JPS59731A - Join operation processing system - Google Patents

Abstract

PURPOSE:To execute a join operation among plural relations at high speed by adopting the multi-stage structure of plural consitution of sort boxes performing data comparison and sort processing in the form synchronized with data transfer. CONSTITUTION:An exclusive function processor SEP group performs actual data accessing based on an intermediate language sequence for data processing transferred from a command processor CP group. Data distributors DD 11-16 being one building block of the exclusive function processor SEP group share data transfer according to data values to be set with respect to the transferred data. Plural constitution amount the data distributors DD 11-16 and sort boxes SB 31-36 sorting the transfer data is adopted as the multi-stage structure via a communication network CN50. Thus, the join operation between two relations (or called tables) R1 and R2 is performed.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a data processing system using a compound computer consisting of a plurality of processors that realize dedicated functions, a plurality of storage devices, and a communication network that dynamically connects an arbitrary number of these devices. The present invention relates to a data processing method that compares and sorts data in synchronization with streams and processes join operations in relational databases at high speed.

A join operation in a relational database is expressed as follows: RI −ajlθRk IIa,k(i\k) where:
RI, Rk are relations, a, l are relations R
The third attribute of +, a, is the relation R, and the fourth attribute θ is a set of comparison operators (<, <, \.

One of =, 〉, :)).

That is, it is a comparison operation between attributes of different relations. This calculation is performed using conventional computer systems.
As shown in FIG. 1, relation data is transferred from the storage device 3 to the host CPU 2 via the controller 2, and is processed by software from the host CPU.

Furthermore, with the advancement of hardware technology, various database machines have been proposed, including high-speed processing methods for join operations using hashing hardware, sorting hardware, associative memory, bitmaps, etc. Proposed. However, most of the above proposals store data in temporary memory and sequentially retrieve it from the data and process it on the host CPU, which lacks the point of processing synchronized with the data stream, which makes the data transfer time effective. No plans are being made to use it. Some datast +7-
There is also a proposal to perform sort hk based on system processing and use it for join operations. However, the processing of the sorting hardware proposed therein includes data input processing (comparison, data movement processing) to the sorting hardware, and data output processing accompanied by data comparison from the sorting hardware. Requires.

The purpose of the present invention is to eliminate the above-mentioned conventional problems, and to provide a communication network with a plurality of processing units that perform data comparison and sorting processing at high speed in a format synchronized with the transferred data stream. The object of the present invention is to provide a method for high-speed processing of join operations in a pl relational database using a connected multi-stage structure.

To achieve these objectives, the present invention utilizes a communication network that dynamically couples a plurality of processors and a plurality of storage devices that implement dedicated functions, and any number of these devices to synchronize data transfer. With a multi-stage structure of multiple sort boxes that perform data comparison and sorting processing,
Simultaneous range specified note processing for multiple relations, parallel batched search processing between range finger toe sort result data by switching to a batched search processor for all sort boxes, and join operations between multiple relations.
It is characterized by fast execution.

Hereinafter, the present invention will be explained in detail with reference to Examples.

[Configuration of Example]

FIG. 2 shows an example of the configuration of a data processing system that fully realizes the present invention, which includes a plurality of microprocessors, a plurality of storage devices, and an arbitrary number of these devices that realize various dedicated functions. This figure shows a complex computer system consisting of a communication network and a controller group 9 that controls them.

The CP (Command processor) group 6 accepts the compilation results of the high-level data access language transferred from the CPU (host), creates an optimal access (processing) path, and processes it using the following 5FP (5pe).
ciaJ Function processor
) Conversion into intermediate language sequences executed by group 7, transfer of these to 8FP group 7, and integration processing of processing results such as SFP etc. In addition, CPC (Com
mandProcessor Control
ler) controls the allocation/release of the CP group 60 based on the data processing request transferred from the host) CPtJl, and controls the CN (Communication Ne).
(work) 5, it manages information communication with other controller groups 9.

The SFP group 7 is a processor group that manages collective data access based on the intermediate language sequence for data processing transferred from the CP group 6.

