JP3260433B2 - Failure recovery computer system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a failure recovery type computer system which employs a rollback method in which, when an abnormality occurs in information processing by a computer, a retry is automatically performed from a past checkpoint.

[0002]

2. Description of the Related Art Conventionally, as a computer system having a rollback point, for example, there is one disclosed in Japanese Patent Application Laid-Open No. 3-265951.

FIG. 8 explains the rollback method disclosed in the above publication, and shows the concept of generation update. Rollback data is stored in an auxiliary storage area (file) so as not to be destroyed. , Generation management is performed.

As shown in FIG. 8 (a), a rollback point update process (steps 81 and 83) and an application program execution process (steps 82 and 84) are performed in chronological order. If an error occurs during the execution process (steps 82 and 84), the process returns to the immediately preceding rollback point (steps 81 and 83) and the rollback data execution process (steps 82 and 84)
To recover from the abnormal state.

[0005] Here, the update of the rollback point means that the generation of the rollback data is changed at the rollback point, for example, as shown in FIG.

When a rollback point is detected in the course of executing the application program, as shown by the solid arrow in the figure, the new generation (85, 88) which is the latest data at that time.
Is transferred to the file as rollback data of the previous generation (86, 89), and the program proceeds to the next rollback point. If no abnormality occurs, an update process is performed to make the generation one generation old for each rollback point by the above procedure.

In this manner, a plurality of generations such as two generations (new generation 85, previous generation 86) and three generations (new generation 88, previous generation 89, two generations before 90) are performed, and rollback data ( New generations 86 and 89, two generations before last 90, etc.) are formed.

On the other hand, when data of the new generation (85, 88) is executed, the subsequent data is stored in the current generation (87, 9).
However, if an error occurs, the process returns to the immediately preceding rollback point and performs the abnormal time restoration process. That is, in the case of the two-generation management, as shown by the dashed arrow in the figure, the previous generation 86 of the file is transferred to the main storage as the new generation 85, and the copied data of the new generation 85 is executed.

In the case of three-generation management, the last two generations 90 on the file are set as the new previous generation 89, and the old previous generation 89 is created.
Is transferred to the main memory to obtain a new new generation 88. in short,
The process of returning the rollback data to one generation is performed.

[0010] As one of the generation updating methods shown in FIG. 8, the above publication discloses an MMU (Memory Ma).
3 shows a generation update method using Pping Unit (pping Unit).

FIG. 9 shows a generation update method using the MMU, and shows an update process for one logical page number "x".

In FIG. 9, as a content of the logical page number “x” of the MMU conversion map immediately before updating, a flag is set in the write flag area of the physical page number “a” in the main storage unit, and the content is changed to the stored data. It is shown that there was.

Therefore, the physical page number “a” of n generations (old and current generations) at the time of updating is set to the (n + 1) th generation (new and current generations), and the contents of the physical page number “a” are used as rollback data as file pages. The data is transcribed to the area of the number “u”, and the file page number “u” is set as the nth generation (new previous generation). Note that the flag is reset at an appropriate time after the transcription.

In the virtual storage system, basically, the storage contents of the entire software are stored in an external storage device such as a disk in correspondence with a virtual address, and a part of the main storage is copied from the external storage device as real storage. Is basically loaded (Roll-In) and returned to the external storage device when unnecessary (Roll-Out). The MMU is a conversion table mechanism that indicates where the information indicated by the address of the virtual storage is located in the external storage device or the main storage unit.

In the generation management based on the rollback point, an appropriate checkpoint is set in the process of processing in preparation for the occurrence of an abnormality, and when an abnormality occurs, the system is returned to there once, and the processing is restarted from there. R
All-Back-Point. Conventional M
The management method by the MU allows the MMU to manage multiple generations, and if an error occurs, performs an error process such as hardware switching, returns the contents of the MMU to one generation, and automatically resumes from the previous rollback point I was able to do it.

[0016]

However, in the above-described conventional MMU-based management method, the external storage device also has a redundant configuration, and in the case of three-generation management, three generations × 2 units for one virtual page = Since an external storage device for six pages is required and a storage capacity of six times is required, improvement thereof has been desired.

The present invention has been made in view of the above points, and can reduce the storage capacity of an external storage device in a failure recovery type computer system of a rollback point system in which an MMU has generation management. The purpose is to obtain a failure recovery type computer system.

