of a network computer system to which the present invention is applied. In this figure, the network computer system 100 includes a plurality of computers 101.
(101 (1) to 101 (n)) are connected by the network 102. In the computer 101 (1), 103 is a system bus that is a data transmission path between blocks constituting the computer, 104 is a processor that executes various programs, and 105 is a magnetic disk that stores these programs and data. An auxiliary storage device, 106 is a main memory for storing programs and data, and 107 is a network interface for controlling input / output with the network 102.

FIG. 2 is a block diagram of an embodiment of a network computer system to which the distributed shared memory management method according to the present invention is applied.

On the main memory 106, there are an application command sequence and data storage unit 201 and a shared data management command sequence and data storage unit 202. A plurality of application instruction sequences and data storage units 201 can exist in the main memory 106 of one computer. Each application command sequence and data storage unit 201 includes an application command sequence storage unit 230 and a shared data storage unit 210. In order to maintain the consistency of the shared data storage unit 210, the shared data management instruction sequence and data storage unit 202 includes a shared data initialization instruction sequence storage unit 221, a shared data consistency guarantee control instruction sequence storage unit 222, and a shared data consistency guarantee control instruction sequence storage unit 222. There are a data management table 223, a changed page table 224, and a shared data copy storage unit 225.

The application command sequence storage unit 230 stores programs using a shared data initialization command, a consistency guarantee request command, a consistency guarantee cancellation command, and the like. When these instructions are executed, the instructions in the shared data management instruction sequence and the data storage unit 202 are activated, and the processing of guaranteeing the initialization and consistency of the data in the shared data storage unit 210 is performed. In this specification, executing or activating an instruction sequence means that the processor 104 reads and processes the instruction sequence. Further, the word “application” is used as a virtual entity generated by executing the application instruction sequence and the instruction sequence of the data storage unit 201.

When the shared data initialization instruction is issued from the application, the shared data initialization instruction string storage unit 22
The instruction sequence of 1 is executed. The shared data initialization instruction is for executing the shared data initialization processing shown in FIG. 3 described later, and is always issued before the application uses the shared data. This shared data initialization process is
This is processing for setting management information in the shared data management table 223 necessary for managing shared data and securing a memory area for the data size. When the instruction sequence of the transfer control method switching instruction sequence storage section 226 inside the shared data initialization instruction sequence storage section 221 is executed, the transfer control method to be applied is selected based on the size of the shared data as a criterion. Data suitable for the batch transfer process is called "batch transfer process data", and data suitable for the page unit transfer process is called "page unit transfer process data". This attribute is recorded as a data type in the shared data management table 223.
In addition, the batch transfer processing data 211 and the page-unit transfer processing data 212 are shared by the shared data storage unit 210 and the shared data duplication storage unit 2 by a method according to each data type.
Allocate to 25.

When the shared data consistency guarantee request command or the consistency guarantee cancellation command is issued from the application,
The instruction sequence of the shared data consistency guarantee control instruction sequence storage unit 222 is executed. This is the shared data management table 223.
Of the batch transfer processing instruction sequence storage unit 227 or the page-based transfer processing instruction sequence storage unit 228, the transfer control method for ensuring the consistency is determined by referring to the data type of Run. When performing page-by-page transfer processing, in order to transfer only the rewritten page,
The changed page table 224 is referred to. Specific configurations of the shared data management table 223 and the changed page table 224 will be described later with reference to FIGS. 6 and 7, respectively.

FIG. 3 is a flowchart showing an example of shared data initialization processing. When the shared data initialization instruction is issued from the application, the instruction sequence of the shared data initialization instruction sequence storage unit 221 is activated. This flowchart shows the processing executed at this time.

This flowchart consists of a process 301 executed only by the computer in which the application that issued the shared data initialization command exists, and a process 302 by all the computers of the network computer system.

In step 306, the shared data initialization request is transmitted in the form of a packet to all the computers of the network computer system. A packet is a data structure that is a unit when data is transferred between computers. The transmitted request shall reach the computer itself that has issued the shared data initialization request. In this communication, it is assumed that packets are not lost on the network and the order of packets received by each computer is the same.

