JPH0738165B2 - Database management method - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a database processing device and, in particular, a database management mechanism, and more particularly to query processing and buffer management based on a query processing procedure, which are suitable for relay regional databases.

[Background of the Invention]

A database management system (hereinafter abbreviated as DBMS), particularly a relay DBMS, processes an external query expressed in a nonprocedural language to determine and execute an internal processing procedure. The main method of conventional query processing is fixed, and statistical information about the database is not considered.
Directly determining a single internal processing procedure based on rules from a purely logical point of view (eg Smith, JM and
Chang, PYT “Optimizing the Performance of a Rela
tional Database Interface "C.ACM Vol.18, No.10, Oct.1
975, pp.568 to 579) and a plurality of candidate processing procedures selected using various statistical information, a method that determines the most suitable one by cost evaluation (for example, Selinger, PGe
t al. “Access Path Selection in a Relational Datab
ase Management System "Proc.ACM-SIGMOD, 1979, pp.23
~ 34). However, the former has a problem in the validity of the rules that are uniformly set, although the load for creating the processing procedure is small, and the latter has a high possibility of giving the optimal processing procedure. There is a heavy load on management and the creation of multiple candidate procedures and their cost evaluation.

Also, in many DBMSs, in order to reduce access costs, a buffer (database buffer) that holds a partial copy of the database stored in the external storage device is used.
It is prepared in the main memory. If the data once fetched to the database buffer as the processing target is left even after the immediate processing is finished, when the data is needed again, the access to the external storage device with the expensive access cost is required. Since the can be omitted, system performance is improved. However, since the capacity of the database buffer is limited, it is necessary to replace (replace) the contents, and the skill of the system affects the performance of the system. Therefore, selection criteria are important for deciding what kind of data should be retained during replacement.

One of the possible selection criteria is the LRU (Least Recently Used) algorithm, which is well known in general buffer storage. This algorithm is based on the assumption that data with a longer time since last access is less likely to be accessed again. However, in database processing, there is a mixture of processing of sequentially accessing database data and processing of randomly accessing database data, in which case the above assumption is often violated. Therefore, in database buffers it is not advisable to rely solely on simple LRU algorithms. There is a conventional technique aiming at solving this problem, which manages the buffer size according to the hit rate, but it is not sufficient yet (Japanese Patent Laid-Open No. 59-5483, G.
M. Sacco et al. “A Mechanism for Managing the Buffe
r Pool in a Relational Database System Using the H
ot Set Model "Proceedings of the 8th International
Conference on Very Large Database, 1982).

[Object of the Invention]

An object of the present invention is to improve the query processing performance by automatically modifying the determination rule of the internal processing procedure for the query processing in the DBMS, and further to reflect the individual internal processing procedures in the replacement control. To improve the efficiency of the database buffer.

[Outline of Invention]

In the query processing according to the present invention, the internal processing procedure to be executed is selected by the first rule base that gives a single internal processing procedure, and this rule base is used for statistical information (system characteristics, system characteristic, Information about database characteristics, processing method characteristics, etc.) is used to select the internal processing procedure. In parallel with this, some internal processing procedures are selected by another rule base for comparison purposes. This rule base also uses the statistical information of the database system, but the internal processing procedure selected by it is not limited to one. All the internal processing procedures selected by the processing of these two systems are subjected to evaluation such as cost evaluation and compared, and the first rule base is modified according to the result. For example, rules are added or modified to provide the lowest cost internal processing procedure. As a result, the system continues to improve the performance in a self-learning manner.

Further, in buffer management according to the present invention, database reference characteristic information (access type and frequency, obtained from the internal processing procedure determined in the query processing optimization process,
The expected number of accesses to each page) is used as buffer control information and is referred to when selecting a buffer to be replaced. For example, under the LRU algorithm, pages that have reached the scheduled access count are replaced with priority. This improves the hit rate and the efficiency of the buffer.

Example of Invention

FIG. 1 shows an outline of an example of a DBMS to which the first invention of the present application is applied. The DBMS 2 includes a system control unit 3 that manages input / output in addition to management and control of the entire system, a database logic processing unit 4 that analyzes an inquiry and generates an appropriate internal processing procedure code, and a generated processing procedure. It includes a database physical processing unit 5 for executing the access to the database 7 and the like according to the code. The input of the inquiry and the output of the response (process result) to the inquiry are performed via the input / output unit 1. The database logic processing unit 4 implements the query analysis unit 41 that analyzes the query, the optimization unit 42 that applies a predetermined rule to the analysis result to determine the optimum processing procedure, and the determined optimum processing procedure. And a code generation / logical process execution unit 43 for performing the related required process.

