JPS59146339A - Information retrieving system - Google Patents

Abstract

PURPOSE:To perform key adding and deleting processings in a high speed by performing a parent node searching processing in a high speed in a balanced tree where there is a probability that the update processing is required up to a parent node in case of addition or deletion of a key. CONSTITUTION:With respect to a balanced tree such as a B<+>-tree or the like, it is necessary to perform the tracing-back processing from the root to a leaf node, where a key should be put or is already put, before addition or the deletion of the key. At this time, node numbers from a root node 201 to leaf nodes 205-208 are stored on a table 7 of a main storage device 4 in order. When the node number of the parent node is required, this table is searched to retrieve the node number. The node number storred in a case just preceding a case including the pertinent node number in this table is the node number of the parent node.

Description

[Detailed Description of the Invention] [Field of Application of the Invention] The present invention relates to an information retrieval mechanism using a balanced tree in a computer system consisting of an arithmetic processing unit, a main storage device, an auxiliary storage device, and an input/output device. An information retrieval method that allows you to quickly add and delete information.

[Prior art]

BACKGROUND ART Balanced trees have conventionally been used in digital computer systems to search for information corresponding to a key. The representative one is called B-'f)ree, and various modifications of B-Tree are also being considered. For B-T r e, D
o Comer：”The Ul）iquitou3
33-'pree'AcM Computing 5u
rveys.

VOl, 11. No, 2. pp121-138
checking ...

A typical variation of B-Tree is 13”-Tree.
Ashi, widely used.

Now, consider adding a node to a Node that is full of B"-'l'ree entries. In this case, a new N
A key can be added by securing an Ode and moving half of the keys from a full N Ode to a new N Ode, and the normal process is called 5-plit processing. When performing 5plii processing, the parent N of the target N0de
It is also necessary to update Ode.

On the other hand, in B''-'l'ree, control is performed so that the intra-Node tsf entry rank is between half of the maximum number of entries that can enter the Node and the maximum number of entries.For this reason, by deleting entries, If the number of entries is less than the order, it is necessary to transfer entries from the neighboring Node or to merge the neighboring Nodes. At this time, as with the 5plit process, the parent
Update e.

Conventionally, when updating the parent N Ode, R,oot
Retracing the search starting from N Ode, we searched for a Node that included a pointer pointing to the corresponding N□ae. Normally, water is stored on an auxiliary storage device such as a magnetic disk device (abbreviated as a disk), so access to the auxiliary storage device is required to trace back from the R,oot Node.
This was a performance issue.

[Purpose of the invention]

The purpose of the present invention is to speed up the process of searching for the parent NOde in a balanced tree, such as a B''-Tree, which may require update processing up to the parent N0de when adding or deleting keys. The purpose of this method is to speed up the process of adding and deleting keys.

[Summary of the invention]

In a balanced tree such as a 13"-Tree, before adding or deleting a key, it is necessary to trace from the Root to the Leaf Node where the key should be or is contained. At this time, the Root No.
From de to the corresponding Le af Node'! JT (7
) R the NOde number (the number that identifies the Node)
,oot are stored in a table on the main memory in order. When the Node number of the parent N□de is needed, this table is searched and The N Ode number stored before the case containing the Node number is the N of the parent N0de.
□de number. In this way, the parent No. d
You can know e.

[Embodiments of the invention]

An embodiment of the present invention will be described below with reference to FIGS. 1 to 11. FIG. 1 shows the overall configuration of the present invention. l is a magnetic disk device (hereinafter abbreviated as disk), 2 is a 13"-Tr
The index file consists of
Bitmap managed with -off, 22 is B"-Tre
This is the part that includes the main body of e. 201-208B" -
201 is 1oo in the Node that constitutes T r e
t Node, 205 to 208 are Leaf No.
It is de. 3 is a file label area, and 31 to 32 are file labels.

The file label is i-j:, 311 is the file sector address, 312 is the pit map management information, 313 is the Root Node No. 4' (i.e., B"
-Tr e N0d in order from the beginning of the area of the e main body 22
(number attached to e), 314 contains the key length, and 315 contains the record size.

4 is a main memory, 5 is a group of programs for the system that controls B"-Tre, 51 is a program for adding entries, 52 is a program for deleting entries, and 53 is a program for searching for entries. 54. 55 is a subroutine used by the program 51.52, and is a subroutine that will be described later.
'N4. 'l [1111 N Ode (7J)
A program for registering the N Ode number, and 55 is a program for obtaining the Node number of the parent of the specified N Ode.

