JPH10320255A - Input/output device controller and method for check point/rollback type computer - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an I/O device controller and a method for a check point/ rollback type computer capable of preventing the issue of an I/O request to a disk from being delayed. SOLUTION: Since a computer system 10 connected to hard disks 11, 12 having partitions A, B and consisting of a processor 15, a main memory 13, a control table 14, and a disk controller 16 can issue an I/O request to a secondary storage device without waiting for a check point and distribute access frequency to respective disk drives 11, 12, the performance of the system at normal processing time can be improved.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a process in which a checkpoint is periodically collected and processing is performed. Checkpoint to avoid temporary system failure by executing
A failure recovery method of an input / output control method in a computer system having a rollback mechanism and storing and reading data from / to a secondary storage device, that is, an input / output device in a checkpoint / rollback computer The present invention relates to a control method, an input / output device control device in a checkpoint / rollback type computer using the method, and a method.

[0002]

2. Description of the Related Art Conventionally, processing is performed while periodically collecting checkpoints, and when a failure occurs in the system, the state of the system is returned (rolled back) to the time of the immediately preceding checkpoint, and processing is performed. In a computer system which has a checkpoint / rollback mechanism for avoiding a temporary failure of the system by re-executing and storing / reading data with / from a secondary storage device, Even after the rollback due to the occurrence, it is necessary to guarantee the consistency between the state of the main memory and the processor and the I / O device (such as the contents of the disk). For this reason, in the normal checkpoint-rollback method, an I / O request from a program is issued to a disk after the issuance is determined by the next checkpoint acquisition.

[0003]

However, in the above input / output control method, issuance of at least a request to change (write) the state of the input / output device is delayed after the next checkpoint collection process. In other words, since the requests delayed until the checkpoint point are issued after the next checkpoint processing, the checkpoint-rollback method degrades the performance of the computer system for achieving high reliability. I was

Accordingly, the present invention has been made in view of the above circumstances, and solves the above-mentioned disadvantages, and provides an I / O for a disk.
An input / output device control method for a checkpoint / rollback computer that does not delay the issuance of an O request, and an input / output device control device and method for a checkpoint / rollback computer using this method are provided. The purpose is to:

[0005]

According to the present invention, in order to achieve the above-mentioned object, the processing is performed while periodically collecting checkpoints, and when a failure occurs in the system, the system is reset to the time of the immediately preceding checkpoint. It has a checkpoint-rollback mechanism to avoid temporary system failure by performing a rollback to return the state and re-executing the process, and stores and reads data to and from the secondary storage device. In a computer system, a plurality of data blocks are allocated for one piece of data on the secondary storage device, and which of the plurality of data blocks is a checkpoint image at the time of the last checkpoint and the latest state Means having a management table for recording whether or not the data is written to the secondary storage device. When the request is received, a data block that is not a checkpoint image is selected from a plurality of data blocks corresponding to the data received by referring to the management table on the secondary storage device, and the selected data block is Means for recording, in the management table, information indicating that there is a latest data block, not a checkpoint image, and for writing data to the selected data block; and a means for reading data from the secondary storage device, Means for reading data from the data block designated as the latest data block in the management table among the plurality of data blocks corresponding to the data requested by referring to the management table on the secondary storage device; When processing points, the above management table In the above-mentioned management, it is determined that the block registered as the latest data block among the plurality of data blocks corresponding to all data on the secondary storage device by referring to the data is the latest and is a checkpoint image. Means for registering in the table, and when the system is returned to the latest checkpoint state and processing is resumed, referring to the above-mentioned management table, for all data, of a plurality of corresponding data blocks on the secondary storage device Means for registering in the management table that a block registered as a checkpoint image is a checkpoint image and the latest one in the management table. In the output device controller. According to such a configuration, in a checkpoint-rollback computer system, an I / O request to a secondary storage device can be issued without waiting for a checkpoint, and the frequency of access can be reduced by each disk. Since the data can be distributed to the drives, the performance of the system during normal processing can be improved.