[0018]

In order to achieve the above object, a fault recovery type computer system according to the first aspect of the present invention uses a rollback point system in which an MMU (memory management unit) has generation management. Oite the fault recovery type computer system, virtual memory addresses and real memory addresses
MMU conversion method that manages conversion to and from multiple generations
When realizing generation management by using
Determine the external storage device to be used, and
At the time of out, the rollback information is used for future generations.
The data is also written to the external storage device at the same time .

A fault recovery type according to claim 2 of the present invention.
The computer system has n (n> 1: integer) external storage devices.
With a memory management unit (MMU)
Recovery of Rollback Point Method with Cost Management
Type computer system, at the rollback point
Switching the external storage device at generation update
Different external storage devices are used for different generations.
The difference from the previous generation is calculated for each page of the external storage device.
It is characterized in that it is stored as the presence or absence of a change between the two.

[0020]

[0021]

[0022]

[0023]

Further, fault recovery type computer system according to claim 3, in claim 1, wherein the fault recovery type computer system, each intergenerational content of 3 n-1 in generation of MMU generation management, n-2 Generation If they are the same, the page content of the n-2 generation is rolled in from the side of the n-2 generation external storage device.

Further, fault recovery type computer system according to claim 4 is the fault recovery type computer system according to claim 1, when the MMU generation management, each page of the external storage device
The change status between two generations before and after is stored.

Further, the fault recovery type computer system according to claim 4 is the fault recovery type computer system according to claim 2 and the fault recovery type computer system according to claim 2.
The rollback information at the time of rollout in n generations
To another external storage device for future generations at the same time.
It is characterized in that

Further, the fault recovery type computer system according to claim 5 is the fault recovery type computer system according to claim 1 or 2, wherein the MMU hardware includes only a conversion mechanism between a current generation virtual address and a real address. The MMU mapping information of the old generation is stored in an external storage device, and is used for rollback processing when an error occurs.
It is characterized by being loaded on hardware.

[0028]

According to the present invention, the following operation is achieved by adopting the above configuration. 1) According to the first aspect, the required capacity of the external storage device is reduced. 2) According to claim 2, continuity of processing is maintained even if disks used in the n generation and the n + 1 generation are different. 3) According to claim 3, the amount of MMU management information is reduced. 4) According to claim 4, the amount of hardware of the MMU is reduced. 5) According to claim 5, the amount of transfer to the external storage device related to the rollback point process is reduced. 6) According to the sixth aspect, the operation is the same as that of the fifth aspect. 7) According to the seventh aspect, the load on the external storage device can be made uniform and the reliability can be improved. 8) According to claim 8, the processing at the rollback point is facilitated. 9) According to the ninth aspect, the operation is the same as that of the eighth aspect. 10) According to claim 10, MMU hardware is reduced.

[0029]

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

According to the present invention, in order to reduce the storage capacity of an external storage device, each generation is distributed and accumulated on a plurality of disks instead of placing all of the plurality of generations on a working system of a conventional external storage device. FIG. 1 shows an embodiment in which the capacity is reduced and two-generation management is performed.

As shown in FIG. 1, the disk specified by the MMU is switched for each rollback point, and the disk to be used is determined for each generation. Using a generational disk, the process can be pulled back (rolled back) and restarted.

FIG. 2 shows the hardware configuration diagram. That is, this embodiment shows an application example of a computer system having a redundant configuration.
1, an MMU hardware 2, a main storage unit 3, an external storage device control unit 4, and a main / redundancy management unit 5.
The virtual address specified by the PU 1 is converted into a real address by the MMU hardware 2, and the
1 is read and written.

In this case, since the virtual storage method is used,
If the virtual address specified by the EPU 1 does not exist in the main storage unit 3, a page fault occurs, and the contents of the unnecessary main storage unit 3 are temporarily evacuated to the disk (roll-out) as in the normal virtual storage system, and the specified address is specified. The content of the virtual address is loaded from the disk (roll-in).

The MMU hardware 2 shown in FIG. 2 may be a current-generation virtual / real conversion mechanism only. The contents of the old-generation MMU are saved in the disk, and the MMU hardware is first used at the time of rollback processing when an abnormality occurs. M as a method to load to 2
The MU hardware 2 can be reduced.

