JP3120039B2 - Input / output control method with check recovery function and input / output control device with check recovery function - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an input / output control method suitable for a computer system having a checkpoint restart function, and more particularly, to all input / output control methods when a secondary storage device is multiplexed. An input / output control method with a check recovery function that dramatically improves the input / output processing capacity of the computer system, eliminating the need to wait for the output process to wait until the next checkpoint is obtained, and a check recovery function It relates to an input / output control device.

[0002]

2. Description of the Related Art In recent years, the spread of computer systems has been remarkable, and higher reliability has been required. And
A checkpoint restart mechanism exists to improve the reliability of the computer system.

Conventionally, a checkpoint is periodically taken to perform processing, and when a failure occurs in the system, the checkpoint is restarted from the collected checkpoint to recover from the failure. In the computer system having the mechanism, the memory state after the recovery and the re-execution state of the process and the I / O
In order to maintain consistency with the state of the O device (for example, the contents of the disk), it is necessary to execute only a determined I / O request that is always issued even after recovery. Therefore, even if an I / O request occurs, its execution (processing) is normally suspended until the next checkpoint processing is started, and is executed collectively after the checkpoint processing is completed.

[0004]

In the conventional input / output control method as described above, the I / O request issuance processing is performed without any trace (RE) when the memory state is restored.
AD) aside, at least write (WRI
TE) is delayed from the original issue time to the checkpoint collection time.

On the other hand, in such a fault-tolerant computer system, multiplexing is usually performed on disks in order to improve reliability.
In this case, in order to maintain data reliability, it is necessary to issue I / O requests to each of the multiplexed disks sequentially instead of in parallel. The operation principle at this time will be described with reference to FIG.

Here, in file management, it is assumed that each file is redundantly managed by a master system and a slave system.
These files may be placed on different disks and multiplexed, or may be placed on different logical partitions on the same disk and multiplexed.

For example, if a write request is issued from an application program ((1) in FIG. 8), the file management system first issues an instruction to a device driver to write the request into a main file (FIG. 8). 8 (2)).

When the completion of the writing is notified from the device driver ((3) in FIG. 8), the file management system subsequently issues an instruction to the device driver to write to the slave file ((3) in FIG. 8). 4)).

When the completion of writing to the slave file is notified ((5) in FIG. 8), the file management system notifies the application program of the completion of writing ((5) in FIG. 8). 6)).

Therefore, a write request in a system in which the above-mentioned checkpoint mechanism is applied and a file is duplicated is first issued until a checkpoint is issued. Unless the processing is completed, one theoretical I / O request is not completed, so that there is a problem that the processing time is long as a whole. The operation principle at this time will be described with reference to FIG.

For example, if a write request is issued from an application program ((1) in FIG. 9), the file management system waits for the issuance of this request until the next checkpoint is obtained (FIG. 9). (2)). When the checkpoint is acquired ((3) in FIG. 9), the file management system issues an instruction to the device driver to write the waiting request into the main file ((4) in FIG. 9). )).

The subsequent operation is the same as the operation shown in FIG.
When the completion of the writing is notified from the device driver ((5) in FIG. 9), the file management system subsequently issues an instruction to the device driver to write the slave file ((6) in FIG. 9). )), When the completion of writing to the slave file is notified ((7) in FIG. 9), the completion of writing is notified to the application program ((8) in FIG. 9).

That is, for a write request to the secondary storage device, an extra processing time is consumed only for the period shown in FIG. 9 (2), and the response time of the entire system is increased. There was a problem.