This intermediate language includes, for example, relational algebraic operations (such as the above-mentioned join operation, projection operation, and selection operation), sorting operations, and appropriate set operations.

List operation creation processing 9 includes appropriate algebraic operations 1, etc.

8 FPC (5special Function)
processorController) is
The SFP group (1) controls, for example, the SFP group 7 dynamically based on the data access intermediate language sequence being transferred from the CP 6 (a plurality of 5FPs 7 can be dynamically assigned to one intermediate language sequence). control of allocation/release control and information communication with other controller groups 9 via the CN 5'e.

The BS (J3+offer Storage) group 8 is used to temporarily store data transferred from the storage device 3 such as a magnetic disk, or to hold intermediate results processed by the SFP group 7, and is used to store IC memory, magnetic BSC (1) consisting of bubble memory, COD memory, etc.
3uffer S storage Control
r) controls the 80% broadcast/release control of the BS group and information communication with other controller groups 9 via the CN5.

Furthermore, 88 (Secondary 3torag
e) Group 3 is a storage device for storing a database, such as a large-capacity magnetic disk device. It is assumed that these devices are designed to increase the speed of transfer by serial transfer. 88 C (S secondary 8 torag
The eController) controls data transfer from the SS group 3 or data transfer from other devices to the 88 group 3, and controls information communication with other controller groups 9 via the CN 5'z.

FIG. 3 shows an embodiment of the present invention, and is a data distributor DD (Data) istrib@, which is a component of the SFP group 7 and distributes data transfer according to set data values. ter ) 1
1 to 16 and 5B31 to 36 for sorting the transferred data in multiple stages via multiple groove bottoms 1cN50 (Figure 3 shows two rations (also referred to as tables) R with a two-stage configuration.
This figure shows a processing form for performing a join performance between s and FL.

FIG. 4 shows another embodiment of the present invention, in which data distributors DD17-28 and 5B37-48'1CN
50i through three rations R, , R2, and R3
This shows a processing form for executing join calculation processing between.

FIG. 5 is an internal configuration diagram of the data distributor DD200. The DD 200 is a device that separates data, and includes a comparison operator register 203 in which a comparison operator is set, a comparator 204 that performs comparison processing based on the contents of the register 203, and a comparison between the comparator 204. A latch circuit 205 for setting a result, a data register DR 206 for setting comparison data for comparison with the transferred data, and a data register DR 207 for setting the transferred data to be compared. Based on the latched contents of the latch circuit 205, the data register D
Selector switch 5W20 to change the destination of the contents of R.
8, and a control circuit 202 that controls each of the above circuits.

FIG. 6 shows the internal structure of cells constituting the sort box SB'r that performs sorting processing.

This cell transfers data from other cells, comparators 306, data registers DR, 308 and DR, 30, which will be described later.
a delay circuit 304 that controls data input to 9, and a delay circuit 304 that controls data input to the 1-data register DR, 30;
Comparison operator register 30 in which a comparison operator is set for comparison with data in 8 or DR, 309.
5 and the data transferred from other cells and the DR, 308 or DR based on the contents of the register 305.
,.

A comparator 306 that performs comparison processing with the data in 309, and a latch circuit 307 that sets the comparison result of the comparator.
and a plurality of selector switches 310 for switching data transfer.
~312 and two data registers DR, 308, DR, 3 in which data transferred from other cells is set.
09 and a control circuit 302 that controls each of the above circuits. The sorting bots 5B31-48 consist of a linear array of the cells described above.

[Operation of the embodiment]

(5) Equicombination of two rations (eQL+1−j
oin) processing operation 3rd. Fifth. This will be explained using FIGS.

Now, assume that the data stream of the pair of join 0 attribute and sequence number (added in the order of data transfer in the DF'CB) of relation R+1 is as follows:
(This is called R, stream.) (101, 11, (105, 2), (102, 3).

(106,4), (103,5), (107,6).

(111,71, (104,8), (115,9).

