JP3019012B2 - How to share files - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a file sharing method in a loosely-coupled multiprocessor system in which a plurality of computers each having a CPU, a storage device (memory) and a file device are connected to a network.

[0002]

2. Description of the Related Art Generally, in a file sharing system such as a client / server system, a file server has a specific computer, particularly a computer having better functions and performance than other computers, or a server function. Those that are provided are assigned, and the central management of the file is performed by this.
On the other hand, in a system using a so-called loosely-coupled multi-processor configuration in which a plurality of computers of the same performance are connected to a network to improve the total performance, a file stored in each computer is referred to by another computer, and the Sharing. The present invention relates to a file sharing method in the latter system.

Various methods are conceivable as a method of connecting a plurality of computers, for example, computers A, B, C, and D, to a network and sharing a file F held by a certain computer A with all the computers. For example, another computer that intends to use the file F, for example, the computer B, obtains an entire copy of the file F from the computer A, stores it in its own computer, and refers to and updates the stored copy. There is a way. This has the advantage that the computer B can continue to process the file F even if the communication with the computer A is interrupted thereafter. The other computers C and D other than the computer B can also use the file F by obtaining a copy of the file F from the computer A. Further, when sharing the file F among a plurality of computers, maintaining consistency between the master of the file F and its copy becomes a problem. As a method for this, the computer B updates the copy of the file F. Update the contents of computer A promptly in case
There is a method to notify the master and update the master. At this time, if the other computers C and D hold copies of the file F, the computer A can maintain the consistency by rewriting those copies.

[0004]

As a method of using a file held by a certain computer by another computer, the above-mentioned method can be considered. However, similarly to a conventional client / server system, a file serving as a master can be used. Since the computers that own the files are statically fixed, the load on the computer that holds the frequently used shared files increases abnormally, and the load between the computers becomes uneven, which may reduce the throughput of the entire system. There is. That is, in the above-described example, if the file F is a file frequently used by each of the computers A to D, the computer A transfers a copy of the file F to a request source in response to a request from another computer B to D. And the process of maintaining consistency must be undertaken.

An object of the present invention is to distribute the processing load on a shared file by dynamically changing the computer holding the master of the shared file.

If the computer holding the master of the shared file F is statically fixed as described above, if the computer fails and goes down, a clear copy of the shared file F is made at that time. Can continue processing the shared file F, but other computers cannot use the file until the master computer is restored.

[0007] It is another object of the present invention to allow all remaining computers to continue using the shared file as much as possible, even if a failure occurs in the computer holding the master of the shared file. It is in.

[0008]

SUMMARY OF THE INVENTION The present invention relates to a file sharing method in a loosely coupled multiprocessor system in which a plurality of computers each having a CPU, a memory, and a file device are connected to a network. The computer that intends to use the shared file receives the transfer of the copy from the computer that owns the master of the shared file, stores it in its own file device, and holds the master or copy of the shared file in its own file device. The computer that performs the operation stores, in its own memory, the shared file name, an identifier indicating whether the shared file is the master, a clear copy having the same contents as the master, or a dirty copy that is an old copy, and an identifier of the computer that owns the master. In addition, for all computers holding the master or the copy, a directory for a reference chain including a bidirectional pointer for creating a reference chain for connecting the computer holding the master as the highest level is generated, and a dirty copy of the shared file is generated. Before using a shared file, a computer with a clear copy of the shared file or a computer that has the master,
Transfer the copy of the shared file from the computer closest to the own computer and make it a clear copy, and if the computer with the master or clear copy of the shared file updates the shared file, copy the updated shared file. Each computer cooperates to update the reference chain directory relating to the shared file so that the new master and the others are all dirty copies.

According to the present invention, in addition to the above configuration, the computer holding a copy of the shared file monitors the presence or absence of a failure in the nearest higher-level computer on the reference chain relating to the shared file. If the failure of the computer with the master of the shared file is detected by, the computer with the failure is removed from the reference chain, and another computer with a clear copy of the shared file becomes the new master in order to become the new master. It is characterized in that the computer cooperates to update the reference chain directory concerning the shared file.

Further, the present invention provides, in addition to each of the above-described configurations,
When a failure of a computer having a copy of the shared file is detected by the monitoring, the computers cooperate to update the reference chain directory for the shared file so as to remove the failed computer from the reference chain. It is characterized by.

[0011]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing an example of a loosely coupled multiprocessor system to which the present invention is applied. In the loosely-coupled multiprocessor system of this example, four computers A, B, C, and D are connected to each other via a bus BUS so that they can communicate with each other. The number of computers does not need to be four, and the network topology does not necessarily need to be a bus type.

Each of the computers A to D is a computer A
As exemplified in the figure, a CPU 1, a file device 2 connected to the CPU 1, a memory 3 connected to the CPU 1, and a communication device 4 connected between the CPU 1 and the bus BUS are provided.

The file device 2 stores a shared file 13 and its directory 12. As shown in FIG. 2, the directory 12 manages a name (shared file name) 121 of the shared file 13 and an identifier 122 indicating whether the shared file 13 is a master file or a copy. Here, Master is set in the identifier 122 when the file is a master, and Copy is set in the identifier 122 when the file is a computer. When a plurality of shared files 13 are stored in the file device 2, the directory 12 exists for each shared file.

The memory 3 includes a reference chain directory 10.
, A shared file management program 11, and an application program (AP) 14.

The reference chain directory 10 is created for each file shared by a plurality of computers A to D. As shown in FIG. 3, the reference chain directory 10 has a shared file name 101 and a file type identifier 102 indicating whether the shared file is a master, a clear copy having the same contents as the master, or a dirty copy which is an old copy. And a master pointer 104 pointing to the computer holding the master, and two pointers 103 and 105 for creating a reference chain in which all the computers holding the master or the copy have the computer holding the master as the highest level. And manage. Here, one pointer 103 is a local master pointer indicating an identifier of a computer at a level immediately above the own computer on the reference chain, and the other pointer 104
Is a reference pointer indicating the opposite, that is, the identifier of the computer at the immediately lower level of the own computer on the reference chain.

Here, the file type identifier 102 includes
When the own computer holds the master file, Master is used. When the computer has a copy (clear copy) equivalent to the master file, Clear Cop is used.
When y has a different (old) copy (dirty copy) due to the update of the master file, Dirty Copy is set. The local master pointer 103 points to a computer located one level higher than itself in the reference chain as described above, but the computer indicated by this does not necessarily have the master file. File type identifier 1
If 02 is a Master, 103 is Don't Ca
re. The master pointer 104 points to the computer having the master file as described above. When a reference chain is created, the master pointers 104 of all reference chain directories on the chain except the master point to the same computer. If the file type identifier 102 is Master, 104 is Don't
Care. As described above, the reference pointer 105 indicates a computer whose reference level is one level lower than itself, and is NULL when there is no reference level.