The rule applied by the optimization unit 42 specifies a single processing procedure to be executed by using various statistical information regarding the database system (system characteristics, database characteristics, processing method characteristics, etc., which will be described later). For example, this rule
Analysis of queries by forming a table consisting of various combinations of predicate selection rate, processing target data size, access path, and access path data size values, and a group of processing procedure information corresponding to each combination one-to-one. From the resulting combination of specific values of the various items, the corresponding processing procedure is determined. A set of such rules is called a rule base, and its management is performed by the rule base management unit 421.

The optimization unit 42 is connected to the cost evaluation optimization unit 6. The cost evaluation optimizing unit 6 optimizes the main part of the query by cost evaluation in parallel with the optimizing unit 42, and compares the result with the cost of the processing procedure created by the optimizing unit 42. , The rule base of the optimizing unit 42 is modified if necessary. As shown in FIG. 2, the cost evaluation optimization unit 6 includes a query analysis unit 6 that performs equivalent conversion and analysis of queries.
1, a processing procedure candidate generation unit 62 that selects some processing procedure candidates from the analysis result, a cost evaluation unit 63 that performs cost evaluation of the processing procedure candidates with respect to the number of external storage accesses, CPU usage time, and the like, and the necessary It includes a processing method management unit 66 that supplies information, a system characteristic management unit 67, a database characteristic management unit 68, and a rule base management unit 69. The processing method management unit 66 manages information regarding available processing functions (logical processing function, hatching function, etc.) and supplies the information to the processing procedure candidate generation unit 62. The system characteristic management unit 67
Information on the characteristics of the computer system (buffer size, speed of CPU and main memory, etc.) is managed and supplied to the cost evaluation unit 63. The database characteristic management unit 68 manages information about database characteristics (various data sizes in the database, physical structure of the database, type and presence of access paths such as indexes and links), and the cost evaluation unit 63.
Supply to. The rule base management unit 69 is the query analysis unit 61.
Manages a rule base for selecting the processing procedure candidate in the processing procedure candidate generating unit 62 in the equivalent conversion in step (1). The rule base for selecting the processing procedure candidates is basically the same as the rule base for the optimizing unit 42 described above, except that there may be a plurality of processing procedures to be selected (similarly. An example of such a rule base is Japanese Patent Application No. 59-173387 (also disclosed in Japanese Patent Laid-Open No. 61-52 ').

The cost evaluation unit 63 also receives the processing procedure determined by the optimization unit 42 via the preprocessing unit 65 and evaluates the cost. The pre-processing unit 65 converts the processing procedure information from the optimizing unit 42 regarding codes and other formats so as to be suitable for the processing in the cost evaluation unit 63. The optimum procedure selection unit 64 determines the minimum cost processing procedure from the evaluation result of the cost evaluation unit 63. The report output unit 70 is for outputting the processes of the processing procedure candidate generation unit 62, the cost evaluation unit 63, and the optimum procedure selection unit 64 as needed, and the line 103 is the various management units 66 to 69. It is for initial setting of.

FIG. 3 is a flowchart of the processing performed by each of the above-mentioned units according to the present invention. An inquiry from the user is given to the DBMS from the input / output unit 1 through the line 100 (see FIG. 3).
0), the system control unit 3 creates information (inquiry identification code, etc.) necessary for managing the processing of the inquiry, and manages the processing thereafter (201). The query analysis unit 41 performs syntax analysis and semantic analysis on the query received from the system control unit 3 (202), and the optimization unit 4 receiving the result.
2 refers to the rule base management unit 421,
The optimum processing procedure based on the statistical information is determined (203).

