JPH07105060A - Record access management method for database management system - Google Patents

Abstract

PURPOSE:To restrict the number of records stored in a page, to suppress the contention of page exclusion, to obtain exclusive contention performance equal to a record exclusive system through the use of a page exclusive system and to prevent the application of page turning-out by LRU management accompanying the lowering of access frequency. CONSTITUTION:Record access management 8 consists of the four programs of the correspondence 21 of the record and a page, page management 22 page exclusion 23 and LRU management 24, and of a buffer area 26. The hold management of the record stored in the page included in a data set 41 on a disk device 4 and the exclusive control among the records are performed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a record access management system in a database management system having a page exclusion control mechanism for records.

[0002]

2. Description of the Related Art In a database management system, a data operation unit is a record (also called a tuple). The database management system processes operation requests (called transactions) for a series of data from a plurality of user programs in parallel. The database management system controls inter-transaction exclusive access control of records, which is an operation unit, mapping between records and pages, which is an input / output unit of an external storage device, management of pages in memory, and eviction management when free space is insufficient. Includes record access management.

A transaction in a database management system requires exclusive control for a record being updated in order to maintain data consistency while continuing. In response to this request, the database management system performs exclusive control so that the record being updated by a certain transaction is not referred to from another transaction until the completion (called commit) of the transaction.

"J. Martin's DB2" Nikkei BP (1
(Published January 991) As shown in Chapter 19 Lock Function, in the conventional database management system, the minimum unit of exclusive control is a page.

The method of storing the record in the page is described in "An Introduction to Database Systems Volume 1", published by Addison Wesley (An.
Introduction to Database Systems Volume 1 Fifth Ed
ition, Addison-Wesley (1990)). Also, for the method of managing the empty space of each page, see "Transaction Progressive; Concept and Techniques", published by Morgan Kaufmann (Transaction).
Processing: Concept and Techniques, Morgan Kaufman
1993, (pp757-760)).

On the other hand, in a benchmark job assuming banking business used for performance evaluation of online systems,
Each transaction has a hierarchical relationship of a branch, a teller, and an account, and the lower the level, the larger the number of records in the database. In this case, the branch records have a very high probability that different transactions access different records within the same data page at the same time. Also, this probability is low for account records and somewhat high for cashier records. When such a benchmark job is executed in a system that performs page-based exclusion, access conflict occurs for records such as branch records that have a high probability of simultaneous access by different transactions to the same data page, causing performance degradation. Become.

As a countermeasure against this, "Transaction Recovery Method Supporting Fine Granularity Looking and Partial Rollbacks Using Right Ahead"
"Logging" AMC MDS (ARIE
S: A Transaction Recovery Method Supporting Fine−
Granularity Locking and Partial Rollbacks Using Wr
ite-Ahead Logging, CM TODS) Vol.17 No.1 (1992) pp
94-162), a system in which a database management system performs exclusion on a record basis has been announced.

[0008]

When the record-based exclusive control method is introduced into the conventional page-based exclusive control system database management system, the following problems occur.

(1) The record exclusive control method is more complicated than the page exclusive control method. Since different transactions update different records in the same page at the same time, information commonly updated such as record management information in the page. Management is complicated. Further, the free space in the page released by a certain transaction can be immediately used by another transaction, and thus may be used by another transaction. At this time, if the original transaction ends abnormally and you try to roll back the update processing, there is no space, so the record cannot be restored in the original page. Therefore, it is necessary to recover the same record on another page. On the other hand, in page exclusion, the free space in the page remains until the transaction is committed.
This problem does not occur because it is not released to other transactions.

(2) It is difficult to shift from the page exclusive control method to the record exclusive control method The conventional database management system controls other parts on the premise of the page exclusive control method. In particular, the failure recovery process is significantly changed when the record exclusive control method is used, and thus the amount of modification is enormous.

[0011]

[Means for Solving the Problems]

(1) A computer system including a processor, a memory, and an external storage device. As a part of a database management system on the memory, records are associated with pages that are units of input / output of the external storage device, page management, and page exclusion. It includes a record access management program consisting of control, and the record access management associates a record with a new page according to a predetermined upper limit of the number of records per page, in response to a record addition request. If the page is not in the memory, the management reads the page from the external storage device, then adds the record to the page, and when the record is exclusively occupied or released, the correspondence between the record and the page is the page corresponding to the record. Page exclusion, the page exclusion is the exclusive occupation or release of the corresponding page. Record access management system to perform.