In step 307, it is checked in the shared data management table 223 whether or not the shared data number (described later in FIG. 6) has been registered for the key number specified by the initialization request. If the shared data number has not been registered, transfer control method switching processing is performed (step 30).
8). At this time, when the data size does not meet a certain criterion, it is treated as batch transfer processing data, and when the criterion is satisfied, it is treated as page unit transfer processing data.

At step 309, a registration field (entry) for a new shared data number is secured in the shared data management table 223, and the data type decided at step 308 is recorded.

In step 310, the shared data holding computer list is a table which records the identifiers of the computers which have already issued the shared data initialization request. In this list,
It is checked whether the identifier of the computer that issued the initialization request is registered. If it is not registered yet, the process moves to step 311 to register it in the list. Also, the shared data copy storage unit 225 is allocated.

In this embodiment, it is premised that a plurality of applications can exist in one computer. Registered in the shared data holding computer list is when there is an application that has already issued the initialization instruction before the initialization instruction currently being processed in the computer that issued the initialization instruction. If the shared data management table 223 has a registration field for the specified key number but is not already registered in the shared data holding computer list, an initialization command is issued by an application other than the computer that issued the initialization command. It corresponds when it is done.

Shared data holding A in step 312
The P list is a table in which the identifiers of the applications in the own computer that have already issued the initialization request for shared data are recorded. It is checked whether or not the identifier of the application that issued the initialization request is registered in this list. If it has not been registered yet, the process moves to step 313 and is registered in the list.

In step 314, a memory area for shared data is secured in the same manner as in the shared data copy storage unit 225.

When already registered in the shared data holding AP list (when the processing reaches step 315)
Means that the application has already completed the initialization processing of the shared data (that is, the application has issued the initialization command for the same shared data more than once), so an error is issued to the application. To notify.

When the use of the shared data is completed, post-processing such as releasing the allocated area of the shared data copy storage unit or deleting information from the shared data management table is required. At this time, in order to recognize whether or not another AP is using the shared data, the shared data management table 223
The shared data holding computer list 603 and the shared data holding AP list 604 in FIG.

FIG. 4 is a flow chart showing an example of the transfer control method switching process at the time of initialization of the shared data shown in step 308 of FIG.

In step 401, the total area used page number is the number of pages included in the shared data area, assuming that the shared data is allocated on the memory in accordance with the page boundaries. In the present embodiment, if the number of pages used in all areas is 3 or more, it is treated as page unit transfer processing data (step 402). However, the division here is rounded down to the nearest integer.

If the number of pages used in all areas is less than 3, the process proceeds to step 404, and if the number of pages used is less than 1, it is treated as batch transfer processing data. If the number of pages used in all areas is 1 or 2, the process proceeds to step 406. The “final page usage amount” in step 406 is the remainder value obtained by dividing the shared data size by the page size. If this value is equal to or larger than the page size / (pages used in all areas + 1), it is treated as page unit transfer processing data, and if not, it is treated as batch transfer processing data (step 407).

The condition for switching shown in FIG. 4 is an example, and the present invention is not limited to this.

FIG. 5 shows an example of a method of allocating shared data to the memory. The downward direction indicates the upper address, the page unit transfer processing data 501 is allocated from the lower to the upper part of the shared memory area 500, and the batch transfer processing data 502 is allocated from the upper to the lower part. In the page-unit transfer processing data 501, if different data exist on the same page, independent transfer processing cannot be performed, so the page-unit transfer processing data 501 is allocated according to the page boundary.

The unit of data to which the shared data number is given is a unit such as an array or a structure which can be logically grouped and handled as one.

FIG. 6 shows an example of the shared data management table 223. It contains the information needed to manage shared data.