In the processing of the rollback point after the normal processing of each generation is completed, the contents of the page changed in the generation are saved to the disk used in the generation, and the software register at the rollback point is rewritten. It is necessary to write all the contents, work contents, status information, and the contents state of the program and the like on the disk. At the same time, the next generation process is assigned to a different next disk. Therefore, at the time of generation update, it is necessary to ensure that the internal state used for the nth generation is carried over to the next disk. is there.

In order to reduce the inheritance of the internal state,
By taking advantage of the fact that the program part does not usually change between generations, and that the change is only a part between generations, the current generation and the past generation are compared on a page-by-page basis, and changes are made. FIG. 3 shows a rollback point processing method in which only a certain page is transferred between disks or from the main storage unit to the next generation disk.

In FIG. 3, the n generation is the generation that has been processed so far, and the n + 1 generation is the next generation after this rollback point.

In the MMU conversion map shown in FIG. 3, the presence / absence of change indicates whether or not the n generations have been written and the contents of the page are different from the (n-1) th generation. When the write flag of each page in the main storage unit in FIG. 9 is rolled out (including the rollout at the rollback point), and the writing is performed, the “change flag” of this MMU conversion map is changed. The presence indicates that the page is different between the n generation and the n-1 generation.

The MMU conversion map indicates that only the working M
The MMU information on the backup side is saved and stored in the MMU escape area shown in FIG. 2 while being generation-managed.

In the process of the rollback point in FIG. 3, first, the real address is the content in the main storage unit from the MMU working map used in the nth generation and the write “YES” in FIG.
Is rolled out to the disk # 1, and the “change presence / absence” flag of the corresponding MMU virtual address is changed to “present”. After all the evacuation processing of the main storage unit is completed, the current MMU map is also saved in the disk # 1, and the saving to the disk # 1 in the internal state is completed.

Next, in preparation for starting the n + 1 generation,
In the example of FIG. 3, the disk # 1 is used for the n generation, and the disk # 2 is used for the n-1 generation and is used as a backup for the n generation. Conversely, in the next n + 1 generation, disk # 2 is used and disk # 1 is backed up. At this time, disk # 2 needs to take over the final internal state of n generations and proceed with processing.

By the way, in this embodiment, the disk # 1,
Since # 2 is alternately used / reserved, pages of n generations that have no change can be used as they are at the start of the (n + 1) generation for the (n-1) generation. For example, in FIG. 3, if there is no change in the page content of the address Y of the disk # 1 used in the nth generation, the page content of the address x of the disk # 2 of the (n-1) th generation is the same as this, and The n + 1 generation may use the page at address x of disk # 2, and if the MMU map is updated, disk # 2
There is no need to transfer between # 1 and # 2.

Therefore, in the example shown in FIG. 3, only the page portion of the "changed" flag of the "changed" flag of the MMU conversion map need be transferred from the disk # 1 or the main storage unit to the disk # 2. The transfer amount can be reduced.

In the embodiment of FIG. 3 described above, the content of the page written in that generation at the time of processing the rollback point is transferred to the next generation disk.
In the embodiment shown in the figure, instead of transferring all the changed parts at the time of rollback point processing, at the time of roll-in, the contents are first identified and loaded from there, and the data transfer at the time of rollback point processing is performed. It is intended to reduce the amount and increase the speed.

For this reason, in the embodiment shown in FIG. 4, the MMU conversion map is extended so that the n generations can identify which disk has the contents, so that the disk of the n-1 generation and the disk address as the address at the beginning, for example, The # 2 address U is designated, and when the roll-in is performed, the disk is loaded from the U of the disk # 2 of the (n-1) th generation. The data is stored at the address U of the disk # 1.

Note that, in FIG. 4, 〜 indicates the flow of time.

Represents the contents of the initial MMU map in n generations, represents the state immediately after roll-in, and represents the state in which data has been written into the memory and which reflects the once changed "Yes" in the roll-out processing. , Indicate the state after the rollout process.

The embodiment shown in FIG. 5 is an embodiment in which the transfer between disks is unnecessary as in FIG. 4 in the processing at the rollback point in FIG. It will be. That is, when the initial MMU contents in the n generation in FIG. 5 are different from the n-1 generation, the contents have already been written to w of the disk # 1 which is the disk in the n generation. The data is read from w of the disk # 1 into the main storage unit.