(2) In the system of (1), the record is divided into a plurality of groups (normally set corresponding to the relation), and only the records belonging to the same group are associated with the same page. A record access management method characterized by setting the maximum number of records per page for each group.

(3) In the system of (1), the record access management is the LRU management for determining the order of expelling the records read in the memory to the external storage device from the oldest used page (called LRU page). The record access management method is characterized in that a specific page is excluded from the ejection determination of the LRU management.

(4) In the system of (3), the record access of (2) is introduced, and the page corresponding to the group designated in advance is excluded from the target of the ejection judgment of the LRU management. Management method.

(5) In the system of (2), LRU management for determining the order of ejecting records to the external storage device is added to the record access management, and a specific part on the memory is page-read in advance for each group. A record access management method characterized in that the LRU management determines the page to be flushed to the external storage device for each group.

[0016]

With respect to the problem that a plurality of records in the same page are simultaneously accessed by different transactions, by setting an upper limit on the number of records to be stored in the same page, this contention is suppressed to a certain probability or less. As a result, the total number of pages is slightly increased, but the number of pages on which access is concentrated originally is small, and even if it is increased slightly, the total number of pages is hardly affected.

On the other hand, in such an intensively accessed page, under the LRU buffer replacement method,
It stays in the buffer all the time, and the number of readings from the external storage device is extremely small. However, in the method of setting the upper limit of the number of records in a page of the present invention, the access frequency to this page is reduced. Therefore, unless the buffer size targeted for LRU is increased in inverse proportion to the decrease in the number of records per page, Will be kicked out of the buffer. Such an increase in the overall buffer size, including infrequently accessed pages, increases the memory requirement for the buffer and is not practical. Therefore, it is necessary to use the following measures together to protect the page from LRU eviction in response to the limitation on the number of records in the page.

By excluding the page in which the number of records in the page is limited from the target of the LRU, the page is made resident in the buffer and the increase of the buffer size is unnecessary.

Further, a buffer is prepared for each group to which an in-page record belongs, a certain buffer space is guaranteed for a page corresponding to a limited number of in-page records, and ejection from the buffer is prevented.

[0020]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a block diagram of the system of this embodiment. The processing device 1 is connected to the disk device 4. The processor 1 has a processor 2 and a memory 3, and the memory 3 has a database management system 5 and an operating system 12.

The database management system 5 comprises a user program 6, relation access management 7, and record access management 8. The user program 6 executes a transaction, which is a unit of processing, by the processes 11-1 to 11-k, and issues an access request 9 (described in SQL language, for example) in the relation unit to the relation access management 7. And get the result as a relation.

The relation access management 7 converts the access 9 for each relation from the user program 6 into the access 10 for each record, requests the record access management 8, obtains the result as a record, and converts it to a relation. , And returns to the user program 6.

The operating system 12 inputs / outputs records to / from the disk device in response to a request for record access management (performed in the form of a system call), waits for a process, and restarts the process (similarly to the system). Done in the form of calls).

FIG. 2 shows details of the record access management 8 and the disk device 4 logically connected to it, which is the object of the present invention. The record access management 8 includes record and page association 21, page management 22, and page exclusion 2
3, LRU management 24, and a buffer area 26. The disk device 4 has a data set 41 including a plurality of pages.

FIG. 3 shows details of the data set 41.
The data set 41 includes page 0 (31-0), page 1
(31-1), page n (31-n). Page 0 (31-0) is a system area for storing page management information and includes a group table 32 and an in-page free size map 33. The group is usually set corresponding to the relation which is the operation target of the relation access management 7.

FIG. 4 shows the details of the group table 32. The group table 32 includes a group column 43 that stores a group identifier, a size column 44 that records the number of pages of the group, an upper limit column 45 that defines the upper limit of the number of records per page of the group, and a start page that records the start page of the group. Column 46, L that specifies whether the page of the group is to be the target of LRU eviction
It comprises an RU exclusion column 47. In the example of FIG. 4, group 1 uses 20 pages, the upper limit of the number of records per page is between pages 1 and 20 on the disk, and it is not subject to LRU eviction (LRU exclusion = Y Indicated by
Is. The group 999 is a special group of the system consisting of empty pages, which is not assigned to a group usable by the user. The upper limits of group 3 and group 999 are set to the maximum allowable value 999.

