JPS63133240A - Contents assurance system for residence table - Google Patents

Abstract

PURPOSE:To extremely improve the performance of the titled system by omitting a process to obtain all contents of a residence table into a nonvolatile medium in a check point mode. CONSTITUTION:A residence table group 22 is divided into optional input units, such as pages. The contents of the table 22 are written into a nonvolatile memory medium 12 if said contents are not stored in the medium 12 when a system is started. Then an input/output unit containing the part of the table 22 updated by a transaction is written into a nonvolatile recording medium which is capable of a high-speed random access when said transaction is through. Thus it is not required to obtain the contents of the table 22 into the medium 12 in a periodical check point mode. As a result, the system performance is extremely improved.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a method for guaranteeing the contents of a so-called resident table, which is a resident area provided in the main memory at the time of system startup, etc. in an information processing system. This invention relates to a method for guaranteeing the contents of a resident table that makes it possible to reduce the overhead of guaranteeing the contents of the resident table and to significantly shorten the recovery time of the resident table in the event of a system failure.

[Conventional technology]

Traditionally, when a system failure occurs, checkpoints are set periodically to ensure the logical consistency of the system without recovering from the point at which the system started, and the memory contents are recorded on a non-volatile storage medium at the time of the checkpoint. It is widely practiced.

In this regard, see, for example, J.N. Gray, Notes on Database Operating Systems (197
8) J (Notes on Database Operation)
ratingSystems、○perating
Systems An Advanced Couple
rse chapter, Springer-Verlag, Ne
w York).

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 medium.

[Problem that the invention seeks to solve]

However, the above conventional technology has the following problems. That is, (1) at the time of a checkpoint, the contents of the memory, that is, the entire contents of the resident table, are acquired into a nonvolatile storage medium, so the writing process becomes a bottleneck, which may reduce system performance.

(2) Recovery processing in the event of a system failure involves reading the contents of the resident table acquired at the time of the checkpoint from the nonvolatile storage medium into the memory, and then recovering the contents using history information after the checkpoint.

The checkpoint interval must be wider than the time required to retrieve the resident table contents to the non-volatile storage medium, and the checkpoint interval cannot be made smaller. Therefore, it takes time to read the history information and update the resident table using the history information. Furthermore, in preparation for the next system failure, the entire contents of the resident table after recovery are acquired into a non-volatile storage medium. These treatments can prolong recovery time.

The present invention has been made in view of the above circumstances, and its purpose is to solve the above-mentioned problems in the conventional resident table content guarantee system, and to transfer all contents of the resident table to a non-volatile medium at checkpoints. To provide a content guarantee method for resident tables that greatly improves system performance by eliminating acquisition processing, and greatly reduces system recovery time in the event of a failure, improving overall efficiency. be.

[Means for solving problems]

The above object of the present invention is, firstly, in an information processing system that performs transaction processing using resident tables made resident in memory, by dividing the resident table group into arbitrary input/output units, and , if the contents of the resident table are not stored in a non-volatile storage medium, the contents are written to the non-volatile storage medium written in 1rI, and when the transaction ends, the pre-update information of the resident table updated by the transaction and the information updated by the transaction are written. The pre-update history information of the input/output unit of another transaction that updated the input/output unit that includes the resident table part is acquired in the history information file, and then the input/output unit is stored in a non-volatile memory that can be accessed at high speed and randomly. The contents of the resident table are guaranteed by writing to a non-volatile storage medium, and in the event of a system failure, the contents of the resident table stored in the non-volatile storage medium are read into memory, and the updated portion due to an incomplete transaction is recovered using pre-update information. This is achieved through a method.