On the other hand, the same query is also sent to the cost evaluation optimization unit 6, and the query analysis unit 61 performs syntax analysis and semantic analysis on the main part of this query (the part required for optimization by cost evaluation). Further, based on the equivalence conversion rule from the rule-based management unit 69, the conditional expression is appropriately managed and transformed (204). Next, the processing procedure candidate generation unit 62 obtains information on applicable processing methods from the processing method management unit 66, applies the processing procedure candidate selection rules from the rule base management unit 69, and applies some processing procedure candidates. Select (20
Five). The cost evaluation unit 63 uses the system characteristic information from the system characteristic management unit 67 and the database characteristic information from the database characteristic management unit 68 to calculate the cost (external The number of storage accesses, the CPU usage time, etc.) are calculated (at 206), and the optimum procedure selection unit 64 compares the costs of these processing procedure candidates to determine the processing procedure with the lowest cost (207).

The minimum cost thus determined is [A].

At that time, the processing procedure determined (203) by the optimization unit 42 is
The line 101 reaches the preprocessing unit 65 where it undergoes format conversion and is then passed to the cost evaluation unit 63. The cost evaluation unit 63 also performs cost evaluation for this processing procedure in the same manner as described above (at 206). The cost calculated by this is set to [B]. The optimum procedure selection unit 64 is
[A] and [B] are compared (208, 209).

In the case of [A] <[B], it is difficult to say that the processing procedure determined by the optimization unit 42 by referring to the rule-based management unit 421 is optimal. Then, the rule that causes the processing procedure of the cost [A] is transferred from the rule base management unit 69 through the line 102 to the rule base management unit 421, and the rule base is corrected to add this rule. Further, if necessary, the rule base managed by the rule base management unit 69 is also corrected via the line 104 (210). Further, if necessary, various statistical information held in the optimizing unit 42 is added / subtracted and corrected (211).

On the other hand, when [A] ≧ [B], the processing procedure generated by the optimizing unit 42 may be considered to be optimal,
The process (210) is omitted, and only the process (211) is performed.

In any of the cases, the execution of the inquiry process is performed by the normal procedure (212 to 214) according to the processing procedure generated by the optimizing unit 42. According to this method, when [A] is smaller than [B], the query processing is executed by a non-optimal processing procedure. However, it is often uneconomical to perform optimization again after the rule base is modified. Therefore, this is not a problem for practical use. The effect of the rule base modification is exerted in the subsequent query processing. However, the method of re-optimizing after modifying the rule base can also be adopted depending on the design conditions.

It cannot be denied that the load increases due to the addition of the cost evaluation optimization process. However, as the processing history increases, the rule base improves, and the frequency of modification decreases.
Therefore, at an appropriate point in time, the entire rule-based modification process or cost evaluation optimization process can be intermittently suppressed or permanently disabled, and even removed.

If the load does not become excessive, the cost evaluation optimization unit 6
It may be embedded inside the optimization unit 42.

FIG. 4 shows an example of a DBMS to which the second invention of the present application is applied. In the figure, the same reference numerals as those in FIG. 1 indicate the same elements. However, the optimization unit 42 is supposed to incorporate the cost evaluation optimization unit 6 of FIG. Database physical processing unit 5
Includes a buffer control unit 51. Database buffer 8
Consists of multiple buffers and is actually part of the main memory. The database 7 is a relay database, which theoretically consists of several relays, each relay being a collection of rows. The database 7 is always stored in the external storage device,
When an access request is made, a certain range of data blocks (called a page) including the corresponding part are continuously sent to the database buffer 8 and if there is a free buffer, it is stored there. If there is no free buffer, Replaces the contents of the buffer selected according to the algorithm described below. If a page including the data requested for access exists in the database buffer 8, that page is used and access to the external storage device is unnecessary.

The optimization unit 42 uses the statistical information as described above,
Determine the internal processing procedure for processing the query. This internal processing procedure includes a specified relay evaluation sequence, a join condition evaluation method, a query condition evaluation method, etc., and based on these, a reference characteristic for each relay, that is, an access type (random or order). ,), Order, frequency, timing, etc., and the number of pages (guaranteed pages) that should be stored in the buffer for each relay and the expected number of low accesses, which are derived from these, The expected number of page accesses and the like can be estimated. These pieces of reference characteristic information are used for replacement control according to the present invention.

FIG. 5 schematically shows buffer control information managed by the buffer control unit 51. Buffer entry (abbreviated as BE below)
The management information groups 300, 310 and 320 are prepared corresponding to the buffers used. A buffer control block (hereinafter abbreviated as BCB) 330 is a management information group regarding the entire buffer to be used. A relay control block (hereinafter abbreviated as RCB) 340 is a management information group prepared in a one-to-one correspondence with respect to some of the relays to be processed.