FIG. 5 shows details of the in-page empty size map 33. The in-page empty size map 33 includes a page number column 51 that defines a page number and an empty size column 52 that indicates an empty size of a page. Figure 5
In this example, page 1 still has a vacancy of 1000 bytes, and page n has a vacancy of 4000 bytes. Page 0 is a system area, and the free size is set to 0 bytes.

FIG. 6 shows details of the page 31 other than the system area. At the end of the page, there is a slot 62 containing a pointer to a record, slot 63-1 is record 64-1, and slot 63-2 is record 64-.
Pointing to 2.

FIG. 7 shows details of the record identifier 71 for identifying the record. The record identifier 71 is composed of a page number 72 and a slot number 73.

FIG. 8 shows the details of the buffer area 26. The buffer area 26 is an area 0 (81-
0) to area m (81-m) and area control block 0 (82-0) to area control block m associated with them
(82-m) and a record access control lock 89 that performs exclusion for serialization of record access control processing.

Area 0 is an area dedicated to reading page 0 of the data set 41 of the disk device, and before the database management system starts its processing.
The contents of page 0 are read in advance from the disk device. Further, the processing of the database management system is interrupted, and the data is written back to page 0 before the power of the system is turned off. Area 1 (81-1) to area m (81-
m) is an area for reading a page used by the user program.

Area control block 0 (82-0) to area control block m (82-m) are control blocks for managing the usage status of the area. The area control blocks 82-0 to 82-m include time stamps 83-0 to 83-m indicating the time of the last use, and LRU exclusions 84-0 to 84-indicating whether or not to exclude the LRU from purging.
m, a disk page number 85-0 to 85-m indicating the page number of the corresponding disk, an exclusive process identifier 86-0 to 86-m indicating the process in the exclusive occupation state,
Group 87-0 to 87-m indicating the group to which the read page belongs, exclusion-waiting process identifier list 8 including a list of identifiers of processes waiting for exclusion
8-0 to 88-m are stored.

Disc page numbers 85-0 to 85-m
Also serves as an empty flag of the corresponding area, and is set to -1 when the area is empty. When the database management system starts its processing, the disk page numbers of the area control blocks of areas other than area 0 are all set to -1. In the area control block of area 0, as shown in FIG. 8, the disk page numbers are 0 and L.
RU exclusion 89 set to Y (yes) and time stamp 8
3 is set to the maximum value of time 999 to prevent eviction by LRU. Further, in the example of FIG. 8, area 1 holds page 1 belonging to group 1 (group 87-
1, disk page number 85-1), the last access is time 101 (time stamp 83-1), is not subject to LRU eviction (LRU exclusion 84-1), and exclusive occupation is not performed (exclusive process identifier). 86-1) is shown.

FIG. 13 shows an external interface when a request for record access management is made from the outside. Hardware register R1 (1 when request is made from outside
31) is an address list 132 consisting of four addresses
Including the address of. Each entry of the address list 132 includes a fixed length request code 133, a record identifier 134, a record length 135, and an address of the record passing area 136. The record has a variable length, but is stored in the range of the length of the record length sentence from the beginning of the record passing area. However, the parameters in these areas are not always used. Table 1 shows the request types and the parameters that are effective as input and output at that time.

[0035]

[Table 1]

The request code always contains valid contents. For example, in addition requests, the other valid inputs are the record length and the record, and the valid outputs are the record identifier.

The process of associating a record with a page will be described with reference to FIGS. 9 and 3. The record-based access 10 performed by the processes 11-1 to 11-k through the relation access management is accepted and processed by the records in the record access management 8 and the page association 21. There are five types of record-based access: record addition, deletion, search, update, and record exclusive release. A record is added by designating a group to which the record is to be added, and a record identifier is designated for other items. When adding or updating a record, the contents of the record to be added or updated are passed along with the access in record units. In associating a record with a page, first, the record access management lock is occupied (block 91). This process is performed by setting the record access management lock 89 to an occupied state by, for example, a Compare and swap command.

If the record access management lock is already occupied by another process, the process loops and waits for its release. Hereinafter, the same processing is performed for all occupation of the record access management lock.

