JPS63259733A - Contents insurance system for volatile storage device resident information - Google Patents

Abstract

PURPOSE:To improve a system performance by limiting a processing which fetches a resident information on a main storage device to a non-volatile storage device at the time of a check point, only for a part which is updated by a transaction, completed after the check point of the previous time. CONSTITUTION:At the time of the start of a system, etc., the non-volatile storage device 24 which can be accessed by a high speed random access in order to insure the contents of the restoration object information of a resident table or a resident data base, made resident on the main storage device 21, is connected to a CPU/main storage device 21. The resident information in a volatile main storage device 21 is identified at every optional sized input output unit and at the time of the periodic check point, the input output unit updated by the transaction completed after the check point of the previous time, is written to the non-volatile storage device 24. Thus, the quantity of the resident information to be fetched to the non-volatile storage device 24 at the time of the check point is reduced, and a resident information contents insurance overhead can be reduced.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a content guarantee method on a volatile storage device such as a main memory device in a data processing system that performs transaction processing, and particularly relates to a content guarantee method for reducing content guarantee overhead and system This invention relates to a content guarantee method for resident information on a volatile storage device that enables shortening recovery time from failure.

[Conventional technology]

Traditionally, in data processing systems that perform transaction processing such as online systems, checkpoints are set periodically to maintain the logical consistency of the system even in the event of a system failure due to a power outage, etc. Recording information in main memory to non-volatile storage is described in Notes on Database Operating Systems by J.N. Gray (1978).
) Operating Systems: An Advanced Course Chapter (J, N, Gray: Note
s onData Ba5e Opsrating S
operating systems
: An Advanced Course, R,B
ayer, R.M.

Graham and G, Seegmuller
(eds, ) Springer-Verl, ag.
1978). According to this method, when a system failure occurs, it is possible to restore the logical consistency of the system using the contents and history information stored in the nonvolatile storage device.

[Problem that the invention seeks to solve]

The above conventional technology has the following problems.

(1) Since the entire contents of the recovery target information on the main storage device are acquired into the non-volatile storage device at the time of a checkpoint, the writing process may become a bottleneck and reduce the performance of the resystem.

(2) In recovery processing when a system failure occurs, the logical consistency of recovery target information is restored using history information after the checkpoint. The checkpoint interval needs to be wider than the time required to acquire resident information to the nonvolatile storage device, and cannot be made too short. For this reason, the range of history information necessary for system failure recovery expands, and it takes time to read the history information. Furthermore, in preparation for the next system failure, all contents of the recovery target information on the main storage device after recovery are acquired into the nonvolatile storage device. These processes may prolong system failure recovery time.

An object of the present invention is to improve system performance by limiting the process of acquiring resident information on the main storage device to a nonvolatile storage device from checkpoint 1 to the portion updated by completed transactions since the previous checkpoint. By reducing the amount of historical information read during system failure recovery, and reducing the amount of processing required to retrieve resident information to nonvolatile storage after recovery, system failure recovery time can be significantly shortened. be.

[Means for solving problems]

In order to solve the above problems, the present invention provides a three-period system that identifies resident information that resides in a volatile device such as a main memory and that should be recovered in the event of a failure for each input/output unit of arbitrary size. At the time of a checkpoint, the contents of the input/output unit updated by the transaction completed since the previous checkpoint and the uncompleted transaction identifier that updated the input/output unit are recorded in a nonvolatile storage device, and the uncompleted transaction is updated. Record history information in a history information file. Examples of nonvolatile storage devices include semiconductor storage devices that have been made nonvolatile by a battery device, etc. (for example,
A storage device that allows high-speed random access, such as a semiconductor disk or a non-volatile expanded storage device, is used. When a system failure occurs, the contents of each input/output unit recorded in the nonvolatile storage device are first read into the resident information area in the volatile storage device. Next, for uncompleted transactions recorded at the time of the checkpoint that did not end normally, we eliminate the influence based on the update history information in the history information file, and update the 1st to 3rd runs completed after the checkpoint. The logical consistency of the resident information content is restored by reflecting the content based on historical information.

[Effect]

The effects of the above means are summarized below.