Next, the shared file management program 11 stored in the memory 3 of FIG. 1 manages all the files shared by the plurality of computers A to D in cooperation with the shared file management program 11 of another computer. It is a program to be performed. The AP 14 is an application program that performs a predetermined process by referring to and updating a shared file.

When referring to and updating a shared file, the AP 14 has a reference chain directory 10 of the shared file to be referenced and updated in the memory 3 and has a file type identifier 102 of D as shown in the flowchart of FIG.
If it is not an irty copy (YES in S1, S
(NO in 2), the process proceeds to step S5, and the shared file is referred to and updated. On the other hand, if the reference chain directory 10 of the shared file to be referred to or updated does not exist in the memory 3 (NO in S1), a copy acquisition request is issued to the shared file management program 11, and control is performed from the shared file management program 11. Is returned to step S5, and the shared file is referenced and updated. Even if the reference chain directory 10 of the shared file exists in the memory 3, if the file type identifier 102 is DirtyCopy (YES in S <b> 2), the clear copy to the shared file management program 11 is performed. Then, after the control is returned from the shared file management program 11, the process proceeds to step S5 to refer to and update the shared file. In step S5, when the shared file is updated, the updated shared file name is notified to the shared file management program 11 at the end of a series of updates.

The shared file management program 11 includes an AP
When a copy acquisition request is issued from 14, if the requested shared file exists in the file device 2 of the own computer, a reference chain directory 10 of the shared file is generated in the memory 3 and control is returned to the AP 14. If the file does not exist in the file device 2 of the own computer, a copy of the shared file is obtained from the computer having the master of the shared file and stored in the file device 2, and the reference chain directory 10 of the shared file is stored in the file device 2. It is generated on the memory 3 and the control is returned to the AP 14.

Also, the shared file management program 11
When a clear copy request is issued from the AP 14, a clear copy of the shared file or a computer having a master and receiving the transfer of the copy of the shared file from the computer closest to the own computer on the reference chain relating to the shared file. Make a clear copy. Note that the clear copy of the shared file is performed separately from the clear copy request from the AP 14 by using the idle time of the process. That is, the shared file management program 1
1 is to check the file type identifier 102 of the reference chain directory 10 on the memory 3 during the idle time of the process, and if it is Dirty Copy, a computer having a clear copy or master of the shared file, A process of receiving a copy of the shared file from the computer closest to the own computer on the reference chain relating to the shared file and performing a clear copy process.

Further, the shared file management program 11
If the shared file name is notified after the update of the AP 14 is completed, and if the shared file is not the master, each of the updated shared files is set as a new master and all the other computers are set as dirty copies. In cooperation with the shared file management program 11 of the computer, the reference chain directory relating to the shared file is updated.

In addition, a shared file management program 11
Monitors the presence or absence of a failure in the computer at the immediately higher level on the reference chain relating to the shared file, and if a failure is detected, performs recovery processing in cooperation with the shared file management program 11 of another computer. Perform.

The operation of this embodiment will be described below. In addition,
For convenience of explanation, attention is focused on a certain shared file File0, and this shared file File0 is
The operation in the case where C refers to and updates is described, but the operation is the same for other shared files.

FIGS. 5 to 9 are flowcharts showing an example of processing of the shared file management program 11. 5 shows a flow of a routine related to a process executed when the AP 14 notifies the file name after the update of the shared file File0 is completed, and FIG. 6 relates to a process of clear copy of dirty copy. FIG. 7 is a flowchart of a routine related to a process for acquiring a copy of the shared file File0 that does not exist in the file device 2 of the own computer from the computer having the master. FIGS. 8 and 9 are flowcharts of other computers. The flow of a routine related to the failure monitoring and the processing at the time of failure detection is shown, respectively. In FIGS. 5 to 9, Fi
le0-X (X = 102-105) is the shared file F
The fields correspond to the fields 102 to 105 (FIG. 3) of the reference chain directory 10 of ile0.

FIGS. 10 to 17 show the contents of the reference chain directory of each computer and the state of change with respect to the shared file File0. In the drawings of each figure, A, B, and C drawn to the left of the reference chain directory correspond to the computers A, B, and C, respectively, and a solid line connecting them represents a chain of references. The higher the level, the higher the level. Also, C.I. in the directory for reference chains. Copy, D.S. Copy is Cle
ar Copy, Dirty Copy.
Further, DC stands for Don't Care.

(1) Operation Example 1 In this operation example 1, an operation in which the computer B receives the transfer of the shared file File0 from the computer A in order to use the shared file File0 in the initial state will be described. Here, as an initial state, it is assumed that the shared file File0 is stored in the file device 2 of the computer A, and the other computers B, C, and D have no copy. In addition, the memory 3 of the computer A stores the shared file File0 in FIG.
It is assumed that a reference chain directory 10A as shown in FIG.

In the computer B, there is no copy of the shared file File0 in its own file device 2. Therefore, there is no reference chain directory 10 for the shared file File0 on the memory 3. Therefore, when the AP 14 attempts to use the shared file File0, the shared file File0 is transmitted to the shared file management program 11.
Is issued.

Since the shared file File0 does not exist in its own file device 2, the shared file management program 11 of the computer B requests the computer A having the master of the shared file File0 to transfer the file via the bus BUS. . Here, the shared file File
As a method of knowing the computer A having the master of 0,
For example, a method in which the computer B inquires each computer of the master destination of the shared file File0 through the bus BUS to know it, a method in which a list of the directories 12 stored in the file device 2 of another computer is obtained and examined, and the like can be used. .

The shared file management program 11 of the computer A transmits the shared file File from the computer B.
When the transfer of 0 is requested, the shared file File0 is read from its own file device 2 and transferred to the computer B via the bus BUS.

The shared file management program 11 of the computer B stores the received shared file File0 in its own file device 2, and stores the directory 12 in FIG.
Create in the format as shown in. At this time, the file name 121 contains the name of the shared file File0,
Copy is set in the identifier 122 indicating whether the file is a master file.

Next, the shared file management program 11 of the computer B, as shown in FIG.
And the file type identifier 102 as the name of the shared file File0. A reference chain directory 10B in which Copy is set is created in the memory 3.

Next, the shared file management program 11 of the computer B starts executing the routine RG shown in FIG. First, the identifier of the computer A which is the absolute master is set in the master pointer 104 of the created reference chain directory 10B (step RG1). Next, the inter-process communication PI is transmitted to the computer A which is the absolute master.
Notify by NT5 that itself (computer B) has obtained a copy of shared file File0 (Refe
notification (reception notification) (step RG2).

The shared file management program 11 of the computer A that has received the inter-process communication P-INT5 starts executing the routine RH shown in FIG. in this case,
The reference chain directory 10A of the shared file File0 of the computer A has the contents as shown in FIG. 10A, and the master pointer 102 is Master and the reference pointer 105 is NULL. To step RH5,
The reference pointer 105 of the reference chain directory 10A is changed to the identifier of the computer B notified by the inter-process communication P-INT5. Then, it notifies the computer B that the local master is itself (computer A) by the inter-process communication P-INT6 (step RH6).