(109,10), (113,11), (108,12
) Also, assume that the data stream of the join attribute and sequence number pair of relation R2 is as follows: (This is called the R2 stream.) (121,1), (101,2), (112,3) .

(105,4), (111,5), (122,6J.

(103,7), (112,8), (104,9).

(115, 113, (101, 12), (117, 13
).

(112, 153, (102, 167) Therefore, let us consider the equi-connection between the above two relations R1 and Ut.

(a) The currently usable pair (DD, 8B) is a relation join attribute (JoinAttri).
bute ), the equijoin (eQLli -join ) is processed as follows.

It is assumed that 8B can accommodate all data.

■ Join to DR206 of DDII and 14 - Maximum attribute value (115° for R, 122° for R2
), 1 to C0R2O3”.

and a transfer data information management unit in the control circuit 202 (not shown, the detailed structure is, for example, a utility model registration application 55
-145531. ), set the size of the data field to be transferred, etc.

(2) The R and stream are input to the data distributor DDII through the line 101, and the at stream is input to the data distributor DD14 through the line 104 (they can be input at the same time).

■ Transferred data stream to DDII and D
D14 is processed as follows.

@RI, DR206 of R1 stream DDll is set to the value "'115" or a value greater than this.

First, when (101, 1) is transferred, the control circuit 2
According to the control of 02 (delay circuit can be used), (101,1
In accordance with the input to the comparator 204, the comparison data value "'115" is input from the DR 206 to the t-comparator 204,
Comparison is performed based on the set value °゛<'' of C0FL203.At this time, at the same time, the other DR207 has the above (1
01,1). Control circuit 2021d opens and closes 5W208 according to the result of comparison latch 205, and transfers the data value in DR207 (transfer to SB or next DD)
control. In this case, for 1101"<"105", DDll transfers (101, 1) to 5B31 through the entire line 119. Here, all input data is much smaller than the contents of DR206 of DDII. So all 5
Transferred to B31.

- Processing of R, stream The value "'122" is set in DR2, 06 of the DD 14, and data is transferred to the DD 14 in order starting from data (121, 1). The control and processing inside the DD is the same as that described for R, S) IJ-input processing. In this case, all input data is smaller than the contents of DR206 of DD14, so
It is forwarded to 5B34 through line 128.

■ R, stream 1sB transferred from DDII
31, and each data value is held in either DR308 or DR,309 of one cell in 5B31.

Which one is known from the valid DR flag 303. Similarly, the stream transferred from DD14 has 8B
34, and each data value is held in either DR308 or DR309 of one cell in 5B34.

That is, the R1 stream and the R,stream are sorted as follows. These are respectively referred to as 'stream' and R2' stream.

-R8' stream (generated from R, stream): Cell 1: (101,1), Cell 2: (102°3),
Cell 3: (103,5), Cell 4: (104,8)
, cell 5: (105,2).

Cell 6: (106,4), Cell 7: (107°6)
, cell 8: (108,12), cell 9: (109,10
3, Cell 10: (111°7), Cell 11: (113
,111,Cell 12: (115,9) #几! 'Stream (generated from Rt stream): Cell 1: (101, 2), Cell 2: (101°12), Cell 3: (102, 16), Cell 4: (
103,7), cell 5: (104°9), cell 6:
(105,4), cell 7: (106,103, cell 8:
(111,5).

Cell 9: (112,3), Cell 10: (112,8)
, Cell 11: (112°15), Cell 12: (115
,11), Cell 13: (117,13), Cell 14:
(119,14), cell 15: (121°1), cell 1
6: (122,6) ■ Transfer the R and I streams held in 5B31 to line 122. CN50. It is transferred to 5B34 through line 183 and line 128, and within 8B34, a comparison operation (here, prior to the data transfer of 5B31, 5B34
The COR of each cell in the cell is reset to equal sign 1 = "), and if the transferred data does not match, it controls the 5W 312' and passes the entire line 325.
Transfer to next cell.