At the time of roll-out, the data is written to U of disk # 1 for n generations and
It is also written in x of disk # 2 for the n + 1 generation.

The contents of the MMU for the (n + 1) th generation are initially undecided, or a disk of the (n + 1) th generation (V of disk # 2 in this case) is specified as unchanged.

As a result, at the beginning of each generation, the MMU data is always prepared in the disk used by the MMU in that generation, thereby improving efficiency.

Next, an embodiment of three-generation management is shown in FIG.

In this embodiment, the MMU is provided with difference information between the n generation and the n-1 generation, and between the n-1 generation and the n-2 generation.

If there is no change (difference) between the n generation and the n-2 generation, instead of loading from the n-1 generation in the embodiment of FIG. 4, the n-2 generation as shown in FIG. (In this example, disk # 1), so that the backup disk is not used as much as possible to improve the recoverability at the time of disk abnormality.

In FIG. 7, the difference between the generations is n
The difference from the generation is held in the MMU to facilitate the processing at the time of roll-in.

[0056]

As described above, according to the present invention, the required capacity of the external storage device is reduced by reducing the redundancy of the escape information of each generation to the external storage device in the conventional rollback point system. It has the effect of being able to do so.

Further, by reducing the amount of writing / reading to / from the external storage device, it is possible to increase the speed of the system and improve the responsiveness.

[Brief description of the drawings]

FIG. 1 is a basic conceptual diagram of a rollback method according to the present invention.

FIG. 2 is a configuration diagram according to an embodiment of hardware of the present invention.

FIG. 3 is a conceptual diagram of a rollback point process according to the present invention.

FIG. 4 is an explanatory view showing a case where a rollback point process is reduced by loading from an (n-1) th generation side disk when rolling in for the first time according to the present invention;

FIG. 5 is an explanatory diagram of the present invention in which the write rollback point process is also reduced on the (n + 1) th generation side disk during rollout.

FIG. 6 is an explanatory diagram of an embodiment of three-generation management.

FIG. 7 is a diagram illustrating a first roll-in in three generation management.
FIG. 3 is an explanatory diagram showing an embodiment in which loading is performed from the -2 generation side.

FIG. 8 is an explanatory diagram of a conventional rollback method realized by an application program.

FIG. 9 is an explanatory diagram of a conventional rollback point update process.

[Explanation of symbols]

 1a, 1b EPU 2a, 2b MMU hardware 3a, 3b Main storage unit 4a, 4b External storage device control unit

Claims (6)

(57) [Claims] 1. An MMU (Memory Management Unit)
In a failure-recovery computer system using a rollback point system with generation management, the generation management by the MMU conversion system that manages the conversion between virtual storage addresses and real storage addresses by dividing them into multiple generations. The external storage device to be used is determined in advance for each roll, and when rolling out in n generations,
Data to other external storage devices for future generations.
A failure recovery type computer system characterized by writing at times . (2)n (n> 1: integer) external storage devices
Generation management in MMU (memory management unit)
Recovery type meter with rollback point method
In the computer system, When the generation is updated at the rollback point, the external storage
Switch the storage device to use a different external storage device between successive generations.
Let me use The difference between the current generation and the previous n generations is
It is especially important to memorize whether the page has changed between generations.
Failure recovery type computer system. 3. The fault recovery type computer system according to claim 1 , wherein n-1 and n-2 are managed by three generations of MMU generation management.
A failure recovery type computer system characterized in that, when the contents between the generations are the same, the page contents of the n-2 generation are rolled in from the external storage device of the n-2 generation. 4. The fault recovery type computer system according to claim 1 , wherein each memory of the external storage device is managed during MMU generation management.
A failure recovery type computer system characterized by storing a change status between two generations before and after a page . 5. A fault recovery type computer system according to claim 2.
When the rollout in n generations
Data to other external storage devices for future generations.
Recovery type computer system characterized by writing at the time
M 6. The fault recovery type computer system according to claim 1, wherein the MMU hardware has only a translation mechanism between a current generation virtual address and a real address.
The MMU mapping information of the old generation is stored in an external storage device, and is used for rollback processing when an error occurs.
A fault recovery type computer system loaded on U hardware.