The common key number 601 in the network computer system is a number set on the application side in order to identify the same shared data in the network computer system when the shared data is initialized. The shared data number 602 is an identification number given to the shared data initialization request sent to all the computers of the network computer system in the order in which they arrived at each computer. The shared data number 602 is a number assigned by the management unit (shared data management command sequence and data storage unit 202) for easy management. For example, a serial number is used when searching the shared data management table 223 for shared data information. It has advantages such as fast processing. However, this shared number is not mandatory. The shared data holding computer list 603 describes the identifiers of the computers that have requested the initialization of shared data. Shared data holding AP list 6
Reference numeral 04 indicates the identifier of the application that has issued the shared data initialization command within the own computer. For example, the identifier of this application is UNIX
A process number (trademark of AT & T Bell Laboratories) can be used. Address 60 of shared data copy storage unit
5. The address 606 of the shared data storage unit indicates the start address of each shared data storage unit. The shared data size 607 is the size of the area designated by the application and allocated in the shared data copy storage unit 225 and the shared data storage unit 210 (the example of the figure shows the number of bytes). The data type 608 indicates whether it is page-unit transfer processing data. In this example, "1" indicates page-unit transfer processing data, and "0" indicates batch transfer data.
The registered contents of the items of each entry in this table are all recorded during the shared data initialization process.

FIG. 7 shows an example of the changed page table 224. The changed page table 224 records the pages that have been rewritten between the time when the application issues the consistency guarantee request command for page-unit transfer processing data and the time when the consistency guarantee cancellation command is issued. Specifically, when a page fault interrupt occurs, the shared data number 701 to which the page belongs and the page index 70
2 is checked, and the process of setting the change flag 703 of the corresponding page is performed.

FIG. 8 shows an example of a shared data access function provided to an application for applying the shared data management method according to the present invention.

In order for the application to access the shared data, an initialization process, a consistency guarantee request process, a consistency guarantee cancellation process, and an end process are performed. First, the command sequence of the shared data initialization command sequence storage unit 221 is activated by the command nsm_open for executing the initialization process of 851. The initialization command nsm_open sets the key number, data size, and data type, and the start address and return value 8
A shared data number of 62 is returned. The key number is a number set on the application side in order to identify the same shared data in the computer system, and corresponds to the key number 601 described in FIG. Unless otherwise specified, the data type is classified into either batch transfer process data or page-unit transfer process data by the transfer control method switching process. It is also possible to specify on the application side. Further, the head address of the shared data allocated to the shared data storage unit 210 is returned based on the data size and the data type. These pieces of information are recorded in the shared data management table 223. Further, when the application accesses the initialized shared data, the shared data number of the return value 862 is designated.

When performing a write operation or a read operation to the memory area in the shared data storage unit 210 specified by the shared data number, a series of write and read operations are performed by a consistency guarantee request command 852, consistency guarantee request commands nsm_acquire and 853. By enclosing it with the release command nsm_release, the consistency of shared data is guaranteed.

When the application no longer needs the shared data, the termination processing instruction nsm_cl at 854.
Issue ose. This issuance releases the shared data management command sequence secured in the initialization process and the management information area in the data storage unit 202.

FIG. 9 shows an example of the structure of a network packet 800 transmitted / received by the network interface 107 via the network 102. The network packet 800 has a packet header 801 and data 802.
Composed of. The packet header 801 includes a destination computer 811, a source computer 812, and a shared data number 81.
3 and shared data management command 814. The shared data management command 814 includes a shared data initialization command, a consistency guarantee request command, a consistency guarantee release command, etc. corresponding to the above-described various commands. In the consistency guarantee cancel command, data having the latest contents is entered in the data 802. In this figure, page unit transfer processing data 80
The case of 3 is shown. Page unit transfer processing data 80
3 contains the number of pages 821 stored in the packet, the page index 822, and the page entity 823. Unlike the format of the data 803, the batch transfer processing data is put in the data 802 as a batch.

The shared data initialization command and the consistency guarantee cancellation command are broadcast to all computers. This broadcast is premised on that all computers have the same arrival order of the broadcast data transmitted from each computer. To broadcast, for example, the destination computer 811 of the packet header
Is set to "-1" and the network interface 10
In No. 7, if the destination computer transmits a packet of "-1", it may be decided in advance to broadcast it.