Second, in an information processing system that performs transaction processing using resident tables made resident in memory, the resident table group is divided into arbitrary input/output units, and when the system starts, the contents of the resident tables are is not stored in a non-volatile storage medium, the contents are written to the non-volatile storage medium, and among these, for the input/output unit updated by the transaction at the end of the transaction, the input/output unit that has not been updated by the incomplete transaction The unit is written to the nonvolatile storage medium, and the input/output unit updated by the incomplete transaction acquires the updated history information to the history information file, and when a system failure occurs, the contents of the resident table stored in the nonvolatile storage medium are transferred to the memory. The updated portion of the input/output unit of the completed transaction, which has not been reflected in the non-volatile storage medium because it was updated by an uncompleted transaction, is updated using the resident table update history information. This is achieved by a resident table content guarantee method that is characterized by restoring the logical consistency of the table.

[Effect]

In the first aspect of the present invention, when guaranteeing the contents of resident tables, the resident table group is divided into arbitrary input/output units such as pages, and the contents of the resident tables are stored in a non-volatile storage medium at the time of system startup. If the transaction does not exist, the contents are written to the non-volatile storage medium, and at the end of the transaction, the input/output unit including the resident table part updated by the transaction is written to the non-volatile storage medium that can be accessed at high speed randomly. There is no need to acquire resident table contents to a non-volatile storage medium at the time of a checkpoint, and the processing at the checkpoint does not become a bottleneck and degrade system performance.

Furthermore, when recovering from a system failure, only the history information for uncompleted transactions needs to be read, reducing the time required to read the history information. Regarding updating of the contents of the resident table stored in the non-volatile storage medium, only the input/output unit corrected by recovery needs to be written, and the writing time is reduced. Therefore, the time required for recovery will be shortened.

In addition, in the second aspect of the present invention, when guaranteeing the contents of the resident table, the resident table group is divided into arbitrary input/output units such as pages, and the resident table is not stored in a nonvolatile storage medium at the time of system startup. However, when a transaction ends, the input/output unit updated by the transaction is written to the non-volatile storage medium, and the input/output unit updated by the incomplete transaction is written to the non-volatile storage medium. Acquire history information into a history information file. Therefore, there is no need to acquire the contents of the resident table into a non-volatile storage medium at the time of one periodic checkpoint, and the processing at the checkpoint does not become a bottleneck and degrade system performance.

Furthermore, the historical information that needs to be read during system recovery is
This is a completed transaction whose input/output unit is not reflected in the nonvolatile storage medium because it was updated by an unfinished transaction, and the time for reading history information and the time for update processing using updated history information are reduced. After failure recovery, the process of writing resident table contents to the nonvolatile storage device requires only the input/output units updated in the recovery process, reducing the write time. Therefore, the time required for recovery will be shortened.

〔Example〕

Embodiments of the present invention will be described in detail below with reference to the drawings. In this embodiment, a case will be described in which the input/output unit for dividing a resident table group is a page.

FIG. 2 is an overall configuration diagram of an information processing system showing an embodiment of the present invention, and is an information storage medium such as a magnetic disk that stores information such as a database]2. A non-volatile storage medium 14 capable of high-speed random access such as a semiconductor storage device for guaranteeing the contents of the resident table. History information file 1
The third medium, such as a magnetic disk or magnetic tape, is used by the CPU.
/ connected to memory 11.

FIG. 3 is a diagram showing the contents of the memory, in which a resident table group 22 consisting of resident tables A to E, etc. and a pre-update information buffer 25 exist in the memory 21, and an update page management table 23 in normal times. but.

During failure recovery, the failure recovery page bitmap 24 is
Exists in memory 21.

When dividing the resident table group 22 into pages, which are input/output units, the relationship between the arrangement of the resident tables in the memory and the pages may be arbitrary, and a plurality of resident tables may be arranged in one page. In this embodiment, resident tables A and B are 1
placed within the page.

Resident table C occupies one page, and resident table E is divided into two pages.

The pre-update information buffer 25 is provided to temporarily store the resident table pre-update information of the transaction being executed until it is acquired in the history information file 13. The updated page management table 23 is a table that registers pages that include resident table parts updated during transaction execution, and exists during normal operation.