The replacement of the buffer is basically controlled by the LRU algorithm. BE300-32 for this LRU control
0 is incorporated into the LRU chain including BCB330. That is, the LRU head pointer 331 of the BCB 330 points to the BE corresponding to the newest access buffer. This is BE300. BE300 LRU forward pointer 301
Next, point to the BE that corresponds to the new access buffer, and so on.
0 and 320 form the LRU front chain in this order.
Conversely, the LRU tail pointer 332 of BCB 330 points to the BE corresponding to the oldest accessed buffer, which is BE 320 in the illustrated example. BE320's LRU backward pointer 322 points to the BE corresponding to the next oldest access buffer, thus BE320,310,300
However, they form the LRU rear chain in this order.

For each relay to which the RCB 340 is assigned, overlapping the above LRU chain, the BE corresponding to the buffer holding the page of that relay is incorporated into each relay LRU chain. That is, the relay LRU head pointer 341 of the RCB 340 points to the BE corresponding to the one having the newest access in the buffer holding the page of the relay, and sets this to BE310.
The relay LRU forward pointer 313 then points to the BE corresponding to the next new access buffer, which is BE 320 in the illustrated example. The relay LRU tail pointer 342 of the RCB 340 and the relay LRU rear pointers 324, 314 of the BE 320, 310 form a relay LRU back chain similar to the LRU back chain starting at the LRU end pointer 332 of the BCB 330. In the example shown, B
The relay LRU forward pointer 303 and the relay LRU backward pointer 304 of the E300 are not used. RCB340 is
It is assigned to relays that are accessed sequentially in a relatively short period of time. For example, in a join operation for relays A and B, if all rows of relay B are sequentially accessed for each row of relay A, RCB 340 is assigned to relay B.

The buffer pointers 305, 315, 325 of the BE 300, 310, 320 point to the corresponding buffers. The page access counters 306, 316, 326 indicate the expected page access times (the access times for the entire page) for the pages held in the corresponding buffers. This value is "0"
Alternatively, it can be a positive integer. For each page to be processed, it is set in an appropriate control block by the initialization process and is replaced at the same time as that page until it becomes "0". "1" is decremented each time is swept from the buffer. It is desirable to keep pages with positive values in the buffer as much as possible because they will be accessed again later. The empty buffer pointer 333 of the BCB 330 points to one of the empty buffers. The entry buffer number 334 indicates the number of buffers (that is, all buffers in use) incorporated in the LRU chain.

The RCB 340 entry buffer number 343 indicates the number of buffers incorporated in the relay LRU chain managed by the RCB. The guaranteed page count 344 holds the number of supported relays that it is desirable to store in the buffer. By keeping the number of pages included in the relay LRU chain at least equal to the guaranteed number of pages, it is possible to prevent a so-called slurring phenomenon (a phenomenon in which an external storage device is accessed each time a data request is made). Low access counter
345 indicates the expected number of accesses of the row of the corresponding relay, and "1" is decremented each time the row of any page of the relay is accessed. If this value reaches “0”, the relay LRU chain managed by that RCB does not need to be maintained and is therefore released.
This release prevents the occupation of the Bahua group by some relays.

In addition to these, a directory (not shown) indicating which page of the database the data held in each buffer is is provided in the buffer control unit 51 as in the normal buffer memory. .

6A to 6E show an example of a buffer control procedure performed by the buffer control unit 51 in the data read (FETCH-DATA) operation. Fig. 6a shows FETCH-DATA control 4
Shows the entire 00. First, the directory is checked (401) to see if the requested page exists in the buffer. If the required page exists in the buffer, it will be accessed. Therefore, it is checked whether the buffer is included in any of the relay LRU chains (402), and if it is included, the corresponding BE is placed at the head of the relay LRU chain and the chain is connected. Is updated (403). However, if the buffer holding the required page is not found, the buffer acquisition (GET-BUFFER) processing 410 acquires the buffer and transfers the required page to the buffer from the external storage device. This processing will be described later in detail.