The shared file management program 11 of the computer B that has received the inter-process communication P-INT6 executes the step RG3 of the routine RG of FIG. 7 and executes the local master pointer in the reference chain directory 10B of FIG. 10B. 103 shows inter-process communication P-INT6
The identifier of the computer A which is the local master notified by the
Is set to NULL.

By performing the above operations,
The contents of the reference chain directories 10A and 10B relating to the shared file File0 of the computers A and B are as shown in FIG.

(2) Operation Example 2 In the operation example 2, the operation when the computer A updates the shared file File0 in the state of FIG. 10C will be described.

When the AP 14 of the computer A updates the shared file File0 and issues a notice, the shared file management program 11 executes the routine RA of FIG.
First, since the file type identifier 102 of the reference chain directory 10A is Master, the process proceeds from step RA1 to step RA7, and the reference pointer 105 becomes N.
Since it is not ULL, the process proceeds to step RA8. In this step RA8, the computer (B) indicated by the reference pointer 105 is notified of itself (computer A) as the new local master and the absolute master for the shared file File0 by the inter-process communication P-INT1.

The shared file management program 11 of the computer B that has received the inter-process communication P-INT 1
The routine RC of FIG. 5 is executed. First, the shared file F
Since the file type identifier 102 in the reference chain directory 10B of file ile0 is not Master, the process proceeds from step RC1 to step RC3, and the file type identifier 102 is set to C.ile0. From Copy to D.S. Change to Copy. Next, since the reference pointer 105 is NULL, the process proceeds from step RC4 to step RC7, and the identifier of the computer A which is the absolute master location of the reference chain directory 10B is set in the reference pointer 105. Next, the computer A, which is the new local master and the new absolute master notified by the inter-process communication P-INT1, is set as the local master pointer 103 and the master pointer 104 (step RC8).
(Substantially unchanged), reference pointer 105
Communication P-I to the computer A indicated by
The computer A is notified as the local master pointer 103 and the master pointer 104 by NT1.

The shared file management program 11 of the computer A that has received the inter-process communication P-INT 1
The routine RC of FIG. 5 is executed. In this case, since the file type identifier 102 of the reference chain directory 10A of the shared file File0 is Master, the process proceeds from step RC1 to step RC2, where the inter-process communication P
-Send a Master notification to the computer B, which is the sender of INT1, by the inter-process communication P-INT2.

The shared file management program 11 of the computer B, which has received the inter-process communication P-INT2,
The routine RD shown in FIG.
The reference pointer 105 of the reference chain directory 10B is set to NULL (step RD1).

By the above operation, the state shown in FIG. 10C is changed as shown in FIG. As described above, when the computer A having the master of the shared file File0 performs the update, the computer B having the clear copy changes the state of the shared file File0 to the dirty copy.

(3) Operation Example 3 In this operation example 3, in the state of FIG.
A description will now be given of an operation for receiving the latest content of the shared file File0 and clear-copying the shared file File0 managed by itself.

The computer B executes the routine RE of FIG. 6 when the clear copy request of the shared file File0 is issued from the AP 14 or when the process is idle. First, the file type identifier 102 of the reference chain directory 10B of the shared file File0 is checked to determine whether or not it is a dirty copy (step RE1). In the state of FIG. 10D, since the copy is a dirty copy, the local master pointer 1 of the reference chain directory 10B is
03, the inter-process communication P-
The master file of the shared file File0 is requested by INT3 (step RE2).

The shared file management program 11 of the computer A, which has received the inter-process communication P-INT3,
The routine RF of FIG. 6 is executed. First, the shared file F
The file type identifier 102 of the reference chain directory 10A of ile0 is checked to determine whether it is a dirty copy (step RF1). In the state of FIG. 10D, since the copy is not a dirty copy, the shared file File0 is read from its own file device 2, and the inter-process communication P-
The signal is transmitted to the computer B by INT4 (step RF3).

The shared file management program 11 of the computer B that has received the inter-process communication P-INT 4
In step RE5 of the routine RE in FIG. 6, the reception completion is determined, and the received shared file File0 replaces the shared file File0 of the own file device 2. Next, the file type identifier 102 of the reference chain directory 10B of the shared file File0 is set to D. From Copy to C.I. Change to Copy.

By the above operation, the shared file File0 existing in the computer B has the latest contents, and the reference chain directory 10B is stored in FIG.
It is changed as shown in (e).

(4) Operation Example 4 In this operation example 4, an operation when the computer B updates the shared file File0 in the state of FIG. 10C will be described. FIG. 11A shows the same state as FIG. 10C again, and FIG.
(B) and (c) show.

[0049] On the computer B, the shared file File0
Is updated, the shared file management program 11
The routine RA of FIG. 5 is executed. First, if the file type identifier 102 of the reference chain directory 10B is a Mast
Therefore, the process proceeds from step RA1 to step RA2, and the file type identifier 102 is set to C. Change from Copy to Master. At this time, the identifier 122 of the directory 12 of the shared file File0 of the file device 2 is also changed to Master. Next, the computer A indicated by the master pointer 104 of the reference chain directory 10B is notified that the shared file File0 has been updated by the inter-process communication P-INT0 (step RB1).

The shared file management program 11 of the computer A, which has received the inter-process communication P-INT0,
The routine RB shown in FIG.
The file type identifier 102 of the reference chain directory 10A of D. Change to Copy (Step RB1). At this time, the identifier 122 of the directory 12 of the shared file File0 of the file device 2 of the computer A is also changed to Copy.

On the other hand, after the execution of step RA3, the shared file management program 11 of computer B executes step R3.
The process proceeds to A4, and it is determined whether or not the reference pointer 105 of the reference chain directory 10B is NULL. FIG.
In the case of 1 (a), since it is NULL, the identifier of the computer A set in the master pointer 104 of the reference chain directory 10B is set in the reference pointer 105 (step RA5), and then the local master pointer of the reference chain directory 10B. After setting Don't Care to the master pointer 103 and the master pointer 104 (step RA6), the inter-process communication P-IN is transmitted to the computer A set to the reference pointer 105.
By T1, the self (computer B) is notified as the new local master (new LM) and the absolute master (new AM) for the shared file File0 (step RA).
8).

The shared file management program 11 of the computer A that has received the inter-process communication P-INT 1
The routine RC of FIG. 5 is executed. First, the shared file F
Since the file type identifier 102 of the reference chain directory 10A of ile0 is not Master, step RC
1 and proceeds to step RC3, and sets the file type identifier 102 to D.1. Change to Copy (no substantial change since it has already been changed in step RB1). next,
Since the reference pointer 105 is not NULL, the process proceeds from step RC4 to step RC5, where the new local master and the new absolute master notified by the inter-process communication P-INT1 from the computer B are referred to as the local master pointer of the reference chain directory 10A. 103 and the master pointer 104 are set. Then, the computer B is notified as the local master pointer 103 and the master pointer 104 to the computer B indicated by the reference pointer 105 of the reference chain directory 10A by the inter-process communication P-INT1 (step RC6).