The failure recovery page bitmap 24 is a bitmap in which pages into which the resident table group 22 is divided are associated with one bit, and exists at the time of failure recovery.

A nonvolatile storage medium 14 that stores the resident table group 22 is divided into pages of the same size as the pages into which the resident table group 22 is divided.

FIG. 4 is a specific example of this embodiment, where the same page i
There are two transactions 51, 52 that update resident tables A and B stored within.

Transaction 51 starts at time T1 and ends at time T3.
Record a of resident table A is changed to a' at 1 time T5. Note that the table in which record a is changed to a'' is referred to as resident table A''. Further, transaction 52 is started at time T2, changes record b of resident table B to b' at time T4, and is interrupted at time T6 due to a system failure. In addition, record b
The table with b' changed to b' is called resident table B'.

Further, FIG. 4 shows the time T3. Contents of the above-mentioned pre-update information buffer 25 at time T4, history information file 13 at time T5
, and the contents of the nonvolatile storage medium 14 at times T1 and T5.

Next, the operation of the system will be explained by dividing it into operation 1 at the time of system startup, operation at normal time, and operation at the time of system failure recovery.

(1) Operation when the system starts When the system starts, the contents of the resident table are not stored in the non-volatile storage medium due to starting a new system, etc.
Specifies whether the resident table contents are stored on a non-volatile storage medium.

(a) When the contents of the resident table are not stored in a non-volatile storage medium: If the contents of the resident table are not stored in a non-volatile storage medium, such as when starting a new system, the contents of the resident table are created in memory during system startup processing. writes each page of the resident table to a non-volatile storage medium.

(b) When the contents of the resident table are stored in a non-volatile storage medium If the contents of the resident table are stored in a non-volatile storage medium, such as after the planned termination of the system, no processing is necessary.

(2) Normal Operation FIG. 1(a) is a schematic flowchart of the normal operation of the system according to this embodiment. Processing according to the present invention from the start to the end of a transaction is performed in the order of steps 31 to 33. The operation will be explained below according to the diagram.

(a) Registration of updated page (step 31) When the resident table 22 is updated during Nidoranza transaction processing, the resident table pre-update information of the transaction is recorded in the pre-update information buffer 25, and the page containing the updated part is recorded. It is recorded in the updated page management table 23. In the example of FIG. 4, transaction 51 records pre-update information a in pre-update information buffer 25 at time T3, records page i in update page management table 23, and transaction 52 records pre-update information at time T4. This corresponds to the recording of pre-update information to the information buffer 25 and the recording of page i to the updated page management table 23.

(b) Processing when a commit instruction is issued (step 32) This process is performed when a commit instruction indicating the end of processing is issued from a program currently being executed.

i) Acquisition of pre-update information (step 321): For the page updated by the transaction recorded in the update table management table 23, obtain the pre-update information of the transaction and the page of the other transaction that updated the page. The pre-update information is written from the pre-update information buffer 25 to the history information file 13. In the example shown in FIG. 4, the transaction 51 writes the pre-update information a and b from the pre-update information buffer 25 to the history information file 13 at time T5.

ii) Writing page step 322): The page updated by the transaction recorded in the updated page management table 23 is written to the nonvolatile storage medium 14. In the example of FIG. 4, the transaction 51 writes page i containing resident tables A' and B' to the nonvolatile storage medium 14 at time T5. The writing position on the page can be one overwrite or a different position from the previous one.If you write to a different position from the previous one, even if a system failure occurs during writing, the contents of this page will not be destroyed. , which has the advantage of enhanced fault tolerance.

(c) Acquisition of process completion information (step 33): After the processing in step 32 is completed, process completion information (PJ) indicating that the transaction process has been completed is obtained.
) is acquired in the history information file 13. Once this process is completed, the transaction is considered complete, and the results are guaranteed from then on. The transaction number and the like are recorded in the process completion information. 4th v! i
In the example, transaction 51 transfers process completion information (P J (51)) to history information file 13 at time T5.
This applies to the process of acquiring.