In this way, it is checked again whether the buffer targeted for this access is included in any of the relay LRU chains (404), and if it is included, the corresponding RCB low access is made. The counter is decremented by "1" (40
5), then the resulting low access counter is "0"
It is checked (406) whether or not: If this is less than or equal to "0", the page access counter of each BE included in the relay LRU chain managed by that RCB is set to "0", and all of the relay LRU pointers are reset to reset the page access counter. The relay LRU chain is released (407). Finally, the BE corresponding to the buffer holding the required page is placed at the beginning of the LRU chain, and the chain is updated (408). However, if the process 404 finds that the buffer in question is not included in any of the relay LRU chains, the process immediately moves to the process 408.

Details of the GET-BUFFER process 410 are shown in Figure 6b. In this process, first, it is checked whether there is a free buffer (411), and if there is a free buffer, one of them is assigned to the new page (415). However, if there is no free buffer, the buffer to be replaced is selected according to the present invention as follows. First, the number of entry buffers 343 and the number of guaranteed pages 344 of each RCB are compared (412). If the number of entry buffers in all RCBs is less than or equal to the guaranteed number of pages, in the LRU chain, if there is a BE with a page access counter of “0”, then the BE that is closest to the end of the chain If there is no BE selected and the page access counter is "0",
The last one in the chain is selected (413). On the other hand, when the number of entry buffers of a certain RCB is larger than the guaranteed page, if there is a BE with a page access counter of “0” in the relay LRU chain managed by that RCB, the end of the chain is among them. Is selected, and if there is no BE with a page access counter of "0",
The last one in the chain is selected (414). After processing 413 and 414, page output (PAGE-OUT) processing, respectively
420 continues, and after these and processing 415, page input (PAGE−
IN) processing 430 is performed, and GET-BUFFER processing 410 ends.

Details of the PAGE-OUT process 420 are shown in Figure 6c. First, it is checked whether the buffer selected for replacement is included in any of the relay LRU chains (421), and if not included, the page held in the buffer is immediately stored in the external storage device. Is output to (42
Four). However, if the problematic buffer is included in any of the relay LRU chains, it will be dealt with.
It is checked whether the page access counter of BE is greater than 0 (422), and if it is not greater than 0, it is as it is.
If it is greater than 0, "1" is subtracted (423), and then the page output process 424 is performed. At this time, the value of the page access counter is transferred to the corresponding control block.

The PAGE-IN process 430 is shown in Figure 6d. First, the required page is transferred from the external storage device to the selected buffer, during which the page access counter value of the page is transferred from the corresponding control block (431). Next, it is checked whether the relay to which this new page belongs should be managed by any of the RCBs (432), and if it is not, it ends the PAGE-IN processing and manages it. If it is something that should be received, the BE corresponding to the buffer that received the page is the corresponding relay LRU.
Placed at the beginning of the chain, the chain is updated (43
3).

FIG. 6e shows a buffer initialization (INIT) process 440. B
In the CB / BE initialization processing 441, BCB, BE, and other control blocks are allocated and their contents are initialized. In the RCB processing processing 442, allocation of RCBs and initialization of their contents are performed, and based on the database reference characteristics obtained from the internal processing procedure determined by the optimization unit 42, the guaranteed page number 344 and low access Counter 345
Is set, and the page access counter value for each page to be processed is set in an appropriate control block. In this embodiment, the initial setting value of the page access counter of the relays that do not form the relay LRU chain is "0".

Next, a comprehensive description will be given by using a specific inquiry example. Two sets of relays "Resort" and "Register" shown in Fig. 7 are prepared in the database. The former is accommodated on pages R11 and R12 and the latter is page R2.
It is supposed to be housed in 1 and R22. From this database, it is desirable to obtain "the name of the tourist spot where you can play golf, the transportation cost is 3,000 yen or more and 5,000 yen or less, and the seafood is a specialty and its prefecture name". Be done. This request is expressed by the following query statement.

Select Resort. Name, resort. Prefecture name from resort, cashier where cashier. Name = 'Golf' and (Resort.Fare> = 3 and Resort.Fare <= 5) and Resort. Specialty = 'seafood' and resort. Name = cashier. Resort name In response to this inquiry, the processing procedure determined by the optimizing unit 42 of the DBMS 2 is roughly as follows.

1) From the relay "resort" to the column "fare"
Find one row with a value of "3" or more and "5" or less and the content of the column "specialty" is "seafood".

2) Scan the relay "Register" and select 1) above.
The row of the relay “Register” that has the same content as the column “Name” of the row of the relay “Resort” retrieved in step 4 is retrieved.