In order to achieve the above object, according to the present invention, a process is carried out while periodically collecting checkpoints, and when a failure occurs in the system, the state of the system is returned to the immediately preceding checkpoint. In a computer system that has a checkpoint-rollback mechanism that avoids a temporary failure of the system by re-executing the process in the back, and stores and reads data to and from the secondary storage device. Allocate two data blocks on the secondary storage device for one data, and record which of the two data blocks is the checkpoint image which is the state at the time of the last checkpoint and which is the latest state Means having a management table, and a request to write data to the secondary storage device When the data block received from the management table is referred to, a data block that is not a checkpoint image is selected from two corresponding data blocks on the secondary storage device, and the selected data block is a checkpoint image. Means for writing data indicating that there is no newest data block in the management table, and writing data to the selected data block; and when a request to read data from the secondary storage device occurs, Means for reading data from the data block designated as the latest data block in the management table among the two data blocks corresponding to the data requested by reference on the secondary storage device; To all of the Means for registering in the management table that the block registered as the latest data block among the two corresponding data blocks on the secondary storage device is the latest and is a checkpoint image. When the system is returned to the state of the latest checkpoint and processing is resumed, it is determined that the data is the checkpoint image of the corresponding two data blocks on the secondary storage device with respect to all data by referring to the management table. Means for registering, in the management table, that the registered block is a checkpoint image and the latest one, in the input / output device control device in the checkpoint / rollback type computer. .

According to such a configuration, since an I / O request to the secondary storage device can be issued without waiting for a checkpoint, the performance of the system during normal processing can be improved.

[0008] Further, the present invention achieves the above object,
If a failure occurs in the system while the process is being performed while periodically collecting checkpoints, a rollback that returns the system status to the time of the immediately preceding checkpoint is performed and the process is re-executed. In a computer system that has a checkpoint-rollback mechanism to avoid and store and read data to and from a secondary storage device, one storage device belongs to the secondary storage device for one data item. Means for allocating a plurality of data blocks in the medium and having a management table for recording which of the plurality of data blocks is the checkpoint image at the time of the last checkpoint and which is the latest state; When a request to write data to the secondary storage device is received, A data block that is not a checkpoint image is selected from a plurality of corresponding data blocks on a storage medium belonging to the secondary storage device for data received by referring to the table, and the selected data block is not a checkpoint image. Means for recording the latest data block in the management table and writing data to the selected data block; and when a request to read data from the secondary storage device occurs, Means for reading data from the data block designated as the latest data block in the management table among the plurality of data blocks corresponding to the data requested by reference on a certain storage medium belonging to the secondary storage device, During checkpoint processing, all Means for registering in the management table that the latest data block among the corresponding data blocks on a certain storage medium belonging to the secondary storage device is the latest and is a checkpoint image; When the processing is resumed by returning to the check point state, the block which is the check point image among the plurality of data blocks corresponding to all data on a certain storage medium belonging to the secondary storage device by referring to the management table Means for registering the latest and the checkpoint image in the management table in the above-mentioned management table.

According to such a configuration, the RAI
The performance of the system during normal processing can be improved without impairing the performance and the fault-tolerant function due to D or the like.

[0010]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a diagram showing a system configuration. As shown, a computer system 10 has two partitions A and B similarly addressed on two hard disks 11 and 12, Is provided with a management table 14.

Each row of the management table 14 indicates the state of a data block allocated to each data in the partition A and the partition B. When the state of the data block is “CN”, it indicates that this data block is a checkpoint image (state of the last checkpoint) and is the latest. When the state of the data block is “C0”, this data block is a checkpoint image,
And it is not up to date. When the state of the data block is “ON”, it indicates that the state of the data block is not the checkpoint image and is the latest. When the state of the data block is “00”, it indicates that the state of the data block is not a checkpoint image and is not the latest.

The management table 14 stores A
It is assumed that each data block of the partition is a checkpoint image and is up-to-date (see FIG. 2).

When a read request is issued for a certain data block X, data is read from a corresponding data block of a partition marked as latest (N) in a row related to the data block X in the management table 14 ( 3). FIG.
If so, partition A (step S32 in FIG. 3),
In the case of FIG. 4, the B partition (step S3 in FIG. 3)
3).

When a write request is issued to a certain data block X, a partition in which the checkpoint image (C) is not marked for the data block X in the management table is selected and data is written to the corresponding data block ( 5). In the case of FIG. 2, the B partition is used.
partition. At this time, in FIG. 2, since the latest is not marked in the B partition where the writing is performed,
Erase the latest mark on the partition to mark the B partition as current. In FIG. 4, the B partition on which the writing has been performed is marked with the latest before the writing is performed, so that there is no change in the table.

At the time of a checkpoint (refer to the flowchart of FIG. 6), for each row of the management table 14, if the partition where the checkpoint image is marked is different from the partition where the latest is marked, the checkpoint Turn off the checkpoint image mark of the partition where the image is marked, and mark the partition with the latest mark with the latest and checkpoint image.

At the time of rollback (refer to the flowchart of FIG. 7), for each row of the management table 14, if the partition where the checkpoint image is marked is different from the partition where the latest is marked, the latest is marked. The latest mark of the current partition is erased, and the partition marked with the checkpoint is marked with the checkpoint image and the latest.