(3) Operation during System Failure Recovery FIG. 1(b) is a schematic flowchart of the system failure recovery page according to this embodiment. The processing related to the present invention from the start to the end of system failure recovery is from step 41 to
Steps 44 are performed in sequence.

The operation will be explained below according to the diagram.

(a) Creating a disaster recovery page bitmap (step 41): Creates a disaster recovery page bitmap 24 in which each page that divides the resident table group 22 is associated with one bit, and initializes all bits of the bitmap to 110". become

This bitmap is used to identify pages targeted for recovery processing.

(b) Reading the resident table (step 42): The contents of the resident table stored in the non-volatile storage medium 14 are read into the memory. In the example in Figure 4, resident tables A' and B'
Page i, etc., containing the page i are read into the memory in this process.

(c) Recovery using history information (step 43): Read history information from the history information file 13 into the pre-update information buffer 25. The history information is analyzed, and if the resident table pre-update information is for a transaction that was interrupted due to a system failure before the process completion information is obtained, the resident table in memory is restored with the pre-update information. In the example of FIG. 4, transaction 51 includes process completion information (PJ
(51)) is taken! ), the transaction 52 is considered to have been completed, and since the process completion information has not been obtained, the transaction 52 is considered to be uncompleted.

In order to return the updated portion of the incomplete transaction 52 to its pre-update state, the pre-update information is used to restore resident table B.
' is restored to its pre-update state (resident table B). At this time, "1" is written in the bit position on the failure recovery page bitmap 24 corresponding to the recovered page. In the example of FIG. 4, 6'1'' is written to the bit position on the disaster recovery page bitmap 24 corresponding to page i.

(d) Writing the update page step 44): Search the failure recovery page bitmap 24 and perform the above II I 1
The page corresponding to the bit position in which 1 has been written is written to the nonvolatile storage medium 14 in preparation for the next system failure. In the example of FIG. 4, II L Jj is written in the bit position corresponding to page i that includes resident tables A' and B, so that page is written to the nonvolatile storage medium 14.

According to the above embodiment, in an information processing system that performs transaction processing using a resident table, there is no need to acquire the entire contents of the resident table to a non-volatile storage medium at the time of a checkpoint, so system performance decreases at the time of a checkpoint. There's nothing to do. In addition, in system failure recovery processing, system recovery can be performed using only the history information of uncompleted transactions, which reduces the time required to read updated information, and also corrects updates to resident table contents stored in non-volatile storage media during recovery. Since only the pages that have been written need to be written to the non-volatile storage medium, the writing time is reduced. Therefore, the reduction in input/output time has the effect of shortening system recovery time.

For example, the size of the resident table is 4MB, the division unit of the resident table is a page, the update per transaction is 2 pages (one page is 4KB), the checkpoint interval is 1000, and the resident table is updated once per transaction when a system failure occurs. Assuming that the number of updated transactions is 10, in this embodiment, compared to the conventional method, the amount of history information and the time required for system failure recovery are 10/1000 = 1/100, and the non-volatile storage medium Also updates the contents of the resident table stored in (2X 4
Xl0)/(4X1000) = 1150.

In the above embodiment, the failure recovery page bitmap 24 is used to record pages targeted for failure recovery, but this can also be realized by a list structure in which page identification names are recorded.

FIG. 5 is an overall configuration diagram of a system showing another embodiment of the present invention, and IIA-14A indicate the same components as the components 11-14 shown in FIG. 2 previously. However, the nonvolatile storage medium 14A is composed of a control information area +41 for storing control information, which will be described later, and a resident table guarantee area 142.

The control information area 141 has two areas (
It consists of a control information area O and a control information area 1). Furthermore, the resident table guaranteed area 142 is an area for storing the contents of the resident table, and is divided into slots of the same size as the pages into which the resident table is divided.