For matched data, a pair of sequence numbers is created from the contents of DR308 and DR309, and a pair of sequence numbers is generated on line 324. Line 326.2 in 131°CN50, and line 13
Through 4, B.S.

(assuming that BSI can be used here) stores this pair. At the same time as transferring the pair of sequence numbers, the data to be compared is transferred to the next cell. This operation is performed for all the transferred Rlm' stream data.

In the above example, it would look like this: ・Comparison between R1' stream and R2' stream The 8B34 performs the following operations.

Cell 1: KiR2' (101, 2) = R, I' (101
From ゜1), create a sequence number pair (1゜2) and transfer it to the BS. At the same time, R1' (10t, 1) is transferred to the next cell. The next data R1' (102, 3J) is transferred through this post-transfer delay circuit 304.

In this case, R,'(101,2) From \R,'(102,3), R1' (102,3) is transferred to the next cell be done. The above processing operations are transferred All incoming RI' stream data This is done for each person.

Cell 2: R2' (101, 12) = R1'.

From (101, 1), the sequence Create No pair (1, 12), Transferred to BSL. At the same time, R1' (Transfers 101,1 to the next cell.

The following processing is the same as that for cell 1.

The above processing operations are executed in cells 3 to 16, and each cell generates the sequence NO pairs shown below, and the BS,
is stored in

Cell 1: (1, 2), Cell 2: (1, 12).

Cell 3: (3,16), Cell 4: (5,7).

Cell 5: (8,9), Cell 6: (2,4).

Cell 7: (4,10), Cell 8: (7,5).

Cell 9: no output, cell 10: no output, cell 8 cell 11: no output, cell 12: (9,11), cell 13: no output, cell 14: no output, cell 15: no output, cell 16: No output Currently, the number of active cells in 5B34 is up to 16, and the R1' stream data swept out from cell 16 is 5.
It is overridden by control circuitry (not shown) of B34. Furthermore, as shown in the above example, there is no output from cell 13 onwards, as there is no cell that satisfies the conditions. That is, cell 1
Processing after 3 is meaningless. In order to avoid this, it is also possible to provide the following control. That is,
Check in advance whether the maximum value in the transferred data stream is smaller than or equal to the data value in the cell number in the SB where comparison data is held, and then The first thing to do is to set the active range of cells. Therefore, in the above example, u, 5B3
Activate cells up to cell 12 of cell 1, and
When the data values swept out from the cell 12 are invalidated and the final data value of the R1' stream has finished processing in the cell 12, step (2) can be completed and processing can proceed to the next step.

■ The corresponding CP (see Figure 2) reads the pair of sequence numbers stored in B, S, and based on the read, reads the BS (or multiple BSs) in which the relations R1 and the Tara pull of R1 are stored. (Acceptable) and performs the actual data merging process.

In the pair of sequence numbers above, the first term beam ration R,
The second term shows the storage order in the BS (sometimes stored in the BS with unnecessary attributes removed), and the second term shows that of the ratio R2. Accordingly, the storage address of each Tara pull within the BS can be obtained as follows.

In other words, the sequence number of the relation Rs (Rt) is vt (Vt), the length of the relation is rs (rtl (byte)), and the relation R+ (Rt) in the BS
If the storage start address of the tuple is St (St), then the storage address in the BS of the relation R1 tuple: Adr (几t"rt)=St+(Vs 1)" The storage address in the BS of the relation R2 tuple: A dr (R2' rt)=St + (V2 1)"
rtFurthermore, if the Tara pull is stored across different BSs, an address is generated based on the identifier of the stored BS. In the corresponding CP, each data is read based on the above address and if it is a substantial data join process (semi-join),
It is only necessary to read data from one side).

■ The newly generated data after data combination in CP is sent to host CPU, CP6゜5FP7 or BS.
Transferred to 8th grade.

Cb) Next, a case where multiple DDs and SBs can be used will be shown. Now assume that two DD and 8B pairs are available for each join attribute. At this time, the comparison operation between join attributes is executed as follows. The data used is the same as described above.