10 and 11 are explanatory views of an example of the data transfer control method. FIG. 10 shows batch transfer processing data, FIG.
Reference numeral 1 denotes a transfer control method of page-unit transfer processing data. Both figures show the flow of the processing of the application 905 of the computer 901 that changes the shared data and the shared data consistency guarantee control processing 906, taking the time axis in the vertical direction. Here, shared data consistency guarantee control processing 9
06 is a shared data consistency guarantee control instruction sequence storage unit 22.
The processing performed by executing the second instruction. Further, the shared data storage unit 210 in the application, the shared data management instruction sequence and the data storage unit 20.
2 schematically shows the data content of the shared data duplication storage unit 225 in FIG. In the horizontal direction, shared data consistency guarantee control processing 907 of other computers is arranged, and arrows indicate communication status. The broken line shows the transmission of commands, and the solid line shows the transfer of data. The shared data 1020 to 1025 shown in the shared data storage unit 210 (1) and the shared data duplicate storage units 225 (1) and (2) have the same pattern, which means that they have the same content.

With reference to FIG. 10, a transfer control method of batch transfer processing data will be described. This figure shows the processing when the application writes to one shared data of the batch transfer processing type.

First, the shared data copy storage unit 225 (1)
It is assumed that the latest data 1022 is held inside.
The latest data is not always held in the shared data storage unit 210 (1) of the application. Therefore, when the application 905 issues the consistency guarantee request command 930, the command sequence of the batch transfer processing command sequence storage unit 227 of the shared data consistency guarantee control command sequence storage unit 222 is activated,
Latest data 922 of shared data duplication storage unit 225 (1)
Is copied to the shared data storage unit 210 (1) (94
0). In this state, the application 905 writes and reads the data 920 in the shared data storage unit 210 (1).

When the application 905 issues the consistency guarantee cancellation instruction 931, the batch transfer processing instruction string storage unit 2
The command sequence of 27 is activated, and the shared data storage unit 210
The shared data copy storage unit 225 stores the data 921 in (1).
Copy to (1) (941). Also, the latest data of the data size is collectively transferred to all computers except the self (942).

In this way, the result of the processing performed on the shared data by the application 905 between the consistency guarantee request instruction 930 and the consistency guarantee release instruction 931 is transmitted to all the computers.

With reference to FIG. 11, a transfer control method of page-unit transfer processing data will be described. This figure shows the processing when the application 1005 writes to one shared data of the page unit transfer processing type.
Only the part of processing different from FIG. 9 will be described. After the consistency guarantee request command 1030, the application 1005 writes and reads the shared data 1021. At this time, each time a new page is written, the shared data management instruction sequence and the instruction sequence in the data storage unit 202 are activated, and the changed page table 224 records which page has been written. When the consistency guarantee cancellation instruction 1031 is issued, the instruction sequence of the page unit transfer processing instruction sequence storage unit 228 is activated, and the shared data storage unit 210 is activated.
Only the rewritten page 1040 in (1) is copied to the data copy storage unit 225 (1) (1041). Also,
Only the rewritten page of the latest data 1023 is transferred to all computers (excluding self) (1042). As described above, in the case of page-unit transfer processing data, useless transfer is omitted by transferring only the rewritten page.

Although not directly related to the present invention, in the present embodiment, other A
The applications are synchronized so that P does not access the same data so that the access does not overlap.

FIG. 12 is a flowchart showing an example of the consistency guarantee request process.

When the consistency guarantee request command 852 of the application command sequence storage unit 230 is issued, the command sequence of the shared data consistency guarantee control command sequence storage unit 222 is activated,
The processing shown in this flowchart is performed. Step 1
At 102, the shared data management table 223 based on the shared data number designated by the consistency guarantee request command 852.
To obtain the address 606 of the shared data copy storage unit, the address 605 of the shared data storage unit, and the size 607 of the shared data. In Step 1103, Step 1102
The latest data is copied from the shared data duplication storage unit 225 to the shared data storage unit 210 based on the information obtained in.

FIG. 13 is a flowchart showing an example of the consistency guarantee cancellation processing.