Incidentally, Japanese Patent Application No. 5-305077 (corresponding U
SSN 992219, U.S. filing date: December 1992
On May 17, Applicant (IBM) discloses a system and method for remote data duplication. This system is a duplex system of DSAD (Direct Access Storage Device) , the secondary side is remote, the input data is directly written on the primary side disk, and the secondary side is grouped by sequence checkpoint. Written. However, the checkpoint described in this system means a sequence checkpoint in which a series of information packets are grouped and processed as one sequence unit, and is used for recovering the system described in the present invention . The meaning is different from the checkpoint.
In other words, while the IBM system is a checkpoint for data duplication (to match the sequence of packets sent asynchronously), the present invention provides a system for checking data in a checkpoint recovery type computer system. This is a checkpoint for consistency. Since the IBM system is not a checkpoint restart system, the checkpoint time interval is longer than that of the present invention .

Therefore, the present invention has been made in consideration of the above circumstances, and solves the above-mentioned problems, and obtains the next checkpoint for all input / output processing when a secondary storage device is multiplexed. It is an object of the present invention to provide an input / output control method having a check recovery function that makes it unnecessary to wait until the input / output processing is completed and dramatically improves the input / output processing capacity of a computer system.

[0016]

According to the present invention, in order to achieve the above object, a checkpoint for restarting the interrupted process is periodically acquired in preparation for a case where the process is interrupted due to a failure or the like. An input / output control method applied to a computer system having a checkpoint acquisition function for multiplexing and managing one data in a plurality of storage locations of a secondary storage device.
When a data write request to the secondary storage device is received, the data write request is issued to a first storage location among a plurality of storage locations of the secondary storage device without waiting for the next checkpoint. A first step of performing a data write process and a second check point of the plurality of storage locations of the secondary storage device after acquiring the next checkpoint.
A second step of executing the data write processing on the storage location of the above, and a restart of the write processing in the first step when restarting from the checkpoint acquired immediately before.
The data in the first storage location where the
To read and repair the data stored in the memory location
And a third step .

According to the present invention, when a data write request is received, for example, of files duplicated between a master system and a slave system, the main system file does not wait for the next check point. Perform write processing. On the other hand, for the slave file, the writing process is performed after the writing process of the master file is completed and the next checkpoint is obtained.

Generally, the checkpoint acquisition interval is
In most cases, the next checkpoint is acquired during the writing process to the main file because the time required for the writing process to the main file is sufficiently small. Therefore, there is almost no waiting period when writing to the slave file, and the processing time is greatly reduced.

Here, it is assumed that the operation is restarted from the immediately preceding checkpoint due to a failure or the like. in this case,
In the present invention, since data written to the master file is restored with data of the slave file, no inconsistency occurs. As a result, it is possible to avoid waiting for all the input / output processing when the secondary storage device is multiplexed until the next checkpoint is obtained, and the processing capacity is dramatically improved.

Further, the present invention achieves the above object,
A computer system having a checkpoint acquisition function for periodically acquiring a checkpoint for restarting the interrupted processing in preparation for a case where the processing is interrupted due to a failure or the like. In an input / output control method applied to a computer system that manages data by configuring a parity group in a storage device, when a request to write data to the secondary storage device is received, the data written after the request is updated from the write data of the request A first step of calculating parity, and issuing a data write request to the first secondary storage device among the secondary storage devices without waiting for the next checkpoint, and executing the write data write process. In the second step and after the acquisition of the next checkpoint, a write request for the parity calculated in the first step is issued. A third step of executing the parity write processing on the secondary storage device of No. 2 and, when restarting from the checkpoint acquired immediately before, the second step stores the parity in the first secondary storage device. The written data is compared with the parity calculated in the first step and the previous data.
Before writing from the write data in the second step
And then rewrite the calculated data
And restoring by a fourth step .

According to the present invention, when a data write request is accepted, the write processing is performed without waiting for the next checkpoint. As a result, the parity to be updated is updated when the write processing is completed. After the checkpoint is obtained, the writing process is performed.

Also in this case, the checkpoint acquisition interval is sufficiently small as compared with the time required for the writing process to the file, and in most cases, the next checkpoint is executed during the writing process to the file. Is obtained. Therefore, when performing parity write processing, there is almost no waiting period,
The processing time is greatly reduced.