■Join to DR2o6 of DDII and DD14.
Attribute value “107”, 1 for C0R2O3”,
Furthermore, the size of the data field to be transferred to a transfer data information management unit (not shown) in the control circuit 202, etc. are set. Also, D of DD12 and DD15
R, 206, the maximum value of the join attribute value,
115 for R8. Set 122 for each of R and R, and set the others in the same manner as above.

■ (a) (!: Similarly, the R1 stream and the R,stream are input to the data distributor DD.

■ Each DD processes the transferred data stream as follows.

- R, stream processing The details are omitted because the processing proceeds in the same way as (a). D
Data values larger than 61o7'' set in DII are transferred to the next DD12, and data values smaller than '107'' are transferred to SR31. In addition, the data value transferred to DD12 is set to "1" in DD12.
As a result of the comparison with 15", the data is transferred to 8B32.

- Processing of R2 stream Data values larger than w ``107'' set in DD14 are transferred to the next DD15, and data values smaller than ``107'' are transferred to 5B34. Also, data transferred to DD15 The value is set to DD15.”
122” and is transferred to 8B35.

■ The process proceeds in the same way as in (a), and the sorting results at each SB (SR31's R31' stream, SR32's i
R,,'St IJ-4,5R346'MrR7,'Strino 1.5B35 is R72' stream]
becomes as follows.

a R,,'s) IJ-m (in SR31) -1=R: (101,1, +h2: (102°3)
, -t=nyl: (103,5), -1=/l, 4:(
104,8), Cell 5: (105,2).

Cell 6: (106,4), Seven Lua: (107°6
) * R, 2' Story, z, (inside SR32) Cell 1
: (108,12), Cell 2: (109,10), Cell 3: (111,7).

Cell 4: (113,11), Cell 12: (115,9
R, 2, 'St I) - Person (SR341' [Cell 1
: (101,2), Cell 2: (101°12), Cell 3: (102,16), Cell 4: (103,7)
, *Le 5: (104°9), Cell 6: (105,
4), Cell 7: (106,10) "Rt2's) ') -m (inside SB 35) Cell 1: (111,5), Cell 2: (112°3), Cell 3: (112, 8), Cell 4: (112, 15 island Cell 5: (115° 11), Cell 6: (117, 13), Cell 7: (
119, 14), cell 8: (121°1), cell 9:
(122, 6) ■ (Similar to ■ in aJ, perform a comparison operation between the above streams and obtain a pair of sequence numbers. According to the value set in each DD, each of the above streams is controlled as follows. In other words, the R11' metric is
2. CN50° line 183. Through line 128,
R2,' is transferred to the 5B34 where the stream is located, and the same processing as shown in (a) (■) is performed, and the pair of sequence numbers is transferred to the line 131. CN50° and the entire line 134 are stored in BS. Similarly, connect the Rj12' stream to line 123°CN50. Line 184. line 12
9 to the 5B35 where the R12' stream is located, and the same processing as above is performed (in order to realize parallel processing, pairs of sequence numbers are stored in the BS).

Through the above processing, the following sequence NO pairs are stored in BS1: (1, 2), (1, 12), (3, 16).

(5, 7), (8, 9), (2, 4J, (4° 10) In addition, the following pair of sequence numbers is stored in BS: (7, 5), (9, 11 ) (As shown by ■ in F4, processing can be sped up by recognizing the range of the ac line cells.

■ Same as ■ in (a) ■ Same as ■ in (a) Next, we will show the equi-combination of three rations.

■ Processing operation 1 for equijoining three relations, join attribute a of relation R3 and attribute a of relation Rt, and then join relation R
1 and attribute b of relation R3. To detailed processing operation (
This is the same as that shown in b), and only an overview is shown here.

■ DI) 17-DD19. DD20-DD22゜DD
23~DD25. and each DR20 of DD26 to DD28
6, respectively, (R1(a.

b), SEQ+t　　　　　Stream,(R,轡.

sEQ+2) stream, (R'l base.