The shared file management program 11 of the computer B that has received the inter-process communication P-INT 1
The routine RC of FIG. 5 is executed. In this case, since the file type identifier 102 of the reference chain directory 10B of the shared file File0 is Master, the process proceeds from step RC1 to step RC2, where the inter-process communication P
-Send a Master notification to the computer A, which is the sender of INT1, by the inter-process communication P-INT2.

The shared file management program 11 of the computer A that has received the inter-process communication P-INT2
The routine RD shown in FIG.
The reference pointer 105 of the reference chain directory 10A is set to NULL (step RD1).

By the above operation, the state shown in FIG. 11A is changed as shown in FIG. As described above, when the computer B having the clear copy of the shared file File0 performs the update, the computer A having the master changes the state of the shared file File0 to the dirty copy, and the computer B replaces the computer A with the shared file. It becomes the master for file File0.

Since the file type identifier 102 of the reference chain directory 10A of the shared file File0 is a dirty copy, the shared file management program 11 of the computer A, as described in the operation example 3, operates as the master computer. The latest file is received from B, and the shared file File0 is made a clear copy. As a result, the state of FIG.
To change.

(5) Operation Example 5 In the operation example 5, in the state of FIG. 10C, the operation is performed when the computer C receives the transfer of the shared file File0 from the computer A in order to newly use the shared file File0. Will be described. FIG. 12 (a)
FIG. 12C shows the same state as FIG. 10C again, and FIG. 12B and FIG. 12C show subsequent transition states.

In the computer C, there is no copy of the shared file File0 in its own file device 2. Therefore, there is no reference chain directory 10 for the shared file File0 on the memory 3. Therefore, when the AP 14 attempts to use the shared file File0, the shared file File0 is transmitted to the shared file management program 11.
Is issued. The shared file management program 11 of the computer C executes the shared file File0
Does not exist in its own file device 2, the computer A having the master of the shared file File0
The transfer is requested via the bus BUS. In accordance with this request, the computer A reads the shared file File0 from its own file device 2 and transfers it to the computer C via the bus BUS. The computer C stores the received shared file File0 in its own file device 2, and creates the directory 12 in a format as shown in FIG. At this time, the name of the shared file File0 is set in the file name 121, and Copy is set in the identifier 122 indicating whether the file is the master file.

Next, the shared file management program 11 of the computer C, as shown in FIG.
And the file type identifier 102 as the name of the shared file File0. A reference chain directory 10C in which Copy is set is created on the memory.

Then, the shared file management program 11 of the computer C starts executing the routine RG shown in FIG. First, the identifier of the computer A which is the absolute master is set in the master pointer 104 of the created reference chain directory 10C (step RG1). Then
Communication between processes P-
By INT5, the self (computer C) notifies (reference notification) that it has acquired (referenced) a copy of the shared file File0 (step RG2).

The shared file management program 11 of the computer A that has received the inter-process communication P-INT5
The execution of the routine RH in FIG. 7 is started. In this case, the reference chain directory 10A of the shared file File0 of the computer A has the contents as shown in FIG. 12A, the file type identifier 102 indicates the Master, and the reference pointer 105 indicates the computer B. , RH4 via step RH7
To the computer B indicated by the reference pointer 105, by the inter-process communication P-INT5, that the computer C has acquired a copy of the shared file File0 (Reference notification).

The shared file management program 11 of the computer B, which has received the inter-process communication P-INT5,
The routine RH of FIG. 7 is executed. In this case, the reference chain directory 10B of the shared file File0 of the computer B has the contents as shown in FIG. Since the copy and reference pointer 105 are NULL, step RH
Then, the process proceeds to step RH5 via RH4 and changes the reference pointer 105 of the reference chain directory 10B to the identifier of the computer C notified from the computer A by the inter-process communication P-INT5.
Then, the inter-process communication PI indicates to the computer C that the local master is itself (computer B).
Notification is made by NT6 (step RH6).

The shared file management program 11 of the computer C that has received the inter-process communication P-INT 6
Step RG3 of the routine RG of FIG. 7 is executed, and FIG.
The identifier of the computer B which is the local master notified by the inter-process communication P-INT6 is set in the local master pointer 103 in the reference chain directory 10C of (b), and the reference pointer 105 is set to N.
Set the ULL.

By performing the above operations,
Shared file File0 of computers A, B and C
The contents of the reference chain directories 10A, 10B, and 10C for the reference chain are as shown in FIG. In this way, in a state where a plurality of computers A and B refer to the shared file File0, a new computer C transmits the shared file File0.
If you get a copy of
Has a clear copy, a new computer C
Is linked to the bottom of the reference chain.

(6) Operation Example 6 In the operation example 6, the operation in the case where the computer C receives the transfer of the shared file File0 from the computer A in order to use the shared file File0 in the state of FIG. I do. FIG. 13 (a) is the same as FIG.
2A, the file type identifier 102 of the reference chain directory 10B for the shared file File0 of the computer B The only difference is that the copy is made.

The shared file management program 11 of the computer C stores the shared file Fi in its own file device 2.
Since le0 does not exist, the computer A having the master of the shared file File0 is requested to transfer the file via the bus BUS. In response to this request, the shared file management program 11 of the computer A reads the shared file File0 from its own file device 2 and transfers it to the computer C via the bus BUS. The shared file management program 11 of the computer C stores the received shared file File0 in its own file device 2, and creates a directory 12 in a format as shown in FIG. At this time, the name of the shared file File0 is set in the file name 121, and Copy is set in the identifier 122 indicating whether the file is the master file.

Next, the shared file management program 11 of the computer C associates the shared file File0 with the shared file name 101 as shown in FIG.
And the file type identifier 102 as the name of the shared file File0. A reference chain directory 10C in which Copy is set is created in the memory 3.

Then, the shared file management program 11 of the computer C starts executing the routine RG shown in FIG. First, the identifier of the computer A which is the absolute master is set in the master pointer 104 of the created reference chain directory 10C (step RG1). Then
Communication between processes P-
By INT5, the self (computer C) notifies (reference notification) that it has acquired (referenced) a copy of the shared file File0 (step RG2).

The shared file management program 11 of the computer A that has received the inter-process communication P-INT 5
The execution of the routine RH in FIG. 7 is started. In this case, the reference chain directory 10A of the shared file File0 of the computer A has the contents as shown in FIG. 13A, the file type identifier 102 indicates the Master, and the reference pointer 105 indicates the computer B. , RH4 via step RH7
To the computer B indicated by the reference pointer 105, by the inter-process communication P-INT5, that the computer C has acquired a copy of the shared file File0 (Reference notification).

The operation up to the above is the same as that of the operation example 5. In the case of the operation example 6, since the shared file File0 of the computer B is a dirty copy, the following operation is different.

