JPH05108388A - Process restoration system - Google Patents

Abstract

PURPOSE:To reduce an overhead for the process restoration by copying only the revised part without copying the entire file image at the check point. CONSTITUTION:An operation system(OS) 101 is provided with a write trap processing part 102, a check point processing part 103, and a process restoration part 104. The write during process execution is detected by the write protect function of an MMU(memory management unit) 105, and the copy of the written part is prepared to revise the copy. By operating a page table 107 of an MM (main storage) 106, the copy is reflected on the copy source at the time of passing the check point during process execution. In the case the process restores the revised data at to the data the time of latest check point, the restoration of the process is performed by making the copy invalid.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a process recovery system for recovering data updated by a process at the nearest checkpoint in an operating system.

[0002]

2. Description of the Related Art When a process cannot be continued due to an error during execution of the process, a checkpoint is set appropriately during the process, and the latest checkpoint is made when the process cannot be continued. It may be necessary to recover the data up to that checkpoint and restart the process from that checkpoint.

For example, in a database system, when a deadlock occurs during execution of a transaction, the data updated from the most recent commit point to the time when the deadlock occurs is returned to the original and the execution is executed from that commit point. Need to restart.

In a multiprocessor system in which a plurality of processors take a process from one execution queue and execute the process, when a certain processor cannot continue execution due to a hardware failure or the like, the process is dispatched. It can be executed by another processor by returning the changed data to the original state, returning the state of the process to the state at the time of execution, and connecting it to the execution queue.

In such a process recovery method, there are the following two methods for restoring the data updated by the process up to the latest checkpoint. (1) A method of copying the entire data image (data area) of the process at the checkpoint, invalidating the data image at that point in time when restoration is necessary, and validating the copied data image.

(2) A method of retrieving a data image at the time of checkpoint based on the history of data update which keeps track of which data the process changed and how.

However, in the method (1), the data image at the checkpoint must be saved,
A memory area for storing the data image is required, and the overhead for copying becomes large.

Further, in the method (2), it is necessary to take a history of data update of the process, the overhead of copying the history is a source, and at the time of restoration, it takes time and effort to restore the data based on the history.

[0009]

As described above, in the conventional method, the overhead of saving the data image, saving the data update history of the process, and restoring the data based on the history is large. SUMMARY OF THE INVENTION It is an object of the present invention to eliminate the above-mentioned drawbacks of the prior art and provide a process restoration method that reduces the overhead for process restoration.

[0010]

In order to achieve the above object, in the present invention, a write protect function of an MMU (memory management unit) is used to detect writing during process execution, and to detect a portion where writing has occurred. Make a copy and make updates to it. Also, by manipulating the page table of the MMU, the copy is reflected in the copy source when the checkpoint passes during the process execution, and the copy is invalid when the data updated by the process is restored at the latest checkpoint. To restore the process.

[0011]

In such a system, the entire file image at the checkpoint is not copied, but only the updated part is copied. Further, at the time of restoration, the history of process execution is not required and it is only necessary to invalidate the copy, so the overhead for restoration of the process can be reduced.

[0012]

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

FIG. 1 shows a block diagram of this embodiment. The operating system (OS) 101 includes a write trap processing unit 102, a checkpoint processing unit 103, and a process restoration unit 104. The operating system 101 realizes virtual storage by the MMU 105. Further, on the main memory (MM) 106, there are a page table 107 and a copy management table 108.

The operation of FIG. 1 will be described with reference to the flow of FIG. 2A shows the flow of processing for the event of "write", (b) shows "checkpoint", and FIG. 2C shows the flow of processing for "process recovery" event. When a "write" event occurs, write protection is applied in advance by the MMU 105 to generate an exception (interrupt) in the MMU (step 1). The exception is input to the write trap processing unit 102 by the write trap from the MMU 105. The write trap processor 102 checks the copy management table 108 to detect whether or not a copy exists (step 2). If there is no copy, a copy of the page in which the writing has occurred is made (step 3) and the page table 107 is updated (step 4). This update is to change the portion of the page table 107 that points to the page where the writing has occurred to point to the newly created copy. After that, the copy management table 108 stores and manages the changes. (Step 5). Then, in step 6, the copy is updated.