FIG. 6 is a diagram showing control information. The page map 61 records a slot address for each page in order to indicate the storage location of page contents with logical consistency that should be guaranteed in the event of a failure, and records slot addresses in the memory 21A of the CPU/memory and in the control information area 141. It is located in The slot usage map 62 is a bitmap indicating whether each slot in the resident table guarantee area 142 is in use or unused, and is placed in the memory 21A.

The current information list 63 includes the starting position in the history information file 13A in which history information is written for each transaction, the updated page number, and the resident table guaranteed area 1.
The slot address written in 42 is recorded and placed in the memory 21A.

The history information list 64 records history information necessary to recover pages updated by completed transactions that have not been written to the resident table guarantee area 142 because they have been updated by incomplete transactions. That is, among the transactions whose update results are not reflected in the resident table guaranteed area 142, the history information position of the transaction whose write position in the history information file 13A is the earliest is recorded. The history information list 64 is placed in the memory 21A and the control information area 141.

The control information map 65 is a bitmap indicating a valid area within the control information area 141, and is placed in the memory 21A. The valid area indicates the control information area 1 when the bit is 0'' and the control information area 01It I 11, for example.

FIG. 7 shows a specific example of the resident table content guarantee process. As shown in FIG. 7(a), when dividing the resident table group consisting of resident tables A to E, etc. in the memory 21A into pages of input/output units, the arrangement of the resident tables in the memory and the relationship with the pages are arbitrary. As described above, a plurality of resident tables may be arranged on one page.

The correspondence between the updated portion of the resident table and the page number is determined by calculating the page number based on the quotient of the displacement of the updated portion from the beginning of the resident table area in the memory divided by the page size. This may be done by creating a table that associates pages with resident tables stored there. A resident table boundary 75 indicates a boundary line between resident tables, and a page boundary 76 indicates a boundary line between pages that divide a group of resident tables. Further, a slot boundary 77 indicates a boundary line between slots that divides the resident table guaranteed area, which will be described later.

The arrangement of the resident tables in this embodiment is similar to that shown in the previous embodiment, where resident tables A and B are arranged within one page, resident table C occupies one page, and resident table E occupies two pages. It is divided into . Update resident tables A and B included in the same page i 3
There are three transactions 71.72.73.

Transaction 71 starts at time T1 and ends at time T
At time T3, record a1 of resident table A is changed to al' at time T3. Here, the table in which record a1 of resident table A is changed to a l l is referred to as resident table A'. Transaction 72 is executed at time T
4, and record b of resident table B at time T6.
Change 1 to bl'.

It is completed at time T8. Here, the table in which record b1 of resident table B is changed to b1'' is assumed to be resident table B'.
5, and record a of resident table A at time T7.
2 to a2', and was interrupted at time T9 due to a system failure. Here, record a of resident table A
2 is changed to a2' as resident table A"
shall be.

Further, FIG. 7(b) shows times Tl, T2. T6. “I”7
The contents of page i in the memory 21A at 21a to 21d, respectively, and time T3. Contents of the history information file 13A and nonvolatile storage medium 14A at T8,
13a, 13b; 14a, 14b respectively
It shows.

The operation of this embodiment will be explained below.

(1) Operation at System Startup The point of specifying whether or not the resident table contents are stored in the nonvolatile storage medium 14A at the time of system startup is the same as in the previous embodiment.

(a) When the resident table contents are not stored in a non-volatile storage medium: System startup processing related to the resident table guarantee method when the resident table contents are not stored in a non-volatile storage medium, such as when starting a new system. , hereinafter shown in i) and ii).

i) Memory 21A in page write system start process
For each page of the above resident table created in
An unused slot is found in the slot usage map 62 in the memory, its slot address is recorded in the page map 61, and the bit of the slot in question is turned on in the slot usage map 62. Thereafter, the page is written to the slot recorded in the page map 61.

ii) Writing control information: Record the bit LLOI+ (bit II 1 )H) in the control information map 65, and write the page map 61 and writing time in the area indicated by the control information map.