(SEQs, SEQ+tl) stream, and (R
3 eggs, SEQ! s) stream, etc. (values related to the underlined attributes above), M < 77 for each COR 203, and a transfer data information management section (not shown in the figure) in each control circuit 202. Set the necessary information in

■ Input the (approximately r (a, b), SEQ+t) data stream to DD17-19 through lines 107-109, and transfer the data stream distributed by each DD to each SB37-39 through lines 137-139. . Similarly, through lines 110-112, D
Input the (Rz (a), SEQ,) data stream to D20-22, and pass the data stream distributed by each DD through lines 146-148 to each 5B40-22.
42 and through lines 116-118 to DD2.
A 6 to 28K (R, (t), 8EQ+8) data stream is input and transferred to each 8B 46 to 48 through all data stream lines 164 to 166 distributed by each DD. These processes are executed simultaneously.

■ (Rr (3-b), 5EQU1) data stream in 5B37 to 39, (R1 R≠SEQ≠,
) data stream by 8B40-42, (R3(off, SEQ≠31) by 5B46-48, each time by the underlined attribute value.

■ (R1(door, b), SEQS,) stream, (R
The sorting results of the t<8), SEQ+t) stream, and the (R1°(h), SEQ+t) stream are set as follows.

・Regarding R□, RIIZ R1,',...
Regarding R111'・R1, RlIZ FL,,',・
..., +(+2j'・R3, RIll
'＋R3! 'e......, l(,3 to 'R■'~
R1 amount' is set to line 140, 141°142, CN50.
Line 189, 190° 191, Line 146, 147
, 148, etc., to 5B40, 5B41.8B42, etc. held by R11' to R25' to each DD17 to D.
Schedule and input based on the data values set in D19 and DD20 to DD22. For example, i=
J, R11' in 5B37 is in 5B40, 8B38
R12' in 5B41 can be input to 8B41, and R11' in 5B39 can be input to 5B42 at the same time. That is, comparison processing of the divided and sorted results can be executed in parallel in each unit. Furthermore, if it is necessary to compare a plurality of sort results with one or more sort results depending on the value set in each DD, input transfer is scheduled accordingly.

As a result of the above comparison process, (R1
(SEQS1.SEQ+t)) The IJ-me schedules the input to line 113 to the BSs reserved for each (because it is parallel processing, the data transfer is (because it would cause a conflict),
Stored temporarily. However, only one SB in the comparison result, for example (R1(1))-(SE
QL.

SEQ+! )) The stream is input directly to DD 23 through line 113 or line 152.

The parallel processing results are sequentially read from the BS, the same processing is performed, and the results of sorting by attribute b into 8B43-45 are held based on the data values set in DD23-25.

■Similar to (R11 parable, (8EQ).

The stream of SEQ, )) is divided and sorted into SB43 to SB45, and these are on lines 158 and 15.
9,160. CN50° line 195, 196, 197
．． Line 1.64.

Through 165.166, (R73Φ).

8EQ+s) is divided and sorted and held in 8B
46, 47.48. etc. each D23~25 and DD26~
Schedule and input based on the data values set in 28.

Parallel processing with multiple input data streams, 5B
From 46 to 48, sequence N is obtained as a comparison processing result.
O triplet (8EQ+,.

A stream of SEQS, SEQS,) is stored across lines 170-172 in a previously reserved BS. If there are few BSs available, 8 B 4
Comparison processing in 6 to 48 is restricted (to prevent storage contention in the BS).

■ C that executes and manages the alignment including the main join operation
P uses itself or a plurality of other free CPs to execute the same process as described in (4) (2) in parallel. If there is one CP, sequential processing is performed.

■ Same as prisoner ■.

0 The above can also be processed as follows.

First of all, ■(i'tt Hibiki, SEQ,), (a, low.

SEQI), (& Sat, SEQ+t), (Rs
Cb) Input the data streams of SEQ+s) into the corresponding DDs, divide and sort the streams based on the data values of DR, 206 in the set DD, and
The sorting result of (Rt to l 8EQ+1) and (R
A pair of sequence numbers (SEQ≠,.