When the consistency guarantee cancellation command 853 of the application command string storage unit 230 is issued, the command string of the shared data consistency guarantee control command string storage unit 222 is activated,
The processing shown in this flowchart is performed. Step 1
In 202, the data type 608 of the shared data number designated by the consistency guarantee cancellation instruction 853 is obtained from the shared data management table 223. In step 1203, the processing is switched depending on whether the obtained data type is page unit transfer processing data or batch transfer processing data.

In step 1204, the changed page table 224 is referred to, and for the shared data of the shared data number, the page rewritten in the section from the consistency guarantee request command to the consistency guarantee cancel command is selected. The page corresponding to the page index 702 in which the change flag 703 is set in the designated shared data number 701 is the target page. In Step 1205, Step 12
The page selected in 04 is copied from the shared data storage unit 210 to the corresponding page in the shared data duplication storage unit 225. In step 1206, the page selected in step 1204 is transferred to all the computers except the self by the page unit transfer processing data 803 of the network packet 800. Each computer that has received the network packet 800 stores the corresponding page in the shared data copy storage unit 225.

If it is determined in step 1202 that the data is not page-unit transfer processing data, step 1207
Then, the shared data having the designated shared data number are collectively transferred from the shared data storage unit 210 to the shared data duplication storage unit 22.
Copy to 5. In step 1208, the shared data is collectively put in the data 802 portion of the network packet 800 and transferred to all computers (excluding the self). Each computer that has received the network packet 800 writes it in the corresponding area of the shared data copy storage unit 225.

All the computers hold the latest data by such a consistency guarantee canceling process.

In the above description, the shared data copy storage unit 2
Although it has been described that the shared data is held in 25, the shared data initialization instruction is first issued when the shared data is first stored by one computer. Next, a consistency guarantee request command is issued. At this time, the shared data is copied from its own shared data copy storage unit 225 to its own shared data storage unit 210, but this data is meaningless. Next, the shared data is written in the shared data storage unit 210. Finally, by issuing the consistency guarantee cancellation command, this shared data is stored in the shared data copy storage unit of all computers.

[0071]

According to the present invention, in a network computer system composed of a plurality of computer systems connected by a network, a distributed shared memory that makes an application appear as if sharing one data. When realizing the above, the transfer control method for guaranteeing the consistency of shared data is switched according to the size of shared data to reduce the amount of transfer data and the number of transfers on the network, and effectively use network resources. can do.

[Brief description of drawings]

FIG. 1 is a block diagram of a network computer system according to an embodiment of the present invention.

FIG. 2 is a block diagram of an embodiment of a network computer system to which a distributed shared memory management system according to the present invention is applied.

FIG. 3 is a flowchart showing an example of shared data initialization processing according to the present invention.

FIG. 4 is a flowchart showing an example of shared data transfer control method switching processing according to the present invention.

FIG. 5 is an explanatory diagram showing an example of a method of allocating shared data to a memory according to the present invention.

FIG. 6 is an explanatory diagram showing an example of a shared data management table according to the present invention.

FIG. 7 is an explanatory diagram showing an example of a changed page table according to the present invention.

FIG. 8 is an explanatory diagram showing an example of a shared data access function according to the present invention.

FIG. 9 is an explanatory diagram showing an example of a configuration of a packet flowing through a network according to the present invention.

FIG. 10 is an explanatory diagram of a data transfer control method of batch transfer processing data according to the present invention.

FIG. 11 is an explanatory diagram of a data transfer control method of page unit transfer processing data according to the present invention.

FIG. 12 is a flowchart showing an example of shared data consistency guarantee request processing according to the present invention.

FIG. 13 is a flowchart showing an example of shared data consistency guarantee cancellation processing according to the present invention.