Here, consider the case where the operation is restarted from the immediately preceding checkpoint due to the occurrence of a failure or the like. in this case,
According to the present invention, the data written to the file is restored by the parity, so that no inconsistency occurs. As a result, it is possible to prevent the input / output processing from waiting until the next checkpoint is obtained, and the processing capacity is dramatically improved.

Further, in order to achieve the above object, the present invention provides a checkpoint acquisition function for periodically acquiring a checkpoint for restarting the interrupted process in case the process is interrupted due to a failure or the like. A data write request to the secondary storage device, wherein the input / output control method is applied to a computer system that manages data by configuring a parity group with a plurality of secondary storage devices. Receiving the request, calculates the updated parity from the write data of the request.
And a second step of issuing the parity write request calculated in the first step without waiting for the next checkpoint, and executing the parity write processing in the second secondary storage device. And a third step of issuing a data write request to a first secondary storage device among the secondary storage devices after the acquisition of the next checkpoint and executing a write process of the write data;
When restarting from the checkpoint obtained just before,
The parity written in the second secondary storage device in the second step is calculated using the parity calculated in the first step.
From the write data of the third step
Of parity and rewrite the restored parity
And repairing the data by a fourth step .

According to the present invention, when a data write request is received, a write process is performed on the parity to be updated as a result of the data write without waiting for the next checkpoint, and the parity write is performed on the data. After the processing is completed and the next checkpoint is obtained, the writing processing is performed.

In this case as well, the checkpoint acquisition interval is sufficiently small compared to the time required for the parity write processing. Therefore, in most cases, the next checkpoint is acquired during the parity write processing. You. Therefore, when performing the data writing process,
There is almost no waiting period, and the processing time is greatly reduced.

Here, consider a case where the operation is restarted from the immediately preceding check point due to a failure or the like. in this case,
According to the present invention, since the parity is restored by the data of the file, no inconsistency occurs. As a result, it is possible to prevent the input / output processing from waiting until the next checkpoint is obtained, and the processing capacity is dramatically improved.

According to the present invention, in a computer system having a checkpoint restart function, for example, multiplexing and managing one piece of data at a plurality of locations to secure data reliability, O request delay can be reduced, and system performance during normal operation can be dramatically improved.

[0029]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. (First Embodiment) FIG. 1 is a block diagram showing an example of a computer system according to a first embodiment of the present invention.

As shown in FIG. 1, the computer system has a CP
U31, memory 37, disk drives 43 and 45, disk controllers 39 and 41 for controlling these disk drives, and a system bus 35 for connecting each of the above units.

FIG. 3A, FIG.
(B), FIG. 5 (a), FIG. 5 (b), FIG. 6 (a), FIG.
The program shown in the flowchart shown in (b) is stored. Although the number of CPUs is one in FIG. 1,
There may be more than one.

FIG. 2 is a functional block diagram showing a schematic configuration of the computer system in the embodiment. The computer system according to the present embodiment is entirely controlled by an operating system 2. The operating system 2 includes a file management system 21 that manages files stored in disks 3a to 3b and the like, and disks 3a to 3b. A device driver 22 for input / output of
And a checkpoint management system 23 for managing acquisition of checkpoints.

In the computer system of this embodiment, when the application program 1 issues a data write request, the write request is sent to the file management system 21.
Handed over to The file management system 21 that has received the write request determines where the data is to be stored and multiplexed, and associates the determined storage location. Here, this data is stored on disk 3a.
It is assumed that the data is multiplexed by being stored in the (master system) and the disk 3b (slave system).

Next, the file management system 21 issues a data write request to the device driver 22 only for the main disk 3a. The device driver 22 receiving this instruction executes the writing process on the disk 3a.