(1) Resident in a volatile storage device such as the main storage device,
Resident information that should be recovered in the event of a failure is identified for each input/output unit of arbitrary size, and at periodic checkpoints, the input/output units updated by transactions completed since the previous checkpoint are stored in a nonvolatile storage device. Write it down.

Therefore, at the time of a checkpoint, the amount of resident information acquired in the nonvolatile storage device is reduced, and the overhead of guaranteeing the content of the resident information at the time of a checkpoint can be reduced. Furthermore, since the resident information acquisition time is shortened, the checkpoint interval can be shortened. Therefore, in the event of a system failure,
System failure recovery time can be reduced because the amount of historical information required to restore the logical consistency of resident information is reduced.

(2) If the exclusive unit for resident information is different from the input/output unit for guaranteeing the contents of resident information, a situation may occur in which multiple transactions update the same person's output unit.

Furthermore, there may be cases where the input/output unit written at periodic checkpoints has been updated by an unfinished transaction. In the present invention, at the time of a checkpoint, the input/output unit updated by the transaction completed after the previous checkpoint is written to the nonvolatile storage device, and then the update history information of the uncompleted transaction that updated the input/output unit is stored in the history. Record in information file. Therefore, the logical consistency of the resident information can be restored by the resident information contents stored in the nonvolatile storage device at the time of the checkpoint and the update history information of the history information file.

〔Example〕

An embodiment of the present invention will be described below with reference to FIGS. 1 to 6.

First, the overall configuration of this embodiment will be explained using FIG. 2.

FIG. 2 is a diagram showing the overall configuration of the data processing system in this embodiment, and shows an information storage medium 22, such as a magnetic disk, that stores information groups such as a database. Guarantees the contents of the recovery target information such as the magnetic disk (magnetic tape or semiconductor storage device) that is the storage medium of the history information file 23, resident tables and resident databases that are made resident on the main storage device at the time of system startup, etc. A non-volatile memory device 24 capable of high-speed random access, such as a non-volatile semiconductor memory device, is connected to the CPTJ/main memory device 21. The nonvolatile storage device 24 has a control information area 2
5 and a resident information guaranteed area 26.

FIG. 3 is a diagram showing the information content on the main storage device,
Resident information group 32, such as resident tables and resident databases.
Update data management unit 35. A nonvolatile storage device management section 36 exists on the main storage device 31.

When identifying the resident information group 32 for each page, which is a logical input/output unit, the relationship between the exclusive control range of each resident information and the page may be arbitrary, and in this embodiment, the exclusive boundary 33a
In the resident information A to D separated by 33d, the resident information A, B, and C are arranged within one page separated by the page boundary 34, and the resident information occupies one page.

FIG. 4 is a diagram showing the information contents of the update data management section 35. The transaction information table 41 exists for each transaction in the execution state, and holds the identifier of each transaction, the start history information address of the transaction, and the identification information of the page in the resident information group 32 updated by the transaction. The update data management table 42 manages page identifiers output at checkpoints, and is composed of two tables so that page identifiers can be registered in separate tables for each checkpoint. The valid information management table 43 manages information indicating the updated data management table 42 that is registered between checkpoints.

The history information management table 44 is a table that records addresses at which history information is to be written next in the history information file 23 at the start of checkpoint processing and after page output. The current information table 45 is a table that records the page identifier to be written to the nonvolatile storage device 24 and the block number in the resident information guaranteed area 26 to which the page is written at the time of a checkpoint.

FIG. 5 is a diagram showing the information contents of the nonvolatile storage device management section 36 and the nonvolatile storage device 24. As shown in FIG. The resident information guarantee area 26 is an area for storing each page that divides the resident information group 32, and is divided into blocks of the same size as the page, and each block is assigned a sequence number from the beginning.

The page table 51 is a table consisting of the above-mentioned block numbers indicating the storage positions in the resident information guaranteed area 26 on the nonvolatile storage device 24 for each page that divides the resident information group 32.

The block usage bitmap 52 is the resident information guaranteed area 2.
This is a bitmap representing the usage status of each block in block 6. In this embodiment, bit 11' is used for the block in use.
, bit '0' is set for unused blocks. The control information area 25 is an area for storing control information such as the page table 51 and block usage bitmap 52, and is composed of two areas so that control information can be stored in a different area for each checkpoint than the previous one. The valid area management table 53 manages information indicating valid control information areas 25.