Hereinafter, 54 is the Node number registration program, and 55 is the parent N.
Call the 0de number search program.

In addition, until now, when adding or deleting a key to a tree, expression 1
However, more accurately, it is correct to say that an entry consisting of a key and a record is added or deleted, and henceforth, it will be described as the latter. 8 is a buffer for reading N'ode and bitmap on the disk. 6 is the area containing the user program, and 61 is the user program itself.Programs 51 to 5 for B+-Tree citywide 11@
Call 3.

62 is an area for setting a key for searching, adding or deleting, and 63 is an area for setting a record to be added. Desired processing can be performed by passing these addresses to the B''-Tre ei131) @ program.

7 is NNLT (Node Numl) Cr Li5t
), root B+-Tree.
'N□de numbers traced from Node to LeafNOde including the target key are stored in order from the beginning.

Intermediate information of the processing of the 33''-'l''ree control program is stored in 64-68. 64 stores the key of the entry to be added. 65 also stores the record body of the en 1 to be added or the N□de number of the child N□de. 66 is en) IJ addition target N0de N
Remember the ode number.

67 is the relative entry number in N□de of the entry to be deleted, and 68 is the entry deletion target・N(+de of N
□Memorize each de number.

Figure 2 shows the detailed structure of Node. 9 is the entire Node, and the length is fixed to one sector regardless of whether it is Root Leaf 6 or N□de other than these. Reference numeral 91 is an area containing control information of the node and has a fixed length. 92 to 93 are double numbers. In the entry, there is a key indicated by 921 and a V code indicated by 922.
Nc of e) d e number (when other than LeafN□de)
including.

911 is the number of entries included in the Node, 912 is the number of entries included in the Node, and 912 is the number of entries included in the Node.
3 stores a flag indicating whether it is a Root Node, and 914 stores a flag indicating whether it is an i, e3f Node.

The number of entries in Node is controlled so that it falls between the maximum number of entries that can fit in N0de and an order that is 1/2 of this value.

Based on the above explanation of the configuration, the process of adding and deleting entries will be described below, and this method will be explained in detail, as well as the effectiveness of this method.

FIG. 3 shows a flowchart of a program for adding second tones. This flowchart will be explained based on the example of B''-Tree shown in FIGS. 4 and 5.

First, the key value is 2 in B"-Tree shown in A in Figure 4.
Consider adding 5 entries. Here, A in Fig. 4 is B"-Tree before addition, and B"-Tr after addition.
ee, 211ba [Loot Node Te, Node
Numbers 1, 212 to 215 are Leaf Node
The Node numbers are j, t, and m, respectively. At step 101 in FIG. 3, all cases of NNLT shown at 7 in FIG. 1 are cleared to zero. At 102 in Figure 3, the first 3
13, the Node number of R, oot Node, i in the example of FIG. 4, is obtained. At 103 in Figure 3, the sector address on the disk of Node number i is determined based on the information on the file label shown at 31 in Figure 1, as is usually done in file management. N□de first
Read to system buffer 8 in the figure. At 104 in FIG. 3, the read Node (7) Node number i is registered in the NNLT.

This process is performed by the Node number registration program shown at 54 in FIG. Details of NNLT and the NODE number registration program are shown in FIGS. 9 and 10.

In Figure 9, in total, each case 71 is 7BANNLT.
~77 consists of 4 bytes, and case numbers are assigned sequentially starting from 71. No registered in NNLT in 71
Note the number of d e. 72 to 77, )3"-'
The NOde numbers when tracing l:'ree from R, oot Node to Leaf N0de containing the target key are entered in order from the beginning. Maximum N□d eO that can be registered
Although the number depends on the number of cases in the NNLT, this value only needs to correspond to the maximum number of stages of the B+-Tree that can be assumed, and here it is fixed at 15 cases due to the maximum size of the file.

FIG. 10 shows a flowchart of the NNLT registration program. 701, the registered N0dC of NNLT
Add 1 to the number (stored at 71 in Figure 9).

In this case, it was initially 0, so after processing 701,
The number of registered N0de is 1.

At 702, the address of the NNLT case to be registered is obtained by adding the number of registered NOde times the number of bytes per case, 4, to the NNLT head address.

In this case, the address for the case shown at 72 in FIG. 9 is calculated.

At step 703, a designated N0de number is set for the case for which the address was calculated at step 702. In this case, Root Nod
The e number i is set in case 72 in FIG.