In most cases, a checkpoint is acquired during this execution, and the checkpoint management system 23 notifies the file management system 21 of that.
When the writing process to the disk 3a is completed, the device driver 22 notifies the file management system 21 of the completion, and the file management system 21, which has already received the checkpoint acquisition notification from the checkpoint management system 23, Immediately subordinate
The device driver 22 is instructed to write data to the disk 3b . Then, when the file management system 21 completes the slave write processing,
It notifies the application program 1 of the completion of the writing process.

On the other hand, when restarting from the immediately preceding checkpoint due to the occurrence of a failure or the like, the file management system 21 instructs the device driver 22 to read data from the disk 3b. The data on the main disk 3a is restored.

Next, the operation procedure of this embodiment is shown in FIG.
This will be described with reference to FIG. FIG. 3 (a)
It is a flowchart which shows the operation | movement procedure at the time of normal.

The device file management system 21 which has received the write request from the application program 1 (Step A1), instructs the device driver 22 only for disc 3a is a write request for the data in the main system, receives this instruction The driver 22 executes this writing process on the disk 3a (step A2).

When the writing process is completed (Y in step A3), the file management system 21 determines whether or not a checkpoint has been acquired (step A4), and if the checkpoint has been acquired (Y in step A4). ), And instructs the device driver 22 to write data to the slave disk 3b (step A).
5). When the writing process is completed (Y in step A6), the application program 1 is notified of the completion of the writing process (step A7).

FIG. 3B is a flowchart showing an operation procedure when restarting from the immediately preceding check point due to a failure or the like. In this case, the file management system 21 first instructs the device driver 22 to read data from the disk 3b (step B1). When the data reading process of the slave disk 3b is completed (Y in step B2), the read data is rewritten to the master disk 3a (step B3 to step B4). As a result, the data on the main disk 3a is restored.

FIG. 4 shows the operation principle of this embodiment. Now, for example, if a write request is issued from an application program ((1) in FIG. 4), the file management system does not wait for the acquisition of the next checkpoint but sends the device driver 22 to the device driver 22 to write to the primary file. An instruction is issued ((2) in FIG. 4).

Since the checkpoint acquisition interval is sufficiently smaller than the time required for the writing process to the main file, in most cases, the next checkpoint is executed during the writing process to the main file. It is acquired ((3) in FIG. 4). Then, (in FIG. 4 (4)) when completing the device driver of the main system of writing <br/> lump can is notified, the file management system, subsequently instruct the device driver to write to the slave file ((5) in FIG. 4), and when the completion of writing to the slave file is notified ((6) in FIG. 4), the completion of writing is notified to the application program (FIG. 4). (7)).

As a result, it is possible to prevent the input / output processing from waiting until the next checkpoint is obtained, and to restore the data written to the master file with the data of the slave file. No alignment occurs.

(Second Embodiment) Next, a second embodiment of the present invention will be described with reference to FIGS. 5A and 5B.

The configuration of this embodiment is the same as that of the first embodiment. However, in the first embodiment, the reliability is realized by providing a master system and a slave system and multiplexing them, whereas in the present embodiment, for example, FIG.
As shown in (a), RAID (Redundant A
rays of Inexpensive Disk
s) level 3 (input data is divided in bit units and stored in a plurality of HDDs. Parity is used for error correction codes) or level 5 (input data is divided into HDD sectors as shown in FIG. 7B). Interleaved in units, distributed to individual HDDs).
This is realized by configuring a parity group from a to 3b.

The operation procedure of this embodiment according to this configuration will be described with reference to FIG. FIG. 5A is a flowchart illustrating an operation procedure in a normal state. The file management system 21 that has received the write request from the application program 1 (step C1) calculates the parity to be updated as a result of writing this data (step C2). Then, the file management system 21
The device driver 22 instructs the device driver 22 to write the data, and the device driver 22 having received the instruction executes the write process (step C3).