Next, the "checkpoint" event will be described. The checkpoint processing is performed by the checkpoint processing unit 103. The processing unit 103 enables the copy by clearing the copy management table 108. In the “process recovery”, a recovery command from the processor is input to the process recovery unit 104, and the contents of the copy management table 108 are changed to the page table 107.
Disable the copy by writing it back to and restart the process from the closest checkpoint.

FIG. 3 is a diagram showing the correspondence between the virtual space and the real space (main memory) and the contents of the page table and copy management table. At this point, a write operation for the page number Va mapped to the page number Ra in the real space 302 occurs (a → a ′ in the virtual space 301), and a copy Ra ′ of the page Ra in the real space 302 is created and updated. think of. In this case, the page number of the page table is updated from Va to Ra ′ as shown in FIG. 3B, and the page number of the copy management table is Va to Ra.
Has been changed to.

As described above, the process P is the file F.
A detailed description will be given of the manner in which the content of is updated over a plurality of pages. FIG. 4 shows the flow of the process P. After passing through the checkpoint 1, the checkpoint 2
During this period, it is assumed that page a is updated to a ', b to b', and c to c '.

FIG. 5 shows an image of the virtual space and the real space of the file F which is the target of execution of the process p. The page numbers Pa, Pb, and Pc in the virtual space have contents a, b, and c, respectively, and the page numbers Ra and R in the real space are included.
It is mapped to b and Rc. FIG. 6 shows a part of the page table of the file F. Process p
Changes the content a of the page number Pa to a'after the checkpoint 1. Similarly, the content b of the page number Pb is changed to b ′, the content c of the page number Pc is changed to c ′, and the checkpoint 2 is reached. The file image is write protected after commit point 1 and a trap file is processed to handle any subsequent writes.
The routine makes a copy of the page that was written to, updates the page table. Also,
The trap routine registers in the copy management table which real space the updated page was mapped to before the update. That is, immediately before the checkpoint 2, the file image is as shown in FIG. The page table and copy management table are shown in FIG. 8 and FIG. 9, respectively.
It looks like. After that, when the check point 2 is passed, the update becomes effective by clearing the copy management table. FIG. 10 shows a file image at time t in FIG. FIG. 11 is a page table at time t. FIG. 12 shows the copy management table at time t. When a deadlock or the like occurs at this time, the page table becomes as shown in FIG. 6 by writing back the contents of the copy management table to the page table, and the copied pages Ra 'and Rb' are invalidated. After that, by restarting the execution of the process p from the checkpoint 1, it becomes possible to recover.

[0019]

As described above, according to the present invention, the entire file image (entire area) at the checkpoint is not copied, only the updated portion (page) is copied, and at the time of restoration, Since the history of process execution is not required and only the copy is invalidated, the overhead for process recovery can be reduced.

[Brief description of drawings]

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

FIG. 2 is a flowchart showing the flow of processing.

FIG. 3 is a diagram showing a state in which writing has occurred.

FIG. 4 is a diagram showing a flow of execution of a process p.

FIG. 5 is a diagram showing a file image of file F.

FIG. 6 is a diagram showing a part of a page table of file F.

FIG. 7 is a diagram showing a file image immediately before checkpoint 2.

FIG. 8 is a diagram showing a page table immediately before checkpoint 2.

FIG. 9 is a diagram showing a copy management table immediately before checkpoint 2.

FIG. 10 is a diagram showing a file image at time t.

FIG. 11 is a diagram showing a page table at time t.

FIG. 12 is a diagram showing a copy management table at time t.

[Explanation of symbols]

 101 ... Operating system 102 ... Write trap processing routine 103 ... Checkpoint processing unit 104 ... Process recovery unit 105 ... MMU 106 ... Main memory 107 ... Page table 108 ... Copy management table

Claims (1)

[Claims] 1. When a predetermined process is executed while setting check points as appropriate, and when the process cannot continue to be executed during process execution, data is collected from the time when the process cannot be continued to the closest check point. In the process recovery method that recovers and re-executes the process from this checkpoint, the memory management unit is used to detect the writing to the memory during the process execution and make a copy of the page where the writing occurred Update, and operate the page table used by the memory management unit to enable or disable the copy to the copy source, and reflect the copy to the copy source at the checkpoint during the process execution. Checkpoint the data updated by A process recovery method characterized in that the copy is invalidated when it is recovered at times.