The shared file management program 11 of the computer B that has received the inter-process communication P-INT 5
The routine RH of FIG. 7 is executed. In this case, the reference chain directory 10B of the shared file File0 of the computer B has the contents as shown in FIG. Because it is Copy
The process proceeds to step RH2 via step RH1, and returns to the computer A that is the source of the inter-process communication P-INT5 via the inter-process communication P-INT7 that the shared file File0 is a dirty copy. Then, the identifier of the computer C notified from the computer A is set to the reference pointer 105 of the reference chain directory 10B (step RH3).

The shared file management program 11 of the computer A that has received the inter-process communication P-INT 7
The routine RI of FIG. 7 is executed, and the inter-process communication P-IN is executed.
At T8, the computer C notifies itself (computer A) as a local master and computer B as a new reference.

The shared file management program 11 of the computer C that has received the inter-process communication P-INT 8
Step RG4 of the routine RG of FIG. 7 is executed, and FIG.
The identifier of the computer A which is the local master notified by the inter-process communication P-INT8 is set to the local master pointer 103 in the reference chain directory 10C of (b), and the identifier of the computer B is set to the reference pointer 105. .

By performing the above operations,
Shared file File0 of computers A, B and C
The contents of the reference chain directories 10A, 10B and 10C relating to FIG. 13C are as shown in FIG. In this way, in a state where a plurality of computers A and B refer to the shared file File0, a new computer C transmits the shared file File0.
If you get a copy of
Is a dirty copy, the new computer C is linked to a higher level of computer B.

Note that the computer B stores the shared file F
Since the identifier 102 indicating whether or not the master file of the reference chain directory 10B of ile0 is a dirty copy, when the clear copy is requested from the AP 14, or using the free time of the process, the shared file Fil is used.
e0 clear copy processing is performed. At this time, since the computer C at a level immediately above the computer B has a clear copy, the shared file management program 11 of the computer B receives a copy of the shared file File0 from the computer C. As a result, the state of FIG. 13C changes as shown in FIG.

(7) Operation Example 7 In this operation example 7, an operation in the case where the computer A updates the shared file File0 in the state of FIG. 14A will be described. It should be noted that the state of FIG.
This is the same as the state shown in FIG.

[0086] The shared file File0 is
Is updated, the shared file management program 11
The routine RA of FIG. 5 is executed. First, if the file type identifier 102 of the reference chain directory 10A is “Mast”
er, the process proceeds from step RA1 to step RA7,
Since the reference pointer 105 is not NULL, the process proceeds to step RA8. In this step RA8, the computer B set in the reference pointer 105
To itself (computer A) as a new local master and absolute master for shared file File0 by inter-process communication P-INT1.

The shared file management program 11 of the computer B that has received the inter-process communication P-INT 1
The routine RC of FIG. 5 is executed. First, the shared file F
Since the file type identifier 102 in the reference chain directory 10B of file ile0 is not Master, the process proceeds from step RC1 to step RC3, and the file type identifier 102 is set to C.ile0. From Copy to D.S. Change to Copy. Next, since the reference pointer 105 is not NULL, the process proceeds from step RC4 to step RC5, where the identifiers of the computer A as the new local master and the new absolute master notified by the inter-process communication P-INT1 are stored in the local master pointer 103 and the master. Set to the pointer 104. Next, the local master pointer 103 is transmitted to the computer C indicated by the reference pointer 105 by the inter-process communication P-INT1.
Itself (computer B) as the master pointer 10
4 and notify computer A (step RC
6).

The shared file management program 11 of the computer C that has received the inter-process communication P-INT 1
The routine RC of FIG. 5 is executed. First, the shared file F
Since the file type identifier 102 in the reference chain directory 10C of file ile0 is not Master, the process proceeds from step RC1 to step RC3, and the file type identifier 102 is set to C. From Copy to D.S. Change to Copy. Next, since the reference pointer 105 is NULL, the process proceeds from step RC4 to step RC7, and sets the computer A which is the absolute master location of the reference chain directory 10C as the reference pointer 105.
Next, the computer A which is the new local master and the new absolute master notified by the inter-process communication P-INT1 is set as the local master pointer 103 and the master pointer 104 (step RC8), and the computer A indicated by the reference pointer 105 is set. On the other hand, the local master pointer 103 and the master pointer 104 are notified by the inter-process communication P-INT1 (step RC9).

The shared file management program 11 of the computer A that has received the inter-process communication P-INT 1
The routine RC of FIG. 5 is executed. In this case, since the file type identifier 102 of the reference chain directory 10A of the shared file File0 is Master, the process proceeds from step RC1 to step RC2, where the inter-process communication P
-A master notification is sent to the computer C, which is the transmission source of the INT1, by the inter-process communication P-INT2.

The shared file management program 11 of the computer C that has received the inter-process communication P-INT2
The routine RD shown in FIG.
The reference pointer 105 of the reference chain directory 10C is set to NULL (step RD1).

By the above operation, the state of FIG. 14A is changed as shown in FIG. As described above, when the computer A having the master of the shared file File0 performs the update, all the computers B and C having the clear copy perform the update of the shared file File at once.
Change the status of 0 to dirty copy.

The computers B and C store the reference chain directories 10B and 10C of the shared file File0.
Since the file type identifier 102 is dirty copy, the clear copy is performed when there is a clear copy request from the AP 14 or using the idle time of the process. In this case, the clear copy processing in the computer B is the same as that in the operation example 3. Computer B
When the computer C clear-copies the shared file File0 after completing the clear-copy of the shared file File0, similarly to the operation example 7, the computer C receives the copy from the computer B at the immediately higher level. . On the other hand, the shared file Fi of the computer B
If the computer C first performs the clear copy of the shared file File0 in a state where le0 is a dirty copy, the operation is slightly different. Therefore, the operation in that case will be described below.

The shared file management program 11 of the computer C executes the routine RE of FIG. 6 when a request for clear copy of the shared file File0 is issued from the AP 14 or when the process is idle. First, the file type identifier 102 of the reference chain directory 10C of the shared file File0 is checked to determine whether or not it is a dirty copy (step RE1). In the case of the state shown in FIG. 14B, since the copy is a dirty copy, the inter-process communication P-INT3 communicates with the computer B indicated by the local master pointer 103 of the reference chain directory 10C.
The master file of the shared file File0 is requested (step RE2).

The shared file management program 11 of the computer B that has received the inter-process communication P-INT3
The routine RF of FIG. 6 is executed. First, the shared file F
The file type identifier 102 of the reference chain directory 10B of ile0 is checked to determine whether or not it is a dirty copy (step RF1). In the case of the state of FIG. 14B, since the copy is a dirty copy, a transmission wait is returned to the computer C by the inter-process communication P-INT4 (step RF2).

The shared file management program 11 of the computer C that has received the inter-process communication P-INT4
The transmission waiting reply is identified in step RE4 of the routine RE, and the transmission request is waited for a predetermined time (step RE4).
3). Then, after a lapse of a predetermined time, the process returns to step RE2, and again sends the inter-process communication P-
The master file of the shared file File0 is requested by INT3.