3) The relay “Register” found in 2) above.
If the row's column "name"is'golf', save the relay "resort" row found in 1) above, otherwise discard it.

4) The above 1) to 3) are repeated until the relay “resort low” runs out.

5) Extract the column “name” and “prefecture name” of the row stored in 3) above.

Based on such a processing procedure, the page access targets are R11, R12, R21, and R22, the number of buffer surfaces used is "3", and the formation of the relay LRU chain in the relay "regulator" is performed. RCB340 is allocated for this purpose, the guaranteed page number is “2”, the row access counter value is “6”, and the page access counter value is “4” for the relay “Register” and the same. It is assumed that it has been decided that the "resort" is "0".

Due to the processing of this query, page access is R11, R12,
The order of the buffer control executed by the buffer control unit 51 in the case of performing R21 and R12 in this order will be described with reference to FIGS. 5 and 6.

Step 0: BCB330, BE300,310,320 by buffer initialization processing 440
Allocate and initialize control blocks such as (44
1) Then, the RCB 340 is assigned to the relay “registerer”, the guaranteed page number 344 and the low access counter 345 are set to “2” and “6”, respectively, and each page of the relay “registerer” R21. Page access counter value "4" is set in the control block corresponding to R22, R22 (442). Each page of relay "Resort" R11, R
The control blocks corresponding to 12 are all set to "0".

Step 1: First, when access to R11 is requested, the FETCH-DATA processing 400 (Fig. 6a) is started. First, the presence or absence of a buffer holding R11 was checked (401), and since it does not exist, G
The ET-BUFFER process 410 (Fig. 6b) is activated. Therefore, it is checked whether or not there is a free buffer (41
1), it turns out to exist and there is 1 free buffer
The face is won (415). This is a buffer corresponding to BE300.

Then, the PAGE-IN process 430 (FIG. 6d) is activated, and the requested page R11 is transferred from the external storage device to the acquired buffer (corresponding to BE300) and the corresponding page counter value is also transferred. “0” is the page counter 30
Set to 6 (431). Then it's relay LRU
It is checked whether or not the page is to form a chain, that is, whether or not the page is a relay “registry” (43
2), otherwise, the PAGE-IN process 430 ends, which ends the GET-BUFFER process 410. Returning to FIG. 6a, it is checked whether or not the BE300 is included in the relay LRU chain controlled by the RCB340 (404), and since it is not included, the process 408 of immediately placing the BE300 at the head of the LRU chain is performed. Done.

At this point, the LRU start pointer 331 and LRU of BCB330
The tail pointers 332 both point to the BE 300, the entry buffer number 334 is “1”, and the page access counter 306 of the BE 300 is “0”. There is no change in the contents of RCB340, and the relay LRU chain has not been formed.

Step 2: Next, when access to R21 is requested, the FETCH-DATA processing 400 is started. Since there is no buffer holding R21 (401), the GET-BUFFER process 410 is activated. Since there is a free buffer (411), the free buffer is 1
The face is won (415). It is assumed that this is a buffer corresponding to BE320.

Therefore, the PAGE-IN process 430 is called, R21 is stored in the buffer corresponding to BE320, and the corresponding page access counter value "4" is set in the page access counter 326 (431). R21 is a relay "Register"
Is a page that belongs to and should form a relay LRU (4
32). Therefore, BE320 is placed at the head of relay LRU chain controlled by RCB340 (433). Returning to FIG. 6A, since the BE 320 is included in the relay LRU chain (404), the low access counter 345 is decremented by “1” (405), and as a result, the value of the low access counter 345 becomes “5”. ", Which is greater than" 0 "(406). Therefore,
Immediately the LRU chain is updated and BE320 is placed in front of it (408).

At this point, the LRU front chain [→ BE320 → BE300] starting at the LRU head pointer 331 and the LRU rear chain [→ BE300 → BE320] starting at the LRU end pointer 332 of the BCB330 are formed. The number of entry buffers 334 is “2”, the page access counter 306 is “0”, and the same 326 is “4”.
Is. Also, RCB340 relay LRU head pointer 3
41 and the same end pointer 342 both point to BE320, the entry buffer number 343 is "1", and the low access counter 345
Is "5".

Step 3: Next, when access to R22 is requested, the buffer corresponding to BE310 is obtained in process 415 in the same order as in step 3, R22 is stored in the buffer in process 431, and the buffer is stored in process 433. BE310 is placed at the head of the relay LRU chain, operation 405 decrements the low access counter 345 by “1”, and operation 408 BE.
310 is placed at the beginning of the LRU chain.