SEQ+t) stream creation and (Rt(b).

Sorting results of SEQ, ++) and (Rla(b),
Pair of sequence numbers (5EQ) Ih, SEQ, +s by comparison operation with the sorting result of 8 EQ4'a)
) streams simultaneously and each stream) IJ-
The system is temporarily stored in multiple BSs, and each BS (8EQ+t, SEQ+t) and (SEQ+t)
EQ, shSEQ4h) e lead, 'tsE respectively
In order to sort by Q+, the free DD and S
Transfer each to H. - Then, a comparison operation is performed on the sorted results as shown in (5) and [F]J, and the triple of sequence numbers ((SEQ+, .

The streams of SEQ+,, SEQS,)) are stored in a reserved BS in parallel or sequentially, and ■ (5).
It is also possible to carry out steps ① and ② mentioned in [■].

The above example shows the processing method for join operations between multiple relations, and it is assumed that all the transfer data within the value set in the DD can be accommodated in the SB corresponding to that DD, but it cannot be accommodated. In this case, use the lines 71 to 84 or couple multiple SBes via CN'. At this time, the value of DR206 in the DD is also changed.

This also makes it possible to sort large amounts of data.

Regarding the transfer of sort results from the SB to the CN50, etc., this can be further speeded up by controlling the transfer of the one with the smaller amount of data.6 [Application of components, design changes] 9 Alternative Measure] In FIGS. 3 and 4, there is strong coupling between the DDs and between the DD and the SB, but this allows for a flexible configuration. This is shown in FIG.

The symbols are the same as those used in FIGS. 3, 4, etc. With this configuration, a determination that yields the same result as that according to FIG. 4 is shown in FIG. In addition, although a specific example has been shown above regarding equi-coupling, it is also possible to perform other types of coupling by changing the set value of the COB between SBs.

As explained above, according to the present invention, the following effects can be obtained.

1.Based on data stream processing, 0(n
) By using multiple SBs that sort by ax(”t r nt )
)+0 (min("s + "t)/"" (1
+ J l ): O(""(nl+n2)). Here, "1 1n!" is the base number of the relations R, , R, respectively, and ieJ represents the number of SBs used to sort the relations R1, R, , respectively.

2. By using CNe, it is possible to dynamically and freely combine a plurality of DD and 8B pairs according to the processing, thereby speeding up the join calculation processing.

[Brief explanation of drawings]

Figure 1 is a schematic diagram of a conventional computer system, and Figure 2 is a schematic diagram of a conventional computer system.
A configuration diagram of a database machine according to an embodiment of the present invention, No. 3
Similarly, FIG. 4 is an explanatory diagram showing an example of join processing between two relations, FIG. 4 is an explanatory diagram similarly showing a join processing method using three relations, FIG. This figure is also an internal configuration diagram of a cell constituting a sort box (SB), and FIGS. 7 and 8 are configuration diagrams showing other application examples.

Claims (1)

[Scope of Claims] 1. A data processing system using a compound computer consisting of a plurality of processors that realize dedicated functions, a plurality of storage devices, and a communication network that dynamically connects any number of these devices; The elements are a comparison operator register that sets a comparison operator, a comparator that performs comparison processing based on the contents of the register, a latch circuit that sets the result of the comparator, and a data register that sets comparison data. , a data register that sets the transferred data to be compared. A data distributor consisting of a switch that switches data transfer based on latch contents and a control circuit that controls the above circuit, a delay circuit that controls data transfer, a comparison operator register that sets a comparison operator, and the register. A comparator that performs comparison processing based on the contents of the comparator, a latch circuit that sets the result of the comparator, multiple switches that switch data transfer, two data registers that set the data to be transferred, and the above circuit. A multi-stage structure in which multiple configurations of sort boxes consisting of a linear array of cells configured with control circuits are connected via a communication network enables data comparison processing by simultaneous processing of multiple data streams. A join operation processing method characterized in that the acquisition of comparison processing results, especially the join operation between relations in a one-routine database, is processed at high speed.