JPH0559463B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a method for controlling simultaneous access to files shared between computers in a loosely coupled multiprocessor system.

[Conventional technology]

In conventional file simultaneous access control in a loosely coupled multiprocessor system, the control unit is a block or a file, and control is not performed on a record-by-record basis.

On the other hand, in a single processor system or a tightly coupled multiprocessor system, simultaneous file access control has been performed on a record-by-record basis, for example, using a configuration as shown in FIG. In this figure, shared exclusive control is a basic function used in file simultaneous access control, and controls the sharing or exclusive use of resources among multiple tasks using the requested logical resource name as a key. In this control, since it is not concerned which physical resource a logical resource name corresponds to, it is possible to associate the logical resource name with a block or with a record. In addition, shared buffer management allocates a shared buffer area so that if the buffers correspond to the same block, the same buffer is shared among a plurality of tasks requesting that block. Tasks that perform simultaneous access in units of records request shared exclusive control by associating records as logical resource names, and file updates are performed using buffers allocated by shared buffer management in association with blocks that include the records. Do it through areas.

The necessity of updating files via a shared buffer in the above example will now be explained with reference to FIG. This figure shows the operation when a buffer area is provided for each task in a single processor system. Task _T1 is
In order to update record _R1 , after exclusively securing record _R1 , block _B1 containing record _R1 is read into the buffer area of the invoking task. In addition, the record
Input/output operation using only _R1 is not possible.

Then task T ₂ updates record R ₂ , so
Exclusively secure record _R2 . Since record R ₁ and record R ₂ are different resources, no waiting occurs. After that, the same block _B1 containing record _R2 is read into the buffer area of the invoking task.
Task T ₁ records in the buffer area of its own task
After changing R ₁ to R ₁ ′, this changed block B ₁ is written, and then task T ₂ changes record R ₂ to R ₂ ′ and then writes this changed block B ₁ .
As a result, since this change from R ₁ to R ₁ ' has not been made in task T ₂ , the update results made in task T ₁ are overwritten by the update results made in task 2 and are lost.

[Problem that the invention seeks to solve]

In the above-mentioned prior art, in a loosely coupled multiprocessor system, even if it is desirable to control simultaneous file access on a record-by-record basis, simultaneous file access control must be carried out on a block-by-block basis. For this reason,
There was a problem in that the frequency of occurrence of waiting due to access conflicts increased, and the processing time of tasks that accessed files simultaneously increased.

The reason why concurrent file access control on a record-by-record basis, which is possible in a single processor system or a tightly coupled multiprocessor system, is difficult in a loosely coupled multiprocessor system is that the input/output buffer area (usually , stored in main memory) cannot be shared between computers. Therefore, if the unit of shared exclusive control is simply a record, the result of updating the shared file is not guaranteed due to the same logic as when a buffer area is provided for each task in the single processor system described above.

[Means for solving problems]

The simultaneous file access control method in a loosely coupled multiprocessor system of the present invention controls simultaneous access to a file on a secondary storage device shared by a plurality of computers making up the loosely coupled multiprocessor system. As a control unit, a block is a control unit that is an integral multiple of the physical input/output unit,
When a task has a record that is a control unit smaller than a block, in response to a record unit file access request that specifies a record and a block that includes the record, a task running on the same computer as the requesting task is requested. In the control between records, if a conflict occurs in the record unit, a wait is performed,
If there is no conflict in the record unit, the access request is permitted, and in the control between the requesting task and a task running on a different computer, if there is no conflict in the block unit. A record sharing exclusive control means that waits for the access request if there is no contention in the block unit, and permits the access request if there is no conflict in the block, and allocation of buffers corresponding to the same block with respect to the shared buffer area that exists for each individual computer. If it is a request, it has a shared buffer management means that performs allocation control so that the same area is shared among multiple tasks requesting the buffer, and when accessing the shared file in record units, the sharing exclusive control is performed as described above. This is performed by a record sharing exclusive control means, and the physical input/output operation is performed via a buffer area allocated by the shared buffer management means.

〔Example〕

An embodiment of the present invention will be described with reference to the drawings.

Referring to FIG. 1, one embodiment of the present invention comprises a loosely coupled multiprocessor system having a shared exclusive control device 1 having a record sharing exclusive control means 11, a shared buffer management means 21, and a shared buffer area 22. The computer 2 includes a computer 2, and a secondary storage device 3 in which files shared among multiple computers exist. The system of this embodiment is composed of two computers 2, and has a shared exclusive control device 1 which is a device exclusively for shared exclusive control and is connected to each computer 2 through a channel.

Furthermore, a magnetic disk device is used as an example of the secondary storage device 3. Note that it is also possible to implement the present invention by replacing one or both of the computers 2 (single processors) with a tightly coupled multiprocessor system. Further, it is also possible to implement the shared exclusive control device 1 by incorporating it into one of the computers 2, or by having both computers 2 share the functions.