Next, system operation will be explained. First, an overview of the system operation will be explained using a simple example.

FIG. 6 shows a group of resident information 32. by transaction execution from system startup to system failure occurrence and recovery. 3 is a diagram showing state transitions of a history information file 23 and a resident information guaranteed area 26 on a nonvolatile storage device 24. FIG. In FIG. 6, resident information A to C are stored within the same page separated by a page boundary 34.

At system start time To, each page of the resident information group 32 is written to the resident information guaranteed area 26. Transaction 61 is started at time T1, changes resident information A to data aha' (this is defined as resident information NA'), and is completed at time T5. Transaction 62 is started at time T2, and is interrupted at time T1o (data b of resident information B is changed to b' and this is designated as resident information B') due to the occurrence of a system failure. transaction 63
is started at time T7, data C of resident information C is changed to C' at time T6, and this is designated as resident information C'), and completed at time T9. Update history information of the resident information updated by transactions 61 and 63 is acquired in the history information file 23 by the time each transaction is completed.

At checkpoint time T6, pages updated by completed transactions after the previous checkpoint (in the case of the first checkpoint after the system start, after the system start) are acquired in the resident information guarantee area 26.

Through this process, the resident information A', B', and C are recorded in the resident information guaranteed area 26. Thereafter, the current page update history information of the incomplete transaction that updated the page updated by the completed transaction is acquired in the history information file 23.

At TIO when a system failure occurs, the information on the main storage device is lost, but the resident information update results of completed transactions up to the previous checkpoint are reflected in the resident information guaranteed area 26 on the nonvolatile storage device 24. .

Therefore, it is possible to guarantee the logical consistency of the system based on this content and the resident information update history information in the history information file 23.

Next, system operation according to the present invention is system start operation 9.
Operation at appropriate times, operation at checkpoints.

We will explain the operation at checkpoint and the operation at transaction failure.

(1) Operation when starting the system The processing according to the present invention when starting the system is as follows (a):
Shown in (b).

(a) In page writing system start processing, the resident information group 32 created on the main storage device is written to the nonvolatile storage device 24. Specifically, for each page of the resident information information 32, an unused block is found in the block use bit map 52, the bit is set to '1', and the block number is recorded in the page table. After that, the page is written to the block recorded in the page table 51.

(b) Control information writing page table 51. The block usage bitmap 52 and the writing start time are written to the control information area 25. Information is recorded in the effective area management table 53 to indicate the written area. Specifically, two areas in the control information area 25 are associated with bit 10' and bit '1', and bit information indicating the area where the control information has been written is recorded.

Bit 10″ is set as an initial value in the effective information management table 43.
Record.

(2) Normal operation Processing according to the present invention from the start to the end of a transaction is shown in (a) and (b) below.

(a) Recording of update page When resident information is updated for the first time during transaction processing, the transaction identifier and the identification information of the update data are recorded in the transaction information table 41. Thereafter, when the current transaction updates the resident information, the identification information of the updated page is recorded in the transaction information table 41 in which the current transaction identifier is recorded.

(b) When the history information acquisition transaction updates resident information, the updated history information is acquired into the history information file 23. At this time, the start address of the first history information of the transaction is recorded in the transaction information table 41.

When the transaction is completed, update the updated page identifier of the completed transaction in Data Management Table 4
Record in 2. Specifically, page identifiers for completed transactions in the transaction information table 41 are recorded in the update data management table 42 indicated by the contents of the valid information management table 43 asynchronously with transaction processing.

The transaction information table 41 for which optical processing has been completed is cleared.

(3) Operation at checkpoint A checkpoint is set at the start of the system or at the time when a certain number of pieces of history information have been acquired from the previous checkpoint, in order to reduce the time required for recovery in the event of a system failure. Among the checkpoint operations, processing according to the present invention is shown in FIG. 1(a). Hereinafter, the explanation will be given according to FIG. 1 (,).