LeafN□d of the Node read at 105 in Figure 3
With reference to the e display (indicated by 914 in FIG. 2),
f Node T: Determine whether h. In this case, L.C.
Since it is not afNode, jump to 106 in FIG. In step 106, the keys of the entries in the read Node are compared in order from the first entry to find an entry whose key value exceeds the key value of the first entry to be added. In this case, the first entry with a key value of 30 is the cough guard. The pointer value of this entry (generally indicated at 922 in FIG. 2).

) to obtain the N□d, e number of the lower N Ode, in this case J.

After this, the process goes to 103 in Figure 3, and the lower Node, that is, the Node of Node number J, is read on the system computer (read on the sofa, and at 104, the Node number registration program is set to FF.
[Load number j
is registered in NNLT. In this case, j is set in case 73 in FIG. 9, and the number of registered N0de of 71 becomes 2.

105 is a Lear Node, so 1
07, and here, enter the entry keys ((i.e. 25
), Nod of N0de to which code and entry are added
Set the e number (i.e. j). Like this 131~
137, the Lea in which the entry to be added is to be entered.
f While tracing with Node i, NNLT K R
, o o t to Leaf t is registered, and information regarding the entry to be added to the work is set. The NNLT will look like C in Figure 4.

At 108 in FIG. 3, the number of entries currently in N0de (stored in 912 in FIG. 2) is compared with the number of entries that can fit in N0de.

The number of entries that can be entered in N□de is determined by the entry length, that is, the sum of the key length and encoding length for Leaf N□de, and the length of the area for storing the key length and pointer value is fixed at 4 bytes for non-Leaf N□de. , and the length of the part containing control (goods) information of N0de (fixed value) is 1 remainder.
Find it by dividing the direct. In the case of A in FIG. 4, the current number of entries is 4, and the number of entries (1) that can be entered in N0de is 4, so in the determination at 108 in FIG. 5, the process proceeds to 109. In 109, R in the control information of N0de
, oot display (indicated by 913 in FIG. 2),
It is determined whether the addition target N0de is a Root Node. In this case, since it is not R,oot, the process proceeds to 110. )
Here, an empty node is obtained by referring to the bitmap indicated by 21 in FIG. In this case, 216 in B of Figure 4
The N0de indicated by is a newly secured N0de'''C, and the Node number is n.

In 111 of FIG. 3, entry addition target N Od
The second half of e is newly secured N Od e
Move to. After that, compare the key of the entry to be added and the maximum key of the N0de containing the first half entry, and if the entry to be added is larger, the newly secured N0de is
d Add an entry to e.

If the entry to be added is smaller, it is added to NOde containing the first half entry.

In this case, an entry is added to N0de indicated by 216 in B of FIG. This process is called 5 plit process.

At 113 in FIG. 3, the parent N0ae number search program is called and
The N'ocie number of is obtained from the NNLT7. Parent Nod
A flowchart of the e-number search program is shown in FIG. This flowchart will be explained by taking as an example a case where the NNLT is as shown in C of FIG. Here, is the Node number of the Node to be added an entry J'? There is 1″.

At 801 of the 11th time, the number of N□de registrations of the NNLT, in this case 2, is set in the counter (which may be an area on the main memory or a lane star). At 802, the address of ■+1st case (in this case, 73 in C of FIG. 4) is calculated, and at 803, the address of N0cieA' (in this case J ) and the specified NOde number +3' (J in this case). In this case, it matches, so 805 Hejanbu l,,
NNLT work case (in this case, 7 of C in Figure 4)
The Node number (i in this case) of the parent node can be obtained from the case shown in 2). If 803 in Figure 11
If there is a close match, the process proceeds to 804, where the counter is subtracted by 1, jumps to 80.2, and the search processing for the case in which the specified Node number is stored is continued. In this way, the Node number of the parent NOde of the designated Node can be obtained.

At 114 in FIG. 3, the N o d obtained at 113
N that read the parent Node with e number and caused 5plits
The pointer of the entry in parent N0de that was pointing to Ode is changed to N0de of N0de added for 5plit.
Change to ode number. That is, 21 at A in Figure 4
Parent 1'Jode 21 that was pointing to Node 2
．． The value J of the pointer of entry 1 is changed to n as shown in B in FIG.

At 115 in FIG. 3, additional entry information for adding an entry pointing to the addition target N0de to the parent N0de is created. In this case, the key value is 15,
The pointer value is j1 and the addition target N0de number is i. After this process, the process returns to step 108 and starts the process of adding an entry to N0de again. In this case, additional target N□d
Since e is full, proceed to 109. Additional target N0 at this time
Since de is R, ootNOde, it jumps by 117. At 117, 5pl shown at 217 in B of Figure 4
Two N Odes are secured, one for the N0 de used for the IT processing and the other for the new Root N Ode indicated by 218. 118 and 119 are the 5 plit processing described above.