In parallel with the above processing, the computer B performs clear copy of the shared file File0 using the idle time of the process. Then, on the computer B, the shared file File0
Is completed, the file type identifier 102 of the reference chain directory 10B becomes C.I. Cop
Therefore, if the inter-process communication P-INT3 is received from the computer C thereafter, the computer B determines NO in the step RF1 of the routine RF, reads the shared file File0 from the file device 2 of the computer B, and reads the shared file File0. The information is transmitted to the computer C by the communication P-INT4 (step RF3).

The shared file management program 11 of the computer C that has received the inter-process communication P-INT4
In step RE5 of the routine RE, completion of reception is determined, and the received file
Is replaced with the shared file File0. Next, the file type identifier 102 of the reference chain directory 10C of the shared file File0 is set to C.I. Change to Copy.

By the above-described clear copy processing, the state of FIG. 14B changes as shown in FIG. 14C.

(8) Operation Example 8 In this operation example 8, the operation when the computer B updates the shared file File0 in the state of FIG. 15A will be described. It should be noted that the state of FIG.
This is the same as the state shown in FIG.

Computer B shares file File0
Is updated, the shared file management program 11
Executes the routine RA of FIG. First, if the file type identifier 102 of the reference chain directory 10B is Ma
Steps RA1 to RA2
The file type identifier 102 is set to C. Copy
To Master. At this time, the file type identifier 122 of the directory 12 of the shared file File0 of the file device 2 is changed to Master. Next, the computer A indicated by the master pointer 104 of the reference chain directory 10B is notified that the shared file File0 has been updated by the inter-process communication P-INT0 (step RA3).

The shared file management program 11 of the computer A, which has received the inter-process communication P-INT0,
The routine RB shown in FIG.
The file type identifier 102 of the reference chain directory 10A of D. Change to Copy (Step RB1). At this time, the identifier 12 of the directory 12 of the shared file File0 of the file device 2 is simultaneously changed to Copy.

On the other hand, after executing step RA3, the shared file management program 11 of the computer B executes step R3.
The process proceeds to A4, and it is determined whether or not the reference pointer 105 of the reference chain directory 10B is NULL. FIG.
In the case of 5 (a), since it is not NULL, step RA5
And proceeds to step RA6 to set Don't Care in the local master pointer 103 and the master pointer 104 of the reference chain directory 10B. -By INT1
Notify itself (computer B) as the new local master and absolute master for shared file File0.

The shared file management program 11 of the computer C, which has received the inter-process communication P-INT1,
The routine RC of FIG. 5 is executed. First, the shared file F
Since the file type identifier 102 of the reference chain directory 10C of ile0 is not Master, step RC
1 proceeds to step RC3, and sets the file type identifier 102 to C.1. From Copy to D.S. Change to Copy. Next, since the reference pointer 105 is NULL, the process proceeds from step RC4 to step RC7, where the identifier of the computer A set as the master pointer 104 of the reference chain directory 10C is referred to as the reference pointer 105.
After that, the identifiers of the computer B as the new local master and the new absolute master notified from the computer B by the inter-process communication P-INT1 are set in the local master pointer 103 and the master pointer 104 of the reference chain directory 10C ( Step RC
8). Then, the inter-process communication P-INT1 notifies the computer A indicated by the reference pointer 105 of the reference chain directory 10C of itself (computer C) as the local master location 103 and the computer B as the absolute master location 104. (Step RC9).

The shared file management program 11 of the computer A that has received the inter-process communication P-INT 1
The routine RC of FIG. 5 is executed. First, the shared file F
Since the file type identifier 102 of the reference chain directory 10A of ile0 is not Master, step RC
1 and proceeds to step RC3, and sets the file type identifier 102 to D.1. Change to Copy (substantially no change). Next, since the reference pointer 105 is not NULL, the process proceeds from step RC4 to step RC5,
The identifier of the computer B as the new local master and the new absolute master notified from the computer B by the inter-process communication P-INT1
0A is set to the local master pointer 103 and the master pointer 104. The local master location 103 and the absolute master location 104 are communicated to the computer B indicated by the reference pointer 105 of the reference chain directory 10A by the inter-process communication P-INT1.
Is notified (step RC6).

The shared file management program 11 of the computer B that has received the inter-process communication P-INT 1
Execute the routine RC. In this case, the shared file Fi
Since the file type identifier 102 of the reference chain directory 10B of le0 is Master, step RC
1 to step RC2, and the inter-process communication P-IN
A master notification is sent to the computer A, which is the source of T1, by the inter-process communication P-INT2.

The shared file management program 11 of the computer A, which has received the inter-process communication P-INT2,
The routine RD is executed, and the reference pointer 10 of the reference chain directory 10A of the shared file File0 is executed.
5 is set to NULL (step RD1).

By the above operation, the state shown in FIG. 15A is changed as shown in FIG. As described above, when the computer B having the clear copy of the shared file File0 performs the update, the computer A having the master changes the state of the shared file File0 to the dirty copy. Also, the computer B becomes the master for the shared file File0 instead of the computer A. Further, the computer C having the clear copy of the shared file File0 also changes the state of the shared file File0 to dirty copy. In this case, the computer A is at the lowest level.

The computers A and C store the reference chain directories 10A and 10C of the shared file File0.
Since the file type identifier 102 is dirty copy, as described in the operation example 8, the latest file is received from the computer B or the like as the master during the idle time of the process or the like, and the shared file File0 is cleared. As a result, the state of FIG.
It changes like

(9) Operation Example 9 In the operation example 9, the operation when the computer C updates the shared file File0 in the state of FIG. 16A will be described. It should be noted that the state of FIG.
This is the same as the state shown in FIG.

The shared file File0 is stored on the computer C.
Is updated, the shared file management program 11
Executes the routine RA of FIG. First, if the file type identifier 102 of the reference chain directory 10C is Ma
Steps RA1 to RA2
The file type identifier 102 is set to C. Copy
To Master. At this time, the identifier 122 of the directory 12 of the shared file File0 of the file device 2 is changed to Master at the same time. Next, the inter-process communication P-IN is transmitted to the computer A indicated by the master pointer 104 of the reference chain directory 10C.
A notification that the shared file File0 has been updated is sent by T0 (step RA3).

The shared file management program 11 of the computer A that has received the inter-process communication P-INT0
The routine RB shown in FIG.
The file type identifier 102 of the reference chain directory 10A of D. Change to Copy. At this time, the shared file File0 of the file device 2 is simultaneously
, The identifier 122 of the directory 12 is changed to Copy.

On the other hand, after the execution of step RA3, the shared file management program 11 of the computer C executes step R3.
The process proceeds to A4, and it is determined whether or not the reference pointer 105 of the reference chain directory 10C is NULL. FIG.
In the case of 6 (a), since it is NULL, the identifier of the computer A set in the master pointer 104 of the reference chain directory 10C is set in the reference pointer 105, and then the local master pointer 103 and the master pointer of the reference chain directory 10C. D to 104
After setting on't Care (step RA6),
It notifies the computer A indicated by the reference pointer 105 of itself (computer C) as the new local master and the absolute master for the shared file File0 by the inter-process communication P-INT1.