(a) Switching the update data management table (recording the exclusive OR of the bit contents of the step valid information management table 43 and bit '1' in the valid information management table 43. With this process, a completed transaction is The updated data management table 42 that records updated page identifiers is switched.

From the transaction information table 41, a list (incomplete transaction list) consisting of identifiers of uncompleted transactions and first history information addresses is created.

(b) Recording history information address 1 (step 11) In the history information file 23, the address (referred to as history information address 1) at which history information is to be written next is recorded in the history information management table 44.

(c) Output of updated page (step 12) The page recorded in the updated data management table 42 indicated by the exclusive OR of the bit content of the valid information management table 43 and bit 11' is written to the resident information guaranteed area 26. Specifically, an unused block is found in the block used bit map 52, the bit is set to '1', and the page identifier and assigned block number are recorded in the current information table 45. Thereafter, the page recorded in the current information table 45 is written to the allocated block. Regarding the transaction that was started during the processing in step 12 and updated the resident information, the identifier and first history information address of the transaction are recorded in the incomplete transaction list created in the processing in step 10.

(d) Acquisition of history information of incomplete transactions (step 13) The incomplete transaction that updated the page recorded in the current information table 45 is determined by the transaction information table 41, and the page update information by the corresponding transaction is stored in the history information file 23. to get to.

(e) Recording history information address 2 (step 14) In the history information file 23, the address at which history information is to be written next (referred to as Jli history information address 2) is recorded in the history information management table 44.

(f) Acquisition of control information (step 15) Above step 1
When the process 4 is completed, the following processes (i) and (ii) are performed for the page recorded in the current information table 45.

(i) Based on the block number recorded in the page table 51, the corresponding bit in the block usage bitmap 52 is set to 10'.

(ji) The block number recorded in the current information table 45 is recorded in the corresponding page of the page table 51.

Thereafter, the page table 51. History information management table 44. An uncompleted transaction list (a list consisting of the identifiers of uncompleted transactions in the transaction information table 41 and the first history information address) and the recording time are written out. Thereafter, the contents of the effective area management table 53 are changed to indicate the control information area 25 written in this process.

(4) Operation during system failure recovery Among the system failure recovery operations, the processing according to the present invention is shown in FIG. 1(b). This will be explained below with reference to FIG. 1(b).

(a) Reading of control information (step 16) Compare the writing times recorded in the two control information areas 25, and add bits to the effective area management table 53 to indicate the area where the control information acquisition time is slower. Record the information and extract it from the page table 51. History information management table 44. Read the incomplete transaction list into main memory.

A block use bitmap 52 is created from the block numbers recorded in the page map 51.

(b) Reading resident information (step 17) Referring to the page table 51, blocks corresponding to each page are read from the resident information guaranteed area 26 on the nonvolatile storage device 24 into the main storage device.

(c) Recovery using B-layer information (step 18) From the time of failure to history information address 2, history information file 23
are read in reverse order to determine which transactions are incomplete and which have a transaction failure among the transactions recorded in the incomplete transaction list. Regarding the transaction, the history information file 23 is read in reverse order from the history information address 2 to the first history information address of the transaction, and the influence on the resident information group 32 is eliminated.

Thereafter, the history information file 23 is sequentially read from the history information address 1 to the point of failure occurrence, and the resident information update result of the completed transaction is reflected in the resident information group 32, thereby restoring the logical consistency of the resident information group 32.

The page identifier targeted during the recovery process is recorded in the update data management table 42.

(d) Writing changed pages Step 19) The pages recorded in the update data management table 42 are written to the resident information guaranteed area 26 in the nonvolatile storage device 24 in the same manner as the checkpoint processing.

(5) Operation when a transaction failure occurs The processing according to the present invention when a transaction failure occurs is shown below.

Transaction management table 4 of completed transactions when a transaction is completed during normal operation
1, the page identifier updated by the transaction is recorded in the updated data management table 42. When a transaction failure occurs, the transaction management table 41 for the transaction is cleared without performing the above processing.

Furthermore, in normal operation, when resident information is updated after update history information is acquired in the history information file 23, the content of the resident information can be guaranteed by modifying the above embodiment as follows.