(b) When the contents of the resident table are stored in a non-volatile storage medium The following describes the processing when the contents of the resident table are stored in a non-volatile storage medium, such as after a planned termination of the system. , ji).

i) Reading control information: Control information area] Compare the write times recorded in the two areas (control information area O and control information area 1) in 41, and select the area where the control information was acquired later. Then, bit information is recorded in the control information map 65, and the page map 61 is read into the memory from the recorded area. A slot usage map 62 is created from the slot addresses recorded in the page map 61.

ji) Reading the resident table: Refer to the page map 61 and read the slot corresponding to each page from the resident table guaranteed area 142 into the memory.

(2) Normal operation FIG. 8(a) is a schematic flowchart of the normal operation of the system according to this embodiment. The processing according to the present invention from the start to the end of a transaction is performed in the order of steps 81 to 85.

(a) Registration of update page (Step 8): When the resident table is updated for the first time during transaction processing, the transaction identification name and the page number containing the updated part are recorded in the current information list 63, and the , when the resident table is updated, the page number containing the updated part is recorded in the current information list 63 in which the transaction identification name is recorded. In the example of FIG. 7, at time T2, page i containing the transaction identifier of transaction 71 and resident table A' is recorded, and at time T6, page i containing the transaction identifier of transaction 72 and resident table B' is recorded. This corresponds to the process of recording page i that includes the transaction identifier of transaction 73 and resident table A II at time T7.

(b) Writing an updated page or acquiring post-update information at the time of a commit command (step 82) This is a process performed when a commit command indicating the end of processing is issued from a program being executed in the Nidoranzaktion. The current information list 63 is searched, and for pages updated by the transaction, pages that have not been updated by incomplete transactions are written to the resident table guarantee area 142, and pages updated by incomplete transactions are written to the post-update history information. Acquire into history information file +3A.

i) Page write processing: Among the pages updated by the transaction, pages that have not been updated by the incomplete transaction are written to the resident table guaranteed area 142. When writing page i, an unused slot is found in the slot use map 62 in the memory, its slot address (supposed to be k) is recorded in the current information list 63, and page 1 is written to that slot. The bit corresponding to the slot k that is newly in use is turned on in the slot use map 62 in the memory.

In the example of FIG. 7, this corresponds to the process of writing page i updated by transaction 71 to slot k at time T5.

ii) History information acquisition processing 2 Among pages updated by the transaction, for pages updated by incomplete transactions, the slot address in the current information list 63 is set to null, and the updated history information is acquired to the history information file 13A. do. At this time, the starting position of the transaction written to the history information file 13A is recorded in the current information list 63. In the example in Figure 7,
At time T8, the updated history information (c
This corresponds to the acquisition process of J(bl')) to the history information file 13A and the acquisition process of its position (history information position 72) to the current information list 63.

(c) Acquisition of process completion information (step 83): After the process of step 82 is completed, the process completion information (PJ) is acquired in the history information file 13A. As mentioned above, once this process is completed, the transaction is considered to have been executed, and the results are guaranteed thereafter. The transaction number and the like are recorded in the process completion information.

In the example of FIG. 7, at time T3, the process completion information of transaction 71 (P J (71)) is acquired into the history information file 13A, and at time T8, the process completion information of transaction 72 (P J (72)) is acquired. This corresponds to the process of acquiring the history information file 13A.

(d) Acquisition of control information (step 84): When the process of step 83 above is completed, δ is added to the current information list 63.
About the self-recorded page.

The following processes i) to 1ii) are performed.

j) Based on the slot address recorded in the page map 61 in the memory, turn off the corresponding bit in the slot use map 62.

ii) For pages whose slot address recorded in the current information list 63 is not null, the slot address is recorded in the page map 61 in the memory, and if the page is recorded in the history information list 64, the slot address is recorded from the history information list 64. Delete the page and its information.

iii) For pages whose slot address recorded in the current information list 63 is null, record the history information position and page number recorded in the current information list 63 in the history information list 64.