When the writing process is completed (Y in step C4), the file management system 21 determines whether or not a checkpoint has been acquired (step C5), and if the checkpoint has been acquired (Y in step C5). This time, a parity write request is instructed to the device driver 22 (step C6). When the writing process is completed (Y in step C7), the application program 1 is notified of the completion of the writing process (step C8).

FIG. 5B is a flowchart showing an operation procedure when restarting from the immediately preceding check point due to a failure or the like. In this case, the file management system 21 first checks the parity calculated in step C2 and C
The data before writing is calculated from the data to be written in step 3 (step D1). And this calculated data is
The data is restored by rewriting the data written in step C3 (step D2 to step D3).

As a result, similarly to the first embodiment, it is possible to prevent the input / output processing from waiting until the next checkpoint is obtained, and to restore the data with the parity, thereby causing no inconsistency. There is no.

Note that the same effect can be obtained by reversing the data writing and the parity writing. That is, as shown in FIG. 6A, when a procedure of performing data write processing after completion of parity write processing (assuming that the next checkpoint has been acquired) is performed, As shown in b), when restarting from the immediately preceding checkpoint, the original parity is restored from the parity calculated in step E2 and the data to be written in step E6 (step F).
1) By replacing the parity written in step E4 with the restored parity (step F2 to
Step F3), restoration is possible.

In the above-described embodiment, for example, the write timing of the master and slave systems and the write timing of data and parity are controlled by the file management system. However, these can be controlled by the device driver. It is.

[0052]

As described in detail above, according to the present invention, 2
This eliminates the need to wait for the next checkpoint to be obtained for all I / O processing when the next storage device is multiplexed, and can greatly improve the I / O processing capacity of the computer system. It works.

[Brief description of the drawings]

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

FIG. 2 is a functional block diagram showing a schematic configuration of a computer system according to the embodiment;

FIG. 3 is an exemplary flowchart for explaining the operation procedure of the computer system according to the embodiment;

FIG. 4 is an exemplary view for explaining the operation principle of the computer system according to the embodiment;

FIG. 5 is a flowchart illustrating an operation procedure of a computer system according to the second embodiment.

FIG. 6 is an exemplary flowchart for explaining the operation procedure of the computer system according to the embodiment;

FIG. 7 is a diagram illustrating a RAID level 3 according to the embodiment;
FIG. 9 is a conceptual diagram for explaining level 5;

FIG. 8 is a diagram illustrating an operation at the time of I / O issuance in a conventional computer system.

FIG. 9 is a diagram illustrating an operation at the time of I / O issuance in a conventional computer system.

[Explanation of symbols]

31: CPU, 35: System bus, 37: Memory, 3
9 disk controller, 41 disk controller, 43 disk drive, 45 disk drive.

Claims (6)