Next, the operation of this embodiment will be described in detail with reference to FIG. 4. First, the configuration of the shared file 31 on the secondary storage device 3, which is subject to simultaneous access control, will be described. The shared file 31 has a hierarchical structure of blocks and records. Among these, a block, also called a page, is a unit in which input/output operations are performed, and is a physical input/output operation unit (which varies depending on the device, e.g.
256 bytes). On the other hand, a record is a logical access unit and is included in any block. Note that it is also possible to define a record that spans multiple blocks, but in such a case, each block to which it belongs is treated as a separate record. Here, all the shared files 31 are given a file ID, which is a file identification name for uniquely identifying the file within the system. Furthermore, each block is given a block ID, which is a block identification name for uniquely identifying the block within the file to which it belongs. Furthermore, each record is given a record ID, which is a record identification name for uniquely identifying the record within the block to which it belongs. In the following, block specification refers to specifying a set of (file ID, block ID), and record specification refers to specifying a set of (file ID, block ID, record ID).

Next, the operation of simultaneous file access control will be explained along the flow of processing requests. First, a task that performs simultaneous file access makes the following request to the simultaneous file access control routine 24.

(1) When reading a record, specify the record and
Make a GET request.

(2) When writing a record, specify the record and
Make a PUT request.

(3) When a logical group of files, usually called a checkpoint section, has been updated, a checkpoint request is made.

The processing procedure in the file simultaneous access control routine 24 is as follows.

(1) GET processing For the record sharing exclusive control means 11,
A lock request is made in shared mode specifying a record. At this time, if necessary, the record sharing exclusive control means 11 makes a wait.

A buffer reservation request is made to the shared buffer management means 21 by specifying a block containing the requested record. At this time, if necessary, the corresponding block is read into the shared buffer area 22 by the shared buffer management means 21.

The requested record is transferred from the buffer obtained by the shared buffer management means 21 to the record storage area of the requesting task.

A buffer release request is made to the shared buffer management means 21 in reference mode, specifying a processed block.

(2) PUT processing For the record sharing exclusive control means 11,
A lock request is made in exclusive mode by specifying a record. At this time, if necessary, the record sharing exclusive control means 11 makes a wait.

A buffer reservation request is made to the shared buffer management means 21 by specifying a block containing the requested record. At this time, if necessary, the corresponding block is read into the shared buffer area 22 by the shared buffer management means 21.

The requested record is transferred from the record storage area of the requesting task to the buffer obtained by the shared buffer management means 21.

A buffer release request is made to the shared buffer management means 21 in update mode, specifying a processed block.

(3) Checkpoint processing A buffer invalidation request is made to the shared buffer management means 21 for all buffers referenced by the requesting task.

For the record sharing exclusive control means 11,
A request is made to unlock all records locked by the requesting task.

In this embodiment, while a certain block is being read into the shared buffer area 22 of one computer 2, any record included in that block is locked, and the record sharing exclusive control means 11, the same block is protected from being read into the shared buffer area 22 of the other computer 2.

Although efficient allocation control of the shared buffer area 22 is also important as a function of the shared buffer management means 21, only the functions related to file simultaneous access control will be described here.

First, the shared buffer management table 23, which is a management table used by the shared buffer management means 21, will be explained with reference to FIG.

Each entry in the management table is created for each corresponding block, and contains the file ID and block ID.
It includes block information consisting of , buffer area information consisting of a pointer to an area, an update flag, an in-use task list, and a reference task list.

Next, the processing procedure for each request will be explained.

(1) Buffer reservation processing Check whether the requested block already exists on the shared buffer management table 23.
If it exists, have the requesting task refer to the buffer whose contents have already been read. If it does not exist, a free buffer area is allocated and the requested block is read and referenced by the requesting task. In either case, the requested task is registered in the buffer management entry's in-use task list and reference task list.

If there is no necessary free buffer area, the buffer corresponding to the management entry whose busy task list is empty is released and allocated to the request block. At this time, if the buffer has an update flag set, writing to the corresponding block is performed before being assigned to the next block. If there are no buffers with no tasks in use, or even if there are, there is not enough space in those buffers, a wait is performed until a free space becomes available.

(2) Buffer release processing The requesting task is deleted from the in-use task list of the buffer management entry corresponding to the requested block. If the request is in update mode, the update flag of the buffer management entry is set.

(3) Buffer invalidation processing The requested task is deleted from the reference task list of all buffer management entries. Here, for buffer management entries whose reference task list is empty (at this time, there should be no tasks in use), if the update flag is set, after writing to the corresponding block. , the entry is deleted.

This buffer invalidation process is necessary in order to prevent blocks that are not locked for all the records they contain from being held in the shared buffer area 22.

Finally, the operation of the record sharing exclusive control means 11 will be explained.