After this, the bit contents of the control information map 65 and the bit LL
The page map 61.I in memory is placed in the area indicated by the exclusive OR of I11. & Layer information list 54. Writes an incomplete transaction list, which will be described later. The incomplete transaction list is a list of transaction identification names other than the transaction currently being committed recorded in the current information list 63. Thereafter, the contents of the control information map 65 are updated to indicate the area written in the current process, and the writing time is further recorded in that area.

In the example of FIG. 7, the process of recording control information related to the commit process of transaction 71 in the control information area 141 at time T3, and the process of recording control information related to the commit process of transaction 72 in the control information area 141 at time T8 are applicable. do.

(e) Acquisition of transaction completion information (step 85
)2 When the process of step 84 is completed, the transaction completion information (TJ) is acquired in the history information file 13A. Specifically, the transaction number is recorded. In the example of FIG. 7, the transaction completion information (T J (71)) of transaction 71 is acquired in the history information file 13A at time T3, and the transaction information (T J (71)) of transaction 72 is acquired at time T8.
This corresponds to the process of acquiring 2 to the history information file 13A.

(3) Operation during system failure recovery FIG. 8(b) is a schematic flowchart of the system failure recovery operation according to this embodiment. Processing related to the present invention from the start to the end of system failure recovery includes steps 91 to 91.
Steps 94 are performed in sequence.

(a) Reading control information (old step): Compare the write times recorded in the two areas in the control information area 141, and write the control information map 65 to indicate the area where the control information was acquired later. Bit information is recorded and a page map 61, history information list 64 . Read the incomplete transaction list into memory. A slot use map 62 is created from the slot addresses recorded in the page map 61.

(b) Reading the resident table (step 92) Referring to the two-page map 61, the slot corresponding to each page is read from the resident table guaranteed area 142 into the memory. 7th
In the example shown in the figure, pages i and the like containing resident tables A' and B are read into the memory in this process.

(c) Recovery using 8 history information (step 93): Search the history information list 64 in the memory, and among the transactions that updated the page to be recovered, history information of the transaction recorded in the history information file 13A earliest. Determine the position. The history information is read from this position in the acquisition type, and the resident table in the memory is updated using the history information after updating the resident table of the transaction that is not recorded in the incomplete transaction list on the recovery target page recorded in the history information list 64. Update. The example in FIG. 7 corresponds to the process of updating resident table B to resident table B' using resident table update information (CJ(bl')).

(cl) Write updated page step 94): The page recorded in the history information list 64 in the memory is written to the resident table guaranteed area 142 in the same manner as the page write processing at the time of commit. In the example of FIG. 7, resident table A'
Since the page i containing the same B' is written in the history information list 64, the page is ejected to the non-volatile storage medium. After this.

The history information list 64 in the memory is reset, and the control information is updated in the same manner as the control information acquisition process at the time of commit, in preparation for a primary system failure.

According to the above embodiment, similarly to the embodiment shown earlier, in an information processing system that performs transaction processing using resident tables, it is necessary to acquire all contents of a group of resident tables to a nonvolatile storage medium at the time of a checkpoint. There is no system performance degradation during checkpoints.

Also, in system failure recovery processing, the history information that needs to be read is for completed transactions with input/output units that have not been reflected in the nonvolatile storage medium because they were updated by incomplete transactions, and 1M history information. Since the reading time is reduced and the amount of updating of the resident table using post-update history information is significantly reduced, the CPU processing time required for updating is also reduced. After failure recovery, the process of writing resident table contents to the nonvolatile storage medium only needs to be performed on pages updated in the recovery process, reducing the write time.

Therefore, there is an effect that the system recovery time is shortened.