When the system is shut down, the system is shut down after storing the management table 14 in the non-volatile storage area for the next boot, or the partition with the latest mark and the check point are checked for each row of the management table 14. If the partition with the image mark is different, the data is copied from the partition with the latest mark to the partition with the checkpoint image mark, and after the contents of the partitions A and B are made consistent, the system is shut down.

Next, an example of updating the management table will be described.

When a read request is issued to the block X of the secondary storage device with the management table shown in FIG.
State is "CN" (checkpoint image,
Therefore, data is read from the block X of the partition A marked “N” (latest). At this point, block X of partition A is the checkpoint image and is the latest (see FIG. 8).

Thereafter, when a write request is issued to the block X, the state of the block X of the partition A is "CN" in the management table 14, so that the state of the block X of the partition A is "C0" and the state of the partition B is "C0". Is changed to “0N”, and data is written to the block X of the partition B. At this point, the checkpoint image of block X is in partition A and the latest data of block X is in partition B (see FIG. 9).

Further, when a read request is issued for block X, since the state of block X of partition A is "C0" in the management table, data is read from block X of partition B. X is the latest data (see FIG. 10).

In this state, when the system is returned to the state of the last checkpoint due to the occurrence of a failure or the like, the block registered as the checkpoint image in each row of the management table 14 is updated, and Register as a checkpoint image. As for the block X, the state of the block X of the partition A becomes “CN”, and the state of the block X of the partition B becomes “00” (see FIG. 11).

Therefore, a read request to the block X generated thereafter is read from the partition A which is the latest and the checkpoint image. When a write request to the block X is generated, the checkpoint image of the partition A is generated. Is written to the block X of the partition B while leaving Here, since the block X of the partition B is marked as the latest state, the subsequent reading of the block X is performed from the partition B (see FIG. 12).

When a checkpoint is taken in this state, the fact that the block X of the partition B that is the latest with respect to the block X is the latest and is a new checkpoint image is recorded on the management table. Then, the I / O processing can be continued (see FIG. 13).

According to the present embodiment, the checkpoint
Even in a rollback computer system, an I / O request to a secondary storage device can be issued without waiting for a checkpoint.

It should be noted that the management table 14 may be provided on a memory which is rolled back when a failure occurs. In this case, if the state after updating the table is used as a checkpoint image of the main memory in the checkpoint processing, the management table is re-read at the time of rollback. No configuration is required.

The present invention is not limited to the above-described embodiment, but can be applied to an input / output control method for a data processing device of a computer system which achieves high reliability by a checkpoint-rollback method. Of course.

[0029]

As described above in detail, according to the present invention, it is possible to provide an input / output device control system in a checkpoint / rollback computer which does not delay the issue of I / O requests to a disk. is there.

That is, according to the present invention, in a checkpoint-rollback computer system, an I / O request to a secondary storage device can be issued without waiting for a checkpoint, and the frequency of access can be reduced by each time. Since the data can be distributed to the disk drives, the performance of the system during normal processing can be improved.
Further, according to the present invention, in a checkpoint-rollback computer system, an I / O request to a secondary storage device can be issued without waiting for a checkpoint, so that the performance of the system during normal processing is reduced. Can be improved.

Further, according to the present invention, in a checkpoint-rollback computer system, the RAI
The performance of the system during normal processing can be improved without impairing the performance and the fault-tolerant function due to D or the like.

[Brief description of the drawings]

FIG. 1 is a diagram showing a configuration of a system according to an embodiment of the present invention.

FIG. 2 is a diagram showing a data management block management table according to the embodiment;

FIG. 3 is a flowchart relating to a reading process according to the embodiment;

FIG. 4 is a diagram showing a data management block management table in a state different from that of FIG. 2 according to the embodiment;

FIG. 5 is a flowchart related to a write process according to the embodiment;

FIG. 6 is a flowchart related to a checkpoint process according to the embodiment;

FIG. 7 is a flowchart regarding a rollback process according to the embodiment;

FIG. 8 is a diagram showing a data management block management table in a different state from that of FIG. 4 according to the embodiment;

FIG. 9 is a diagram showing a data management block management table in a different state from that of FIG. 8 according to the embodiment;

FIG. 10 is a diagram showing a data management block management table in a state different from that of FIG. 9 according to the embodiment;

FIG. 11 is a diagram showing a data management block management table in a different state from FIG. 10 according to the embodiment;

FIG. 12 is a diagram showing a data management block management table in a state different from that of FIG. 11 according to the embodiment;