At this point, the LRU forward chain starting from the LRU head pointer 331 of the BCB330 [→ BE310 → BE320 → BE300] and the LRU
LRU rear chain starting at the end Poita 332 [→ BE300 → BE3
20 → BE310] is formed. Number of entry buffers 3
34 is “3”, page access counter 306 is “0”,
Both 316 and 326 are "4". On the other hand, in RCB340, the relay chain LRU front chain starting from the relay LRU head pointer 341 [→ BE310 → BE320] and the relay chain LRU rear chain starting from the relay LRU end pointer 342 [→ BE320 → BE310] are formed. ing. The entry buffer number 343 is "2" and the low access counter 345
Is "4".

Step 4: Next, when the access to R12 is requested, the GET-BUFFER processing 410 is started in the same manner as the above-mentioned various steps, and it is checked whether or not there is an empty buffer (411). This time there is no free buffer. Therefore, the number of entry buffers of RCB340
343 is compared with the guaranteed buffer number 344 (412). Both are equal to "2". Therefore, by the process 413, BCB33
In the LRU chain managed by 0, the search for BE whose page access counter is “0” is started from the end of the chain, and the corresponding BE 300 is found.

Therefore, the PAGE-IN processing 420 is called, and since this BE300 is not included in the relay LRU chain (421), the contents of the corresponding buffer are immediately output (424), and PAGE
-Proceed to IN processing (430), and R12 is input to the buffer corresponding to BE300 (431). This page is a page of "Resort", and should not be included in Relay LRU Chain (432), so immediately PAGE
-IN processing ends, and so does GET-BUFFER processing. And because this BE300 is not included in the relay LRU chain (404), leave the RCB340 as it is and B
The E300 can be placed at the beginning of the LRU chain.

At this point, the LRU front chain [→ BE300 → BE310 → BE320] starting from the LRU head pointer 331 of the BCB330 and the LRU
LRU backward chain starting from the end pointer 332 [→ BE320 → B
E310 → BE300] are formed. The number of entry buffers 334 is “3”, the page access counter 306 is “0”, and both 316 and 326 are “4”. On the other hand, in the RCB340, a relay chain LRU front chain starting from the relay LRU head pointer 341 [→ BE310 → BE320] and a relay chain LRU rear chain starting from the relay LRU end pointer 342 [BE320 → BE310] are formed. . The entry buffer number 343 is "2" and the low access counter 345
Is "2". FIG. 5 shows the various chains at this point.

If the number of guaranteed pages is "1" by changing the conditions in the above-described example, GET-BUFFE in step 4 is used.
At the time of determination 412 in the R process 410, the entry buffer number 343 is “2”, which is larger than the guaranteed page number “1”. As a result, instead of the processing 413, the processing proceeds to the processing 414 and the relay LR
A BE having a page access counter of “0” is searched in the U chain [→ BE310 → BE320]. However, since the page access counters 316 and 326 are both "4", the BE3 at the end is
20 is selected. Therefore, in the PAGE-OUT processing 420, the page access counter 326 of BE320 is decremented by "1" and becomes "3" (421 to 423), and the buffer corresponding to this BE320 is the target of replacement (424,431). Becomes

Further, if the initial setting value of the low access counter 345 is "2", the low access counter 345 becomes "0" as a result of the processing 405 in step 3, and therefore, the relay processing is executed by the processing 406 and 407. LRU chain is released and page access counters 316 and 326 are both "0"
Is set to. In this case, when performing Step 4,
Since there is no relay LRU chain, processing 412,4 regardless of the value of the entry buffer number 343 of RCB.
By 13, the replacement target buffer is selected from the LRU chain.

The buffer management method of the present invention is not limited to the combination of the rule using the statistical information and the cost evaluation, but any other method (for example, the cost) can be used as long as an internal processing procedure that gives appropriate database reference characteristic information can be obtained. It can also be applied to a DBMS that performs optimization processing by evaluation only, high-status, cost evaluation and high-status combination, etc.).

Any of the inventions can be realized via a software system of a large-scale computer, or can be realized via a composite processor system in which a dedicated processor is prepared for each unit.