Next, it is checked whether the record unit access is a record addition request (block 92). If it is a record addition request, the page of the designated group is searched for the entry of the page free size map from the pages of the designated group. Find (block 910). Here, an entry is searched for which is the free size of the entry> the length of the record to be added and the number of records of the entry <the upper limit of the group to which the record to be added belongs. If not found, an error occurs (block 91).
1).

Next, page management is called to read a page in the area (block 912), a record is added to the page (block 913), and the empty size 52 of the entry corresponding to the page in the page free size map is read. The changed page size is subtracted, 1 is added to the number of records (block 914), a record identifier is generated from the page number and slot number, and is used as a return value to the relation access management (block 915). Finally, the record access management lock is released (block 99), and the process returns to the relation access management (block 916).

If it is not a record addition request, the page number is extracted from the record identifier (block 93) and it is checked whether it is a page exclusion release request (block 9).
4). In the case of a page exclusion release request, page exclusion is called to release the page exclusion (block 98), and block 9
Go to 9. If it is not a page exclusive release request, page management is called to request a page on the area (block 9
5) Apply any one of deletion, search, or update to the record in the page (block 96), and if the record size changes, update the free size of the page entry of the page free size map, and if delete, record Add 1 to the number (block 97) and go to block 99.

The process of the page management 22 will be described with reference to FIGS. 10 and 2. The page management 22 is called from the record and the page association 21 and places a specified page on the area to perform exclusive occupation. First, the disk page number of the area control block is searched (block 101), and it is checked whether or not there is an area including the designated page (block 102). If there is an area including the specified page, the page exclusion 23 is called to exclusively occupy the area (block 104), the LRU management 24 is called to change the access order of the area to the latest area (block 105), Return (block 1015).
If there is no area including the specified page, it is checked whether or not there is an empty area (block 103). If there is, an exclusive page 23 is called to exclusively occupy that area (block 106), and the record access management lock is released ( Block 107) and enter a page into the designated area using the operating system 12 (Block 10).
8) Re-occupy the record access management lock (block 109) and set the disk page number, LRU exclusion, and group of the area control block (block 10).
10) Go to block 105. If there is no free area, the LRU management is called to determine the area containing the page to be ejected from the area (block 101).
1) The page exclusive 23 is called to exclusively occupy the area (block 1012), the record access management lock is released (block 1013), and the contents of the designated area are written to the disk device as a page by the operating system (block 1014), Go to block 108.

The processing flow of the page exclusion 23 will be described with reference to FIGS. 11 and 2. The page exclusion 23 is called from the record and the page association 21 and the page management 22 and exclusively occupies and releases the page and the area storing it. First, it is checked whether it is an area occupation request (block 111). If not, 0 is checked as the exclusive process identifier of the designated area (block 1).
12). If it is 0, the process identifier of the process being executed is set in the exclusive process identifier of the designated area (block 118), and the process returns to the caller (block 11).
15). If the exclusive process identifier is not 0, compare it with the process identifier of the running process (block 1
18), if equal, return to caller (block 1115). If not equal, the area is occupied by another process. At this time, the process identifier of the process being executed is added to the exclusion-waiting process identifier list in the specified area (block 114), the record access management lock is released (block 115), and the process is placed in the waiting state by the operating system (block). 1
16). When releasing the resource, the process is restarted from the waiting state and re-occupies the record access management lock.
(Block 117) and the process from block 112 are repeated. If it is not an area occupation request, it is a page release request, so the area control block is searched to find the area containing the specified page (block 119). It is checked whether the exclusive process identifier of the found area matches the process identifier of the executing process (block 1110), and if they do not match, an error occurs. If there is a match, the exclusive process identifier of the found area is set to 0 (block 111
1) Check if the process identifier list exists in the exclusion waiting process identifier list of the found area (block 112), and if not, go to block 1115. If so, all processes included in the exclusion waiting process identifier list of the found area are restarted (block 11).
13). Next, the exclusion waiting process identifier list of the found area is cleared (block 1114).

Using FIG. 12 and FIG. 2, LRU management 2
The processing flow of No. 4 will be described. The LRU management 23 is called by the page management 22 and holds the access order of areas and selects the oldest accessed area.
First, it is checked whether the request is to update the area access order to the latest (block 121), and if so, the current time is set in the time stamp of the designated area (block 121), and the process returns to the caller (block 12).
4). If the area access request is not updated, the area control block is searched to find the area with the oldest time stamp that satisfies the following conditions. The conditions are exclusive process identifier = 0 and LRU exclusion = N (block 123). Then, the process returns to the caller (block 123).