First, the shared exclusive control table 12, which is a control table used by the record shared exclusive control means 11 shown in FIG. 4, will be explained with reference to FIG. Resources to be controlled are managed in two levels: blocks and records. The block control entry in the first layer contains block information consisting of a file ID and a block ID, the ID of the computer in use, the task queue waiting for the block, and the record control entry corresponding to the record belonging to the block.
Contains a list. The record control entry in the second hierarchy includes record information consisting of a record ID, a locked task list, and a record waiting task queue. Here, for each task in the locked task list, record-waiting task queue, and block-waiting task queue, the computer ID and task ID of computer 2 to which the task belongs and the shared mode and exclusive mode are listed. Lock request types such as ``lock requests'' are registered. Here, the computer ID is an identification name for uniquely identifying each computer 2 in the system, and also a task
The ID is an identification name for uniquely identifying each task within the corresponding computer 2.

Next, the processing procedure of the record sharing exclusive control means 11 will be explained with reference to a flowchart.

First, the lock request processing procedure will be explained in detail below with reference to FIG.

The block is not locked by a task on another computer 2 (corresponding to 71, 72), and there is no lock conflict on a record basis with other tasks on the same computer 2 (73, 74).
) is checked, and if the lock can be permitted, the requesting task is registered in the locking task list of the corresponding record control entry. If locking is not possible, the requesting task is registered at the end of the record-waiting task queue or block-waiting task queue, and is kept waiting until locking becomes possible.

Note that since the double lock processing is the same as the conventional processing, the processing flow is omitted. If the existing lock type indicates shared, the lock type of the request indicates exclusive, and there are other tasks in the shared lock, it is necessary to wait until there are no more tasks in the shared lock. The lock type after double lock processing is shared if both existing and requested locks are shared, and exclusive in any other combination.

Next, Section 8 describes the unlock request processing procedure.
This will be explained in detail with reference to the drawings.

First, the unlock requesting task is deleted from the locked task list of the corresponding record control entry, and if there are no more locked tasks, lock wait release processing is performed. In the record waiting release process 81, the record waiting task queue of the record control entry is sequentially executed from the head, and in the block waiting canceling process 82, the block waiting task queue of the block control entry is again sequentially executed from the head of the block waiting task queue, which is equivalent to the lock process. Processing takes place.

In the above description, only the shared exclusive control in units of records, which is a feature of the present invention, has been explained, but it is also possible to incorporate the conventional shared exclusive control in units of blocks into the shared exclusive control device 1 at the same time. In addition, as has traditionally been done,
A deadlock avoidance mechanism is incorporated into the shared exclusive control device 1, before processing to wait for a lock request, to check whether a deadlock state will occur if the waiting is performed, and if a deadlock state occurs, to reject the lock request without performing the waiting. is also possible.

〔Effect of the invention〕

As explained above, in the present invention, when controlling simultaneous file access in a loosely coupled multiprocessor system, the hierarchies of blocks and records are taken into consideration, and access conflicts between tasks running on the same computer are handled on a record-by-record basis. Shared exclusive control is performed. For this reason, compared to the conventional case of performing shared exclusive control on a block-by-block basis,
This has the effect of reducing the frequency of waits caused by access conflicts between tasks that access files simultaneously, and improving the processing capacity of the system.

[Brief explanation of drawings]

Fig. 1 is a diagram showing an embodiment of the present invention, Fig. 2 is a diagram for explaining record unit file simultaneous access control in a single processor system, and Fig. 3 is a record unit file simultaneous access control without a shared buffer. A diagram to explain the operation,
FIG. 4 is a diagram for explaining file simultaneous access control processing in one embodiment, FIG. 5 is a diagram showing the shared buffer management table 23, FIG. 6 is a diagram showing the shared exclusive control table 12, and FIG. 7 is a diagram showing the shared buffer management table 23. A diagram showing the locking process and FIG. 8 are diagrams showing the unlocking process. In FIGS. 1 to 8, 1...Sharing exclusive control device, 11...Record sharing exclusive control means, 1
2... Shared exclusive control table, 2... Calculator, 2
1... Shared buffer management means, 22... Shared buffer area, 23... Shared buffer management table,
24...File simultaneous access control routine, 3
...Secondary storage device, 31...Shared file.

Claims (1)

[Scope of Claims] 1. In a method for controlling simultaneous access to files on a secondary storage device shared by a plurality of computers constituting a loosely coupled multiprocessor system, physical input/output is used as a control unit of the simultaneous access control. A block is a control unit that is an integral multiple of the unit,
When a task has a record that is a control unit smaller than a block, in response to a record unit file access request that specifies a record and a block that includes the record, a task running on the same computer as the requesting task is requested. In the control between records, if a conflict occurs in the record unit, a wait is performed,
If there is no conflict in the record unit, the access request is permitted, and in the control between the requesting task and a task running on a different computer, if there is no conflict in the block unit. A record sharing exclusive control means that waits for a block if there is a conflict and permits an access request if there is no conflict in the block, and a buffer allocation request corresponding to the same block regarding the shared buffer area that exists for each individual computer. If so, the shared buffer management means performs allocation control so that the same area is shared among multiple tasks requesting the buffer, and when accessing the shared file in units of records, the shared exclusive control is applied to the record. A file simultaneous access control method in a loosely coupled multiprocessor system, characterized in that the file simultaneous access control is performed by shared exclusive control means, and physical input/output operations are performed via a buffer area allocated by the shared buffer management means.