FIG. 13 is a diagram showing a data management block management table in a different state from that of FIG. 12 according to the embodiment;

[Explanation of symbols]

 10 Computer System 11 Hard Disk 12 Hard Disk 13 Main Memory 14 Management Table

Claims (6)

[Claims] 1. A process is performed while periodically collecting checkpoints. If a failure occurs in the system, rollback is performed to return the state of the system to the point of the checkpoint collected immediately before, and the process is executed again. Checkpoint to avoid temporary system failure
In a computer system having a rollback mechanism and storing and reading data to and from a secondary storage device, a plurality of data blocks are allocated on the secondary storage device for one data, and Means having a management table for recording which of the data blocks is the checkpoint image at the time of the last checkpoint and the latest state; and a request for writing data to the secondary storage device Is received, a data block that is not a checkpoint image is selected from a plurality of data blocks corresponding to the received data on the secondary storage device by referring to the management table, and the selected data is The block is not a checkpoint image and the latest data block Means for recording the information shown in the management table and writing data to the selected data block; and when a request to read data from the secondary storage device occurs, by referring to the management table, Means for reading data from the data block designated as the latest data block in the management table among the plurality of data blocks corresponding on the secondary storage device; By referring to the table, of the plurality of data blocks corresponding to all data on the secondary storage device, the block registered as the latest data block is the latest and is a checkpoint image To register the system in the above management table, When the processing is resumed by returning to the state of the data block, by referring to the management table, all the data are registered as a checkpoint image among a plurality of corresponding data blocks on the secondary storage device. A means for registering in the management table that the block being used is a checkpoint image and is up-to-date in the management table. 2. The process proceeds while periodically collecting checkpoints, and when a failure occurs in the system, the state of the system is returned to the time of the checkpoint collected immediately before by a rollback to change the state of the checkpoint. It has a checkpoint-rollback mechanism that avoids a temporary failure of the system by re-executing the processing, and stores data with the secondary storage device.
In a computer system that performs reading, two data blocks are allocated on a secondary storage device for one data, and which of the two data blocks is the checkpoint image in the state at the time of the last checkpoint, And means having a management table for recording whether the data is in the latest state, and when a request to write data to the secondary storage device is received, by referring to the management table, secondary storage is performed on the received data. A data block that is not a checkpoint image is selected from two corresponding data blocks on the device, and information indicating that the selected data block is not a checkpoint image and that there is a latest data block is stored in the management table. And write the data to the selected data block. Means for reading the data from the secondary storage device, when a request for reading data from the secondary storage device occurs, by referring to the management table, of the two data blocks corresponding to the requested data on the secondary storage device, Means for reading data from the data block designated as the latest data block in the management table; and, at the time of checkpoint processing, by referring to the management table, all data is handled on the secondary storage device. Of the two data blocks, the block registered as the latest data block is the latest,
Means for registering a checkpoint image in the management table; and when the system is returned to the latest checkpoint state and processing is resumed, secondary storage is performed for all data by referring to the management table. Means for registering in the management table that a block registered as a checkpoint image among two corresponding data blocks on the apparatus is a checkpoint image and is up-to-date. An input / output device control device in a checkpoint / rollback type computer, which is a feature of the computer. 3. A process is performed while periodically collecting checkpoints. If a failure occurs in the system, rollback is performed to return the state of the system to the checkpoint collected immediately before, and the process is executed again. In a computer system that has a checkpoint-rollback mechanism that avoids a temporary failure of the system and stores and reads data to and from a secondary storage device, A plurality of data blocks are allocated in one storage medium belonging to the storage device, and which of the plurality of data blocks is the checkpoint image that is the state at the time of the last checkpoint and whether it is the latest state Means for storing a management table for recording the data, and writing of data to the secondary storage device is required. Is received, a data block that is not a checkpoint image is selected from a plurality of data blocks corresponding to a received data on a certain storage medium belonging to a secondary storage device by referring to the management table, The selected data block is not a checkpoint image,
Means for recording information indicating that there is the latest data block in the management table, and writing data in the selected data block; and when a request to read data from the secondary storage device has occurred, the management table Of the plurality of data blocks corresponding to the requested data on a certain storage medium belonging to the secondary storage device, from the data block designated as the latest data block in the management table. Means and, at the time of the checkpoint process, of a plurality of data blocks corresponding to all data on a storage medium belonging to a secondary storage device, a block registered as the latest data block is the latest one. To register in the management table that there is a checkpoint image When the system is returned to the state of the latest checkpoint and processing is resumed, by referring to the management table, all data are output from a plurality of corresponding data blocks on a storage medium belonging to the secondary storage device. Means for registering in the management table that the block registered as a checkpoint image is the latest and is a checkpoint image in the management table. I / O device controller. 4. A process is performed while periodically collecting checkpoints. If a failure occurs in the system, a rollback is performed to return the state of the system to the time of the checkpoint collected immediately before, and the process is executed again. Checkpoint to avoid temporary system failure