[Explanation of symbols]

100 ... Network computer system, 101 ... Computer, 102
... network, 103 ... system bus, 10c ... main memory, 107 ... network interface, 201 ... application instruction sequence and data storage unit, 202 ... shared data management instruction sequence and data storage unit, 221 ... shared data initialization instruction sequence storage unit , 222 ... Shared data consistency guarantee control instruction sequence storage unit, 223 ... Shared data management table, 224 ... Change page table, 210 ... Shared data storage unit, 211 ... Collective transfer processing data, 212 ... Page unit transfer processing data, 301 ... Processing of the computer that issued the shared data initialization processing request, 302 ... Processing of all computers in the network computer system, 308 ... Transfer control method switching processing, 500 ... Shared memory area, 501 ... Page unit transfer processing data, 502 ... Batch Transfer processing data, 6
02 ... shared data number, 607 ... data type, 800 ... network packet, 801 ... packet header, 802 ... data,
901 ... A computer that changes shared data, 905 ... An application, 906 ... Shared data consistency guarantee control processing, 210
(1) ... Shared data storage in the computer that makes changes to shared data, 225 (1) ... Shared data duplicate storage in the computer that makes changes to shared data, 225 (2) ... Shared data in other computers Duplicate storage unit, 1001 ... Computer that makes changes to shared data, 1005 ... Application, 1006 ... Shared data consistency guarantee control processing.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Satoshi Yoshizawa 1-280 Higashi Koikeku, Kokubunji City, Tokyo Central Research Laboratory, Hitachi, Ltd. (72) Takehisa Hayashi 1-280 Higashi Koikeku, Kokubunji, Tokyo Hitachi Ltd. Central Research Institute (72) Inventor Kominori Yamada 5030 Totsuka-cho, Totsuka-ku, Yokohama City, Kanagawa Prefecture Hitachi Ltd. Software Development Division

Claims (5)

[Claims] 1. A distributed shared memory management method in a network computer system in which a plurality of computers each having an independent address space are connected by a network, and a shared data area for sharing between the plurality of computers is provided in each computer memory. Then, when the shared data in the shared data area is updated on a certain computer, when the updated shared data is transferred from the computer to another computer, the shared data is updated according to the size of the shared data. Transfers are performed by switching between batch transfer for batch transfer and partial transfer for transfer to a predetermined storage capacity unit including the changed part. When other computers receive the batch transfer, the transfer is performed. Replace the shared data with the corresponding shared data in its own shared data area, and if partial transfer is received, the part of the transferred shared data Distributed shared memory management method, characterized by replacing the shared data portion corresponding only. 2. The predetermined storage capacity unit is a page, and at the time of the partial transfer, a page index representing an identifier of the page is transferred together with data of the page to be transferred. Distributed shared memory management method. 3. A network computer system in which a plurality of computers each having an independent address space are connected by a network, and a shared data area for sharing between the plurality of computers is provided in each computer memory. A network computer system comprising a plurality of control means for ensuring consistency of shared data memory, and means for switching the control means according to the size of shared data. 4. A network computer system in which a plurality of computers each having an independent address space are connected by a network, and a shared data area for sharing between the plurality of computers is provided in each computer memory. Each computer has a shared data storage means for storing shared data used by its own computer, a shared data management table means, a shared data duplicate storage means for storing the latest shared data, and designation of shared data consistency guarantee. Guarantee requesting means for requesting a certain part in the program, a consistency guarantee releasing means for canceling the specification of the consistency guarantee at an arbitrary part in the program, and sharing in the shared data storing means When the data is rewritten, the changed page table means for recording the rewritten page and the consistency guarantee request means are used. When the latest shared data is transferred from the shared data copy storage means to the shared data storage means in response to the request, and the shared data in the shared data storage means is rewritten in response to the consistency guarantee cancellation. And a control unit for updating the corresponding shared data in the shared data copy storage unit of the own computer with the shared data and transferring the shared data to the shared data copy storage unit of another computer. , Regarding the shared data that is the target of updating the shared data duplication and storage means in its own computer and transferred to the shared data duplication and storage means in another computer, the entire shared data is transferred in batch according to the size of the data Selectively switching whether to perform transfer or page unit transfer in which only changed pages are transferred by referring to the page table means A network computer system characterized by 5. The network computer system according to claim 4, wherein each computer is the batch transfer or page-by-page transfer determined based on the identification number of the shared data used by itself and the data size of the shared data. A shared data management table means for storing a data type indicating
The network computer system, wherein the control means recognizes the data type of the shared data by referring to the shared data management table means when transferring the shared data.