The shared file management program 11 of the computer A, which has received the inter-process communication P-INT1,
The routine RC of FIG. 5 is executed. First, the shared file F
Since the file type identifier 102 of the reference chain directory 10A of ile0 is not Master, step RC
1 and proceeds to step RC3, and sets the file type identifier 102 to D.1. Change to Copy (substantially no change). Next, since the reference pointer 105 is not NULL, the process proceeds from step RC4 to step RC5,
The identifier of the computer C as the new local master and the new absolute master notified from the computer C by the inter-process communication P-INT1
0A is set to the local master pointer 103 and the master pointer 104. Then, the local master pointer 103 and the master pointer 104 are communicated to the computer B indicated by the reference pointer 105 of the reference chain directory 10A by the inter-process communication P-INT1.
Is notified (step RC6).

The shared file management program 11 of the computer B that has received the inter-process communication P-INT 1
The routine RC of FIG. 5 is executed. First, the shared file F
Since the file type identifier 102 of the reference chain directory 10B of ile0 is not Master, step RC
1 proceeds to step RC3, and sets the file type identifier 102 to C.1. From Copy to D.S. Change to Copy. Next, since the reference pointer 105 is not NULL, the process proceeds from step RC4 to step RC5, where the identifiers of the computer C which is the new local master and the new absolute master notified from the computer A by the inter-process communication P-INT1 are referred to as the reference chain directory. 10B local master pointer 103 and master pointer 1
Set to 04. Then, the reference chain directory 10
Computer C indicated by reference pointer 105 of B
, The local master pointer 103 and the master pointer 104 are notified by the inter-process communication P-INT1 (step RC6).

The shared file management program 11 of the computer C that has received the inter-process communication P-INT 1
The routine RC of FIG. 5 is executed. In this case, since the file type identifier 102 of the reference chain directory 10C of the shared file File0 is Master, the process proceeds from step RC1 to step RC2, where the inter-process communication P
-Send a Master notification to the computer B, which is the sender of INT1, by the inter-process communication P-INT2.

The shared file management program 11 of the computer B, which has received the inter-process communication P-INT2,
The routine RD shown in FIG.
The reference pointer 105 of the reference chain directory 10B is set to NULL (step RD1).

By the above operation, the state of FIG. 16A is changed as shown in FIG. As described above, when the computer C having the clear copy of the shared file File0 updates the file, the computer C becomes the master for the shared file File0 in place of the computer A even at the lowest level in the reference chain. Further, both the computer A having the master and the computer B having the clear copy change the state of the shared file File0 to dirty copy.

The computers A and B store the reference chain directories 10A and 10B of the shared file File0.
Since the master file identifier 102 is dirty copy, the latest file is received from the master computer C or the like using the idle time of the process, and the shared file File0 is Make a clear copy. Thus, the state of FIG. 16B changes as shown in FIG.

(10) Operation Example 10 In this operation example 10, an operation when a failure occurs in the computer A or B in the state of FIG. The state shown in FIG. 17A is the same as the state shown in FIG.

The shared file management program 11 of each computer monitors the presence or absence of a failure in the nearest upper-level computer on the reference chain relating to the shared file File0, and takes necessary measures at the time of failure detection. The routine RJ of FIG. 8 is executed.

First, the shared file management program 11
Reference chain directory 10 for shared file File0
It is checked whether or not the file type identifier 102 is Master (step RJ1). If it is Master, the process ends. In the case of not being a Master, that is, on the reference chain relating to the shared file File0, the shared file management program 11 of the computer other than the master transmits the shared file Make an alive check request by the inter-process communication P-INT9 (step RF2). Therefore, in the case of FIG. 17A, an alive check request is issued from the computer B to the computer A by the inter-process communication P-INT9, and the alive is requested from the computer C to the computer B by the inter-process communication P-INT9. A check request is issued.

The shared file management program 11 of the computer A that has received the inter-process communication P-INT 9
The routine RK of FIG. 8 is executed, and an alive check reply is sent to the computer B by the inter-process communication P-INT10.
(Step RK1). However, if a failure has occurred in the computer A, such a reply cannot be made. Similarly, the shared file management program 1 of the computer B that has received the inter-process communication P-INT9
1 executes the routine RK and executes the inter-process communication P-IN.
An alive check reply is sent to the computer C by T10 (step RK1). However, if a failure has occurred in the computer B, such a reply cannot be made.

The shared file management program 11 of the computers B and C, which issued the alive check request to the computers A and B, sends the alive check reply from the computers A and B within a predetermined time after the execution of the step RJ2 of the routine RJ. It is checked whether it has been received (steps RJ3, RJ4). Then, when an alive check reply is received from computers A and B within a certain time,
This monitoring processing ends. On the other hand, when the alive check replies are not received from the computers A and B within the predetermined time, it is considered that a failure has occurred in the computers A and B, and the process proceeds to step RJ5 and thereafter.

Now, consider a case where a failure has occurred in the computer A. In this case, the shared file management program 11 of the computer B transmits the local master pointer 10 of the reference chain directory 10B of the shared file File0.
3 and whether the master pointer 104 is equal or not. In the case of FIG.
Reference chain directory 10 for shared file File0
B, the file type identifier 102 is C. It is checked whether or not it is Copy (step RJ6). And C.I. Cop
If it is y, the file type identifier 102 of the reference chain directory 10B is changed to Master so that the master itself becomes the master of the shared file File0 (step RJ7). At this time, the identifier 122 of the directory 12 of the shared file File0 of the file device 2 is M
changed to aster. Also, the local master pointer 103 and the master pointer 104 are set to Don't
Set to Care (step RJ8). Next, it is determined whether or not the reference pointer 105 is NULL.
If it is UL, the process ends at that point (step RJ9). In the case of FIG. 17A, since it is not NULL,
The computer C indicated by the reference pointer 105 is notified of itself (computer B) as a new master by the inter-process communication P-INT11 (step RJ10).

The shared file management program 11 of the computer C that has received the inter-process communication P-INT 11
Executes the routine RL of FIG. First, the identifier of the computer B as the new master notified from the computer B by the inter-process communication P-INT11 is set to the master pointer 104 of the reference chain directory 10C of the shared file File0 (step RL1). Next, it is checked whether or not the reference pointer 105 of the reference chain directory 10C is NULL (step R).
L2) If it is not NULL, the inter-process communication P is sent to the computer indicated by the reference pointer 105.
-The new master (computer B) is notified by INT11 (step RL3), but in the case of FIG.
Since it is UL, the process ends at that point.

By performing the above processing,
The state of FIG. 17A is changed as shown in FIG. As described above, when a failure occurs in the computer A that is the master of the shared file File0, the computer A is removed from the reference chain of the shared file File0, and the computer B having the clear copy becomes the new master.