(57) [Claims] 1. A computer system having a checkpoint acquisition function for periodically acquiring a checkpoint for restarting an interrupted process in case a process is interrupted due to a failure or the like. An input / output control method applied to a computer system for multiplexing and managing one data in a plurality of storage locations of a secondary storage device, wherein when a data write request to the secondary storage device is received, A first step of issuing the data write request to the first storage location among the plurality of storage locations of the secondary storage device without waiting for the next checkpoint, and executing a data write process; After the acquisition of the checkpoint, a second step of executing the data write processing to the second storage location among the plurality of storage locations of the secondary storage device is performed. When restarting from the checkpoint obtained immediately before
The write operation performed in the first step
The data of the first storage location is stored in the second storage location.
And a third step of reading and restoring the read data, and an input / output control method having a check recovery function. 2. A computer system having a checkpoint acquisition function for periodically acquiring a checkpoint for restarting an interrupted process in case a process is interrupted due to a failure or the like. An input / output control method applied to a computer system that manages data by configuring a parity group with a plurality of secondary storage devices, comprising: when receiving a data write request to the secondary storage device, writing the request; A first step of calculating an updated parity from data; and issuing a data write request to the first secondary storage device among the secondary storage devices without waiting for the next checkpoint, and A second step of executing a write process; and writing of the parity calculated in the first step after acquiring the next checkpoint. A third step of executing the write process of the parity to the second of the secondary storage device by issuing a request, when re-starting from the checkpoint which has been acquired immediately before,
The data written in the first secondary storage device in the second step is calculated using the parity calculated in the first step.
From the write data of the second step.
Calculate the data before embedding, and rewrite the calculated data
And a fourth step of writing and restoring . An input / output control method having a check recovery function. 3. A computer system having a checkpoint acquisition function for periodically acquiring a checkpoint for restarting an interrupted process in case the process is interrupted due to a failure or the like. In an input / output control method applied to a computer system that manages data by configuring a parity group with a plurality of secondary storage devices, when a request to write data to the secondary storage device is received, write data of the request is written. A first step of calculating an updated parity from the first and a write request of the parity calculated in the first step without waiting for a next check point, and issuing the parity write request to a second secondary storage device. A second step of executing a write process of the first storage device of the secondary storage device after obtaining a next checkpoint. Third executing the writing process of the write data by issuing a data write request to location
Steps and restarting from the last checkpoint taken,
The parity written in the second secondary storage device in the second step is calculated using the parity calculated in the first step.
From the write data of the third step
Of parity and rewrite the restored parity
And an input / output control method having a check recovery function. 4. A computer system having a checkpoint acquisition function for periodically acquiring a checkpoint for restarting an interrupted process in case a process is interrupted due to a failure or the like. An input / output control device applied to a computer system for multiplexing and managing one data in a plurality of storage locations of a secondary storage device, wherein when a data write request to the secondary storage device is received, First means for issuing the data write request to the first storage location among the plurality of storage locations of the secondary storage device without waiting for the next checkpoint, and executing data writing processing; Second means for executing the data write processing to a second storage location of the plurality of storage locations in the secondary storage device after acquiring the checkpoint; When restarting from the checkpoint obtained just before,
The first storage device in which write processing has been performed by the first means.
Memory data is stored in the second storage location
An input / output control device having a check recovery function, comprising: third means for reading and restoring data . 5. A computer system having a checkpoint acquisition function for periodically acquiring a checkpoint for restarting an interrupted process in case a process is interrupted due to a failure or the like. In an input / output control device applied to a computer system that manages data by configuring a parity group with a plurality of secondary storage devices, when a request to write data to the secondary storage device is received, the request is written. A first means for calculating an updated parity from data; and a second means for issuing a data write request to the first secondary storage device without waiting for a next checkpoint, and executing a write process of the write data. The second means, and after the acquisition of the next checkpoint, issues the parity write request calculated by the first means and issues the second
Third means for executing the parity write processing to the next storage device, and when restarting from the checkpoint acquired immediately before,
The data written to the first secondary storage device by the second means is calculated by comparing the parity calculated by the first means with the parity calculated by the first means.
The data before writing from the write data of the second means.
Data, and rewrite the calculated data to
And an input / output control device having a check recovery function. 6. A computer system having a checkpoint acquisition function for periodically acquiring a checkpoint for restarting an interrupted process in case a process is interrupted due to a failure or the like. An input / output control device applied to a computer system that manages data by configuring a parity group with a plurality of secondary storage devices, wherein when a data write request to the secondary storage device is received, the write data of the request is written. A first means for calculating an updated parity from the first means, and a write request for the parity calculated by the first means is issued without waiting for a next checkpoint, and the parity is written to a second secondary storage device. Issuing a data write request to the first secondary storage device after acquiring the next checkpoint Said third means for performing write processing of the write data, when re-starting from the checkpoint which has been acquired immediately before Te,
The parity written in the second secondary storage device by the second means is compared with the parity calculated by the first means.
The original parity is obtained from the write data of the third means.
Restore and rewrite and restore this restored parity
And an input / output control device having a check recovery function.