At 120, the previous Root N0de is no longer the same, so the root display of N0de shown at 211 in B of FIG. 4 is cleared. At 121 in FIG. 5, create a new Root Node containing two entries pointing to two N0de.
, move the pointer to the R,root Node in the file label (indicated by 313 in Figure 1) to the new Root
Update to Node. As a result, B"-T r e
The shape of e is as shown in B in FIG.

In this way, it is necessary to change the parent N□de in the 5plit process, but from the Root to the Leaf
By remembering these Node numbers when tracing to the Node, you can obtain the parent Node number without retracing from the RootNode,
Entries can be added at high speed.

A case where Spl+ does not need to be processed will be explained using FIG. The key value is 15 in f3''-'I'ree of A in Figure 5.
Consider adding an entry for . 101~ in Figure 3
In step 107, the target L to which an entry is added from Root
trace with eaf Node i, register these Node numbers in NNLT, and set additional en l-IJ information on the work. At 108 in FIG. 3, the number of entries in N Ode is 3 and it is not full, so the process proceeds to 116, an entry is added, and the process ends. This result B”-Tl
ee becomes as shown in B in FIG. In this case, parent N0
There is no need to update de.

Entry deletion processing will be explained using FIGS. 6 to 8. First, consider deleting an entry with a key value of 10 from B''-Tree of A in Figure 7. ”-4:
'ree. In Figure 7, 501 is RootNod
e, 502, 503, 504 and 505 Le
af Node, and the Node number is that, i,
j, A and m.

In FIG. 6, in steps 401 to 403, the trace is traced from the Root Node to the Leaf Node containing the deleted entry in the same way as in the entry addition process, and the traced N□ and the N of de
Register the ode number to NNLT. As a result, NNL'l
tj: As shown in C in FIG. At step 403, deletion entry information, ie, N of the entry to be deleted, is entered at 67 in FIG.
Relative factory number in 0de and deletion target NO in 68
d Set the Node number of e.

At 404 in FIG. 6, the number of entries in the deletion target N0de is compared with the order, that is, 1/2 of the maximum value of the number of entries that can fit in the N0de. The maximum number of entries is determined by the method described in the section on adding entries. No in this case
The number of entries in de is 2, which is the same as the order, so 40
Proceed to step 5.

2 in 405 in Figure 6 and the entry is to be deleted.
Determine whether t is R,oot N□de. This is done in the same way as adding an entry. This seventh
In the example in the figure, it is not R, oot Node, so 4
Proceed to 06.

In step 406 of FIG. 9, the parent N0de number search program is called to search for cases where the Node number of NNLT is stored, and to find the Node of the N0de whose entry is to be deleted.
Find the case in which the e number is stored and obtain the 1'tJQde number contained in the case immediately before this case. This is the Node number of the parent N0de. In this case, the NNLT of C in Figure 7 is searched by j, and the N□de number i in the case immediately before the case containing J is the parent N.
This is the Ode number.

At 407, read the parent N0de.

In 408 of FIG. 6, the deletion target N in the parent Node is
Here, the Node number R of the Node next to Node 503 in FIG. 7 is obtained from the entry next to the entry pointing to 0de, and this Node is read into the system buffer. In step 409, the number of fully read entries in N0de, 3 in this case, is compared with the order, 2 in this case. Since the number of entries is larger than the order, the process advances to 410.

At 410 in Figure 6, 501°50 of A in Figure 7
The N Ode of 2 and 503 is set to be 501° and N□de of 502 and 503 in FIG. 7B. That is, the number of entries specified by the deletion entry information (stored at 67 and 68 in FIG. 1) is made not to become smaller than the order. This process is called balancing process.

This completes the entry deletion process. As a result, B”-
The tree will look like B in Figure 7.

Consider deleting the entry whose key value is 70 from B"-'14ee in FIG.

In FIG. 6, the same processing as in the example of FIG. 7 is performed in steps 401 to 407. However, the NNLT becomes as shown in FIG. 8C, and N Ode 601 and 603 are read into the system buffer. 408, the parent N Od
Obtain the Node number of the next Node from the entry next to the entry of the target Node in e, and read it into the Nodel system buffer (7).

At 409 in Figure 6, the adjacent N c) d e's en)
Compare the IJ number and the order. In this case, the number of entries is 2 and the order is 2, so 411 hegejumps are performed. 4
11, after deleting the specified entry, the left neighbor NO
The remaining entries are moved to d and this Node is returned as an empty Node. This process is called merging process.