In the history information acquisition process during normal operation, the start addresses of the first and latest history information recorded in the history information file 23 for each transaction are recorded in the transaction information table 41. In the checkpoint operation control information acquisition process (step 15), the uncompleted transaction list (transaction information table 41
A list consisting of the identifier of the incomplete transaction and the first and latest history information addresses) is written to the control information area 25. Furthermore, recording of history information address 1 (step 11) is omitted.

In the recovery process (step 18) based on the history information of the system failure recovery operation, the following process is performed. History information file 23 from the time of failure to history information address 2
are read in reverse order, and among the transactions recorded in the incomplete transaction list, it is determined which transactions are incomplete or have a transaction failure, and which are completed. For a transaction that is incomplete or a transaction failure has occurred, the history information is read in reverse order from the history information address 2 to the oldest history information address among the first history information addresses of the transaction, and the resident information group 3 is read.
Eliminate the influence on 2.

After that, the history information file is sequentially read from the oldest history information address among the latest history information addresses of the completed transaction to the point of failure occurrence, and the resident information update result of the completed transaction is reflected in the resident information group 32, and the resident information group 32 logical consistency is restored.

As described above, according to this embodiment, it is possible to solve the problems in the above-described conventional content guarantee method for resident information resident in the main storage device or the like. Further, as an effect of this embodiment, the input/output unit for guaranteeing the contents of resident information can be set to a size that takes system efficiency into consideration, regardless of the exclusive control unit of resident information.

〔Effect of the invention〕

According to the present invention, in an information processing system that performs transaction processing based on resident information resident in a volatile storage device such as a main storage device, at a checkpoint, information is updated by transactions completed since the previous checkpoint. Since only the input/output units that have been input/output are acquired in the nonvolatile storage device, the overhead of guaranteeing the content of resident information at the time of the checkpoint can be reduced. Furthermore, since the time required to acquire resident information at checkpoints is shortened, checkpoint intervals can be shortened. Therefore, the amount of history information required to restore the logical consistency of resident information during system failure recovery is reduced, and system failure recovery time is shortened.

[Brief explanation of drawings]

FIG. 1 is a flowchart of resident information guarantee processing at checkpoint in an embodiment of the present invention (a), a flowchart of system failure recovery processing (b), FIG. 2 is an overall configuration diagram of a system in an embodiment of the present invention, 3 is an information content diagram on the main storage device in this embodiment, FIG. 4 is an information content diagram of the update data management section in this embodiment, and FIG. 5 is a diagram of the nonvolatile storage device management section and the device in this embodiment. The information content diagram in FIG. 6 is an explanatory diagram showing the state transition of the system in a specific example of this embodiment. 21... CPU/main storage device, 22... Information storage medium, 23... History information file, 24... Non-volatile storage bag, 25... Control information area, 26... Resident information Guaranteed area, 31... Main storage device, 32... Resident information group, 33a to 33d... Exclusive boundary, 34... Page boundary, 35... Update data management unit, 36... Non-volatile Storage device management unit, 41... Transaction information table, 42... Update data management table, 43.
... Effective information management table, 44... History information management table, 45... Current information table, 51...
Page table, 52...Block usage bitmap, 53...Valid area management table, 61-62...
transaction.

Claims (1)

[Claims] 1. In an information processing system that performs transaction processing based on resident information resident in volatile storage devices such as main memory, resident information that should be recovered in the event of a failure is determined for each input/output unit of arbitrary size. identifying, at system startup, if the resident information content is not stored in non-volatile storage, writing that content to non-volatile storage; and, at periodic checkpoints, depending on transactions completed since the last checkpoint; recording the updated contents of the input/output unit and the incomplete transaction identifier that updated the input/output unit in a nonvolatile storage device;
recording update history information of the incomplete transaction in a history information file; upon recovery from a system failure, reading the contents of each input/output unit recorded in the nonvolatile storage device into a resident information area in the volatile storage device; For uncompleted transactions recorded at the time of checkpoint that did not end normally, we eliminate the influence based on the update history information recorded in the history information file, and create a history of the updated contents of transactions completed after the checkpoint. Reflecting the information and restoring the resident information content;
A content guarantee method for volatile storage device resident information.