For example, under the same assumption as in the embodiment shown above, let us assume that when a system failure occurs, the number of transactions in which there are pages that are not reflected in the nonvolatile storage medium because they were updated by unfinished transactions is 5; ,
Assuming that the history information position of the earliest written transaction in the history information file is 50 transactions ago, in this embodiment, the amount of history information and reading time required for system failure recovery are 50/1000 = 1/20 Therefore, updating the contents of the resident table stored in the non-volatile storage medium is (2x 5)/(4xt000/4) = 1/10
It decreases to about 0.

In the above embodiment, when determining a valid control information area, the later information writing time recorded in the control information area is used, but the control information map 65 is recorded in the process completion information and the recovery process is performed. When doing so, the history information file 13A may be read in the reverse order of history information acquisition, and an effective control information area may be determined based on the control information map 65 recorded in the first discovered process completion information.

In addition, in the above embodiment, an example was shown in which the resident table is stored in the main memory, but it goes without saying that the present invention should not be limited to this.

[Effects of the Invention] As described above, according to the present invention, in an information processing system that performs transaction processing using a resident table, there is no need to acquire the entire contents of the resident table to a nonvolatile storage medium at the time of a checkpoint. Therefore, the system performance at checkpoint 1 does not deteriorate, making it possible to design a flexible system, and also significantly reducing the processing time for system recovery processing when a failure occurs. This has the remarkable effect of realizing a resident table content guarantee system that improves overall efficiency.

[Brief explanation of the drawing]

FIG. 1 is an operation flowchart of an embodiment of the present invention, and FIG.
3 is a diagram showing the memory contents, FIG. 4 is a diagram showing a specific example, FIG. 5 is a system configuration diagram showing another embodiment of the present invention, and FIG. 6 is a diagram showing the overall system configuration of the embodiment. 7 is a diagram showing the contents of the memory, FIG. 7 is a diagram showing a specific example, and FIG. 8 is an operation flowchart. 11, IIA: CPU/Memory, 13.13A
= MI history information file, 14.14A: Nonvolatile storage medium, 141: Control information area, 142: Resident table guaranteed area, 21, 21A: Memory, 22: Resident table group, 23: Update page management table, 24: Failure Recovery page bitmap, 25: Pre-update information buffer, 5
1.52° 71-73 Nidoranzaktion, 61: Page map, 62 Nislot usage map, 63: Current information list, 64: History information list, 65: Control information map. Figure 2 Figure 3 L)1 Figure 4

Claims (1)

[Claims] 1. In an information processing system that performs transaction processing using a resident table made resident in memory,
The resident table group is divided into arbitrary input/output units, and if the resident table contents are not stored in a non-volatile storage medium when the system starts, the contents are written to the non-volatile storage medium, and the corresponding contents are written at the end of the transaction. Acquires the pre-update information of the resident table updated by the transaction and the pre-update history information of the input/output unit of another transaction that updated the input/output unit that includes the resident table part updated by the transaction in the history information file. After that, the relevant input/output unit is written to a non-volatile storage medium that can be accessed at high speed randomly, and in the event of a system failure, the contents of the resident table stored in the non-volatile storage medium are read into memory, and the updated part by the unfinished transaction is saved. A resident table content guarantee method characterized by recovery based on pre-update information. 2. In an information processing system that performs transaction processing using a resident table resident in memory,
The resident table group is divided into arbitrary input/output units, and when the resident table contents are not stored in a non-volatile storage medium when the system is started, the contents are written to the non-volatile storage medium. Regarding the input/output unit updated by the transaction at the time of termination,
I/O units that have not been updated due to incomplete transactions are written to the non-volatile storage medium, and input/output units that have been updated due to incomplete transactions acquire their updated history information in the history information file, and in the event of a system failure, the non-volatile storage medium is written. The contents of the resident table stored in the storage medium are read into the memory, and among them, the updated portion of the input/output unit of the completed transaction that has not been reflected in the non-volatile storage medium because it was updated by an uncompleted transaction,
A content guarantee method for a resident table, characterized in that the resident table is updated using history information after the resident table is updated to restore logical consistency of the system.