N0 to which the entry was moved by the merging process of 411 in Figure 6
The maximum key of de changes, that is, Node 6 in Fig. 8.
The maximum key of 02 is the virtual maximum key (B”-T
In order to simplify the processing for ree,
“-Key to add to Tree)”,
The en that points to this Node within the parent Node)
Update the IJ key to the new maximum key in the Node,
In order to delete the entry (l) pointing to the deleted Node, enter the information of the deletion entry at 67.1 in Figure 1.
Set to 68.

In this case, the intra-Node relative entry number of the deletion entry is set to 67, and the Node number i of the deletion target N□de is set to 68.

In Figure 6, it is deleted from the parent N Q d e and 9 N Od e
The process returns to 404 to delete the entry pointing to .

At 404, the number of entries in Node is 2, which is equal to the order, so the process advances to 405. At 405, target l'J o d
Since e is R, oot Node, it jumps by 413. At 413, the target N□de is LeafN
Since it is not Ode, proceed to 414. At 414, the number (l) in the target Node is fixed to 2, so 41
Proceed to step 5. In 415, N became 91 by annexation.

Set Node as a new Root Node and return the current Root Node at t. This result 13
”-Tree becomes like 13 in FIG.

On the other hand, if the Node is 11 B''-'pree and the number of entries is less than the order, 401~
The process proceeds through steps 405 and 413, and at step 413 jumps to step 416 to delete the designated entry and end.

In addition, if there are more en 11 than the order in the Leaf Node that includes the entry to be deleted, the process proceeds from 401 to 404 in FIG. It ends.

Furthermore, entry deletion is possible from the lower N0de to Root
If it spreads through Nocie-5 and the number of entries in 1root Nade is 3 or more but less than the order, 416 is jumped at 414 in FIG. 6, the entry is deleted, and the process ends. This is because the number of entries in ROotN Ode is also allowed to be less than the order.

In this way, even in entry deletion processing, when merging and balancing, it is necessary to update the parent Node.
de number can be obtained.

〔Effect of the invention〕

In the 5-plit process by adding an entry to B''-Tree and the balancing or merging process by deleting an entry, the parent N of the processing target N0de is
It is necessary to refer to and update Ode. According to the present invention, it is possible to know the parent N□de without tracing back from the Root N□de, it is possible to refer to or update the parent N0de at high speed, and it is possible to add and delete entries to B”-Tree. can be done quickly.

[Brief explanation of drawings]

Figure 1 shows the overall configuration of the present invention, Figure 2 shows the internal configuration of the Node, Figure 3 shows the flow of a program that adds an entry to B''-4'ree, and Figures 4 and 5. Figure 8 shows an example of deleting an entry from B+~Tree. Figure 9 shows NNLT, Figure 10 shows NNLT.
Flowchart of the program for registering KNode number,
FIG. 11 shows a flowchart of a program for searching the parent Node number from the NNLT.

Claims (1)

[Claims] In an electronic computer system consisting of an arithmetic processing unit, a main memory, an auxiliary storage device, and an input/output device, for a balanced tree in which information is searched using a key, an ivy N is traced from a Root Node to a Leaf Node. A table that stores identification information of Ode in the order in which it is traced. When adding a key to the balanced tree, R,oot Node is added to the Leaf Node where the additional key should not be stored.
When tracing from N0de, the identification information of the nodes is stored in the table mentioned above in the order of tracing, and when the identification information of the parent Node of the N0de being processed is needed to balance the tree, the table mentioned above is stored. A program that performs logical key addition processing to obtain the identification information of the parent N0de by referring to it. Even when deleting a key from the balanced tree, ROOl N0de is added to the LeafN□de that contains the key to be deleted.
When tracing from , the identification information of the traced Node is stored in the above-mentioned table in order, and the identification information of the parent Node of the Node being processed, which is necessary for balancing, is stored in the above-mentioned table in the same manner as the key addition process. For these programs, there is a subroutine that registers the identification information of the specified N0de in the above-mentioned table in the order in which the nodes are traced, and a subroutine that refers to this table to N □ d e
t7) By providing a subroutine that searches for job information and obtains the identification information of the parent's N Ode that has been stored immediately before, the N Od that is being processed, which is necessary at the time of balancing, can be
e's parent's N Od e's identification information is R, ootN□c
An information retrieval method that eliminates the need for retracing processing from ie and allows particularly high-speed addition and deletion of keys to and from a balanced tree.