〔The invention's effect〕

According to the query processing method of the present invention, the rule base for determining the query processing procedure is supplemented and maintained so as to be suitable for the system usage environment by repeatedly improving the dynamics every time a new query is processed, Therefore, a significant performance improvement can be expected compared to the fixed rule.

Further, according to the buffer management method of the present invention, the replacement control of the buffer is performed in a procedure adapted to the processing procedure of each inquiry, and as a result, the efficiency of the buffer is improved,
As a result, the inquiry processing can be performed at higher speed and lower cost than ever before.

[Brief description of drawings]

1 is a block diagram of a database management system to which an embodiment of the first invention is applied, FIG. 2 is a block diagram of a cost evaluation optimizing unit in FIG. 1, and FIG.
FIG. 4 is a flow chart of query processing performed by the database management system shown in FIGS.
FIG. 5 is a block diagram of a database management system to which an embodiment of the second invention is applied, and FIG. 5 is a schematic diagram of buffer control information managed by the buffer control unit in FIG.
FIG. 6 is a flow chart of the buffer control performed by the buffer control section in FIG. 4, and FIG. 7 is a schematic diagram of the relay used for explaining the embodiment. 2 ... Database management system, 4 ... Logical processing unit,
41 …… Inquiry analysis section, 42 …… Inquiry processing optimization section, 42a
...... First rule base management unit, 43 ...... Code generation / logical process execution unit, 5 ...... Physical processing unit, 51 ...... Buffer control unit, 6 ...... Cost evaluation optimization unit, 69 ...... Second rule base Management department, 7 ... Database, 8 ... Database buffer.

Claims (6)

[Claims] 1. Statistical information of a database system, which is performed in a database processing device that analyzes an inquiry from a user to a database, determines an internal processing procedure for obtaining information necessary for answering from the database, and executes the internal processing procedure. A first selection process of selecting a single internal processing procedure to be executed by a first rule base that gives a single internal processing procedure from the analysis result of the query, statistical information of the database system, and the analysis result of the query. The second selection process of selecting at least one pseudo internal processing procedure by the second rule base that gives an internal processing procedure without restriction on the number and the internal processing procedures selected in the first and second selection processes are evaluated and compared. Based on the contents of the second rule base according to the result of the evaluation comparison. And a step of modifying, database management method. 2. The statistical information according to claim 1, which is information on a system characteristic indicating a processing capacity of a database system, a database characteristic indicating a database configuration, and a processing method characteristic indicating an available processing function. A database management method that includes evaluation comparison of the internal processing procedures with respect to the cost of executing each internal processing procedure. 3. A database inquiry is analyzed to determine an internal processing procedure for obtaining required information from the database, and a partial copy of the database data stored in an external storage device is temporarily copied. The process of determining the optimum internal processing procedure from the statistical information of the database system and the analysis result of the query, which is performed in the database processing apparatus executed by using the buffer group stored in, and obtained from the determined internal processing procedure. A database management method comprising: a step of setting database reference characteristic information indicating a mode of database access as buffer control information; and a replacement control step of controlling replacement of the buffer group using the set buffer control information. 4. The database according to claim 3, wherein the database is a relay regional database including a plurality of relays, each relay comprising a plurality of rows, and between the external storage device and the buffer group. The data transfer is performed in units of pages that are data blocks of a predetermined size, and the replacement control process uses a buffer that holds the pages of relays that are sequentially accessed in a relatively short period among the target relays. Relay LR formed corresponding to each relay
The process of incorporating into the U chain, the process of incorporating all buffers holding some pages into a single LRU chain, and the number of pages included in any of the relay LRU chains is included in the buffer control information in advance. Database that includes selecting a replacement target from buffers included in the relay LRU chain when the value is greater than the specified value, or selecting a replacement target from buffers included in the single LRU chain otherwise. Management method. 5. The buffer control information according to claim 4, wherein the buffer control information includes an expected page access count for each page, and the replacement control process corresponds to the page access count corresponding to each page replacement. In the process of selecting a replacement target, a page having a page access count of 0 is preferentially selected in the process of selecting a replacement target. 6. The buffer control information according to claim 5, including the expected number of row accesses for a page included in each of the relay LRU chains, and the replacement control process of each relay LR.
A database management method comprising the step of decrementing the corresponding row access count by 1 each time a page included in the U chain is accessed, and releasing the relay LRU chain when the row access count becomes 0.