The second embodiment of the present invention will be described below. Figure 1
4, FIG. 15 and FIG. 16 show parts different from the previous embodiment. The other parts are the same as in the previous embodiment.

FIG. 14 shows the difference in the initial setting of the area control block. In this embodiment, the area is divided into groups in advance. In the initial setting of the area control block in FIG. 14, the area 1 is divided into the group 1, the area 2 into the group 2, and the areas 3 and 4 into the group 3. Also, LRU exclusion is not used in this embodiment.
Area 0 is a special group 9 as a system area.
It is set to 99.

FIG. 15 shows a block 103 and a block 101 which are different from those of the first embodiment in the flow of page management processing.
1 is shown. In block 103, the empty area is checked not in the entire area but only in the area assigned to the designated group. In block 1011, LRU management is called to determine the area including the eviction page from the area of the designated group.

FIG. 16 shows a modified LR corresponding to this.
A block 123 of U management is shown. The point that the area control block is searched to find the oldest area with the time stamp that satisfies the following condition is the same, but the second condition is changed to group = designated group. As described above, pages are swapped for each group, so if you allocate enough area to a group that sets an upper limit on the number of records in a page, the page will be stored in memory regardless of other groups' areas being expelled. You can leave it in.

[0049]

By limiting the number of records that can be put in the same page, it is possible to suppress the contention of record access in the system where page exclusion is performed to the same level as the system where record exclusion is performed. In addition, the problem that the page is liable to be evicted by the LRU due to the decrease in page access frequency that occurs along with it is specified that the LRU eviction target is specified at the same time as the upper limit of the number of records is set, or the group corresponding to the page grouping is set. It was solved by applying each LRU.

[Brief description of drawings]

FIG. 1 is a block diagram of the system of the present invention.

FIG. 2 is a block diagram of record access management and a disk device.

FIG. 3 is an explanatory diagram of a data set.

FIG. 4 is an explanatory diagram of a group table.

FIG. 5 is an explanatory diagram of a page free space map.

FIG. 6 is an explanatory diagram of a page.

FIG. 7 is an explanatory diagram of a record identifier.

FIG. 8 is an explanatory diagram of a buffer area.

FIG. 9 is a processing flowchart of associating a record with a page.

FIG. 10 is a flowchart of a page management process.

FIG. 11 is a flowchart of a page exclusion process.

FIG. 12 is a flowchart of a process of LRU management.

FIG. 13 is an external interface block diagram of record access management.

FIG. 14 is an explanatory diagram of initial setting of an area control block.

FIG. 15 is a flowchart of a page management process.

FIG. 16 is a flowchart of a process of LRU management.

[Explanation of symbols]

4 ... Disk device, 8 ... Record access management, 21 ...
Record and page association, 22 ... Page management, 23 ...
Page exclusion, 24 ... LRU management, 26 ... Buffer area.

Claims (5)

[Claims] 1. A computer system comprising a processor, a memory, and an external storage device, wherein a record is associated with a page, which is a unit of input / output of the external storage device, as a part of a database management system on the memory. management,
A record access management program including page exclusive control is provided, and the record access management is such that, in response to a request to add the record, the record and the page are associated with each other according to a predetermined upper limit of the number of records per page. If the page is not in the memory, the page is read from the external storage device, the record is added to the page, and the record is exclusively occupied or released. The record access management method is characterized in that the record and the page are associated with each other to obtain the page corresponding to the record, and the page exclusion is to exclusively occupy or release the corresponding page. 2. The record access according to claim 1, wherein the record is divided into a plurality of groups, only records belonging to the same group are associated with the same page, and the upper limit of the number of records per page is set for each group. Management method. 3. The LRU management for determining the order of expelling the records read into the memory to the external storage device as the order from the LRU page that has been used the oldest is added to the record access management according to claim 1. A page is a record access management method that excludes pages from the ejection determination of the LRU management. 4. The record access management method according to claim 3, wherein a group of the records is introduced, and pages corresponding to a group designated in advance are excluded from the target of eviction determination of LRU management. 5. The LRU management for determining the order of ejecting records to an external storage device according to claim 2, is added to the record access management, and a specific part on the memory is allocated in advance as a page read area for each group.
LRU management is a record access management method that determines pages to be flushed to an external storage device for each group.