In this embodiment, in step RJ6 of the routine RJ, the shared file F
If the file type identifier 102 in the reference chain directory 10B of file ile0 is C.I. If it is not Copy,
Although the processing has been completed, as another embodiment, the C.I. It is also possible to search for a computer having Copy and update the reference chain directory in cooperation with the shared file management program 11 of each computer so that the computer becomes the master of the shared file File0.

Next, consider the case where a failure has occurred in the computer B. In this case, the shared file management program 11 of the computer C transmits the local master pointer 10 of the reference chain directory 10C of the shared file File0.
It is checked whether or not 3 is equal to the master pointer 104 (step RJ5). In the case of FIG.
Proceeding to step RJ11, the routine RM of FIG. 9 is executed.

First, the shared file management program 11 of the computer C determines that the computer B has failed (failure LM) with respect to the computer A indicated by the master pointer 104 of the reference chain directory 10C of the shared file File0. Regarding File0, the computer at the immediately lower level of computer B is itself (computer C) (reference),
The notification is made by the inter-process communication P-INT12 (step RM1).

The shared file management program 11 of the computer A that has received the inter-process communication P-INT 12
Executes the routine RN. First, the shared file Fi
It is checked whether or not the reference pointer 105 of the reference chain directory 10A of le0 matches the computer B which is the failure LM notified from the computer C by the inter-process communication P-INT12 (step RN).
1). If they do not match, the reference pointer 10
5 sends the same content as that notified from the computer C to the computer indicated by the inter-process communication P-IN.
Notification is made by T12. In the case of FIG.
Since they match, the process proceeds to step RN2. This step R
In N2, the reference (computer C) notified from the computer C by the inter-process communication P-INT12 is set in the reference pointer 105 of the reference chain directory 10A. Next, the new LM is transmitted to the computer C by the inter-process communication P-INT13.
(Computer A).

The computer C that has received the inter-process communication P-INT 13 sends the new LM (computer A) notified from the computer A to the local master pointer 103 of the reference chain directory 10C of the shared file File0 in step RM2 of the routine RM. ) Is set.

By performing the above processing,
The state of FIG. 17A is changed as shown in FIG. As described above, when a failure occurs in the computer B in the middle of the reference chain of the shared file File0, the computer B is removed from the reference chain of the shared file File0.

[0124]

As described above, according to the present invention, the following effects can be obtained.

The processing load on the shared file is distributed, and the overall throughput is improved. The reason is that the computer holding the master of the shared file is dynamically changed with the file update.

Even if a failure occurs in the computer holding the master of the shared file, all the remaining computers can continue to use the shared file as long as possible. The reason is that when a failure of the computer having the master of the shared file is detected, another computer having a clear copy of the shared file becomes the new master.

If a failure occurs on the computer holding the copy of the shared file, the failed computer can be removed from the reference chain. Thereby, it is possible to prevent the reference chain from being broken in the middle.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an example of a loosely coupled multiprocessor system to which the present invention is applied.

FIG. 2 is a diagram showing an example of the contents of a directory 12;

FIG. 3 is a diagram showing an example of the contents of a reference chain directory 10;

FIG. 4 is a flowchart illustrating a processing example when an AP (application program) refers to and updates a shared file.

FIG. 5 shows a processing example of the shared file management program;
9 is a flowchart of a routine related to a process executed when a shared file is updated.

FIG. 6 shows a processing example of the shared file management program;
It is a flowchart of the routine relevant to the process of clear copy of dirty copy.

FIG. 7 shows a processing example of the shared file management program.
9 is a flowchart of a routine related to a process when a copy of a shared file that does not exist in the file device of the own computer is acquired from the computer having the master.

FIG. 8 shows a processing example of a shared file management program;
9 is a flowchart of a part of a routine related to a process of monitoring a failure of another computer and detecting a failure.

FIG. 9 shows a processing example of the shared file management program;
It is a flowchart of the remaining part of the routine related to the process of monitoring the failure of another computer and detecting the failure.

FIG. 10 is a diagram showing an example of the contents of a reference chain directory of each computer regarding a shared file File0 and an example of how the directory is changed.

FIG. 11 is a diagram showing another example of the contents of the reference chain directory of each computer and another example of the state of change with respect to the shared file File0.

FIG. 12 is a diagram showing another example of the contents of the reference chain directory of each computer and another example of the state of change with respect to the shared file File0.

FIG. 13 is a diagram showing another example of the contents of the reference chain directory of each computer and another example of a change in the shared file File0.

FIG. 14 is a diagram showing another example of the contents of a reference chain directory of each computer and another example of a change state with respect to the shared file File0.

FIG. 15 is a diagram illustrating another example of the contents of the reference chain directory of each computer and another example of a change state regarding the shared file File0.

FIG. 16 is a diagram showing another example of the contents of the reference chain directory of each computer and another example of a change in the shared file File0.

FIG. 17 is a diagram showing another example of the contents of the reference chain directory of each computer and another example of a change state regarding the shared file File0.

[Explanation of symbols]

 A to D computer BUS bus 1 CPU 2 file device 3 memory 4 communication device 10 reference chain directory 11 shared file management program 12 directory 13 shared file 14 AP (application program)

────────────────────────────────────────────────── ─── Continuation of the front page (58) Field surveyed (Int. Cl. ⁷ , DB name) G06F 15/177 682 G06F 12/00 533 JICST file (JOIS)

Claims (3)

(57) [Claims] 1. A file sharing method in a loosely-coupled multiprocessor system in which a plurality of computers each having a CPU, a memory, and a file device are connected to a network, wherein the computer attempts to use a shared file that does not exist in its own file device. Receives the transfer of the copy from the computer that owns the master of the shared file and stores it in the file device of the own computer. The computer that holds the master or copy of the shared file in the file device of the own computer uses the memory of the own computer. On the top, the shared file name, an identifier indicating whether the shared file is the master, a clear copy of the same contents as the master or a dirty copy that is an old copy, an identifier of a computer holding the master, and a master or a copy. Create a reference chain directory containing bidirectional pointers to create a reference chain linking the master computer at the highest level with all computers that have a master, and a computer that has a dirty copy of a shared file ,
Before using the shared file, a clear copy of the shared file or a copy of the shared file is transferred from the computer that has the master and is closest to the own computer in the reference chain related to the shared file. If a computer with a master or clear copy of a shared file updates the shared file, the computers work together to make the updated shared file a new master and all others dirty copies. And updating the reference chain directory for the shared file. 2. A computer holding a copy of a shared file monitors whether there is a failure in a computer at the immediately higher level on the reference chain relating to the shared file. If detected, the failed computer is removed from the reference chain, and each computer cooperates so that another computer having a clear copy of the shared file becomes a new master. 2. The file sharing method according to claim 1, wherein the chaining directory is updated. 3. When the monitoring detects a failure of a computer having a copy of the shared file, the computers cooperate with each other to remove the failed computer from the reference chain. 3. The file sharing method according to claim 2, wherein the directory is updated.