In a computer system having a rollback mechanism and storing and reading data with a secondary storage device, a plurality of data blocks are allocated on the secondary storage device for one data, and It has a management table that records which of the data blocks is the checkpoint image at the time of the last checkpoint and the latest state, and accepts a request to write data to the secondary storage device When a data block that is not a checkpoint image is selected from a plurality of data blocks corresponding to the received data on the secondary storage device by referring to the management table, Indicates that it is not a checkpoint image and that there is the latest data block The information is recorded in the management table, the data is written in the selected data block, and when a request to read data from the secondary storage device occurs, the requested data is referred to by referring to the management table. Of the plurality of data blocks corresponding on the secondary storage device, data is read from the data block designated as the latest data block in the management table, and at the time of checkpoint processing, by referring to the management table, Of the plurality of data blocks corresponding to all data on the secondary storage device, the management table indicates that the block registered as the latest data block is the latest and is a checkpoint image. Register, return the system to the latest checkpoint, and restart the process. In doing so, by referring to the management table, a block registered as a checkpoint image among a plurality of data blocks corresponding to all data on the secondary storage device is a checkpoint image, An input / output device control method in a checkpoint / rollback system computer, wherein the latest information is registered in the management table. 5. The process proceeds while periodically collecting checkpoints. If a failure occurs in the system, the process is re-executed by a rollback that returns the state of the system to the point of the checkpoint collected immediately before. By
In a computer system that has a checkpoint-rollback mechanism that avoids a temporary failure of the system and that stores and reads data to and from the secondary storage device, one data is stored on the secondary storage device. Allocating two data blocks, having a management table for recording which of the two data blocks is the checkpoint image at the time of the last checkpoint and which is the latest state, the secondary storage When a request to write data to the device is received, by referring to the management table, a data block that is not a checkpoint image among two data blocks corresponding to the received data on the secondary storage device. Select the selected data block as a checkpoint image And information indicating that there is a newest data block is recorded in the management table, data is written to the selected data block, and when a data read request from the secondary storage device occurs, the management is performed. From the two data blocks corresponding to the requested data on the secondary storage device by referring to the table, the data is read from the data block designated as the latest data block in the management table, and checkpoint processing is performed. At this time, by referring to the management table, the block registered as the latest data block among the two data blocks corresponding to all data on the secondary storage device is the latest,
In addition, when the system is registered as a checkpoint image in the management table, and the system is returned to the latest checkpoint state and processing is resumed, all data are referred to the secondary storage device by referring to the management table. A check characterized in that, among the two data blocks corresponding to the above, the block registered as a checkpoint image is a checkpoint image and the latest is registered in the management table. An input / output device control method in a point / rollback computer. 6. A process is performed while periodically collecting checkpoints. If a failure occurs in the system, rollback is performed to return the state of the system to the point of the checkpoint collected immediately before, and the process is executed again. In a computer system that has a checkpoint-rollback mechanism that avoids a temporary failure of the system and stores and reads data to and from a secondary storage device, A plurality of data blocks are allocated in one storage medium belonging to the storage device, and which of the plurality of data blocks is the checkpoint image at the time of the last checkpoint and whether it is the latest state A management table for recording data, and receives a request to write data to the secondary storage device. When the data block is selected, a data block that is not a checkpoint image is selected from a plurality of data blocks corresponding to a received data on a certain storage medium belonging to a secondary storage device by referring to the management table, Information indicating that the selected data block is not a checkpoint image and is the latest data block is recorded in the management table, data is written in the selected data block, and a data read request from the secondary storage device is issued. Is generated, by referring to the management table, the latest data block is designated in the management table among the plurality of data blocks corresponding to the requested data on a certain storage medium belonging to the secondary storage device. Data from the data block At the time of the point processing, the management unit determines that the latest data block is the latest one and is a checkpoint image among a plurality of corresponding data blocks on a certain storage medium belonging to the secondary storage device for all data. When the data is registered in the table and the system is returned to the state of the latest checkpoint and processing is restarted, referring to the management table, all data corresponding to a plurality of data on a certain storage medium belonging to the secondary storage device are read. Input / output device control in a checkpoint / rollback system computer, wherein the block which is the checkpoint image among the blocks is the latest and is registered in the management table as a checkpoint image. Method.