JPS61110234A - Data processor - Google Patents

Abstract

PURPOSE:To easily execute restricting calculation at a high speed, by detecting and comparing overlap of records of the 1st and 2nd relations and controlling next read out records in accordance with the compared results and overlap detected results added to the records. CONSTITUTION:Records and overlap detected results added to the records read out from the 1st and 2nd record buffer memories 16 and 17 are respectively set in data registers 22 and 23. A comparator circuit 24 compares the record data with each other and outputs the compared results to control circuits 18 and 19 and, at the same time, to the output controlling circuit 25 of the next stage. Record buffer memories 26 and 27, each of which has a storage capacity which is able to store one record, once store each record from the buffer memories 16 and 17 and preserves each record word until the comparing process is performed to all words of records. The records respectively stored in the buffer memories 26 and 27 in such a way are read out under the control (restricting calculation) of the output controlling circuit 25 and outputted from a selection circuit 28 through an output register 29.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention provides data processing in which constraint operations between two sorted relations can be executed at high speed by relational operations based on Mercy sort processing. Regarding equipment.

(Technical background of the invention and its problems) With the advancement of mass storage devices, the number of computer systems that construct large-scale databases on the mass storage devices is increasing. However, as the database becomes larger, the time required to search for necessary data from the database becomes longer. As one method to solve this problem, various proposals have been made to execute the processing necessary for the search, that is, the relational calculations, using dedicated hardware.

By the way, one of these types of relational operations is the constraint operation (res
triction). This constraint operation, for example, uses two relations (R1) (-condition) as shown below.
, relation (R2), the values of its keys (attribute A and attribute B) satisfy the constraint condition (
For example, by selecting only the records that match (equal), the following new relation (R3) is obtained.

However, conventionally, this constraint operation involves sorting the above two relations in ascending order, and then merging between them.
This is done by performing a sorting process. In other words,
First, a pointer is placed at the first record of both relations, and the key values are compared between the records pointed to by this pointer, and the record with the smaller key value is selected according to the comparison result. However, since the above constraint calculations are processed by always advancing only one pointer and comparing records, including when the key values are equal, there is a problem that the calculation processing time is long. Ta.

[Purpose of the invention]

The present invention has been made in consideration of these circumstances, and its purpose is to easily and quickly execute constraint operations based on merge/sort processing between two sorted relations. The object of the present invention is to provide a data processing device that can perform the following tasks.

(Summary of the Invention) The present invention provides a method for inputting first and second relations consisting of lean records that have been sorted according to predetermined rules and storing them in first and second record buffer memories, respectively. , detects duplicate records in each of the above relations, and stores the duplicate detection results in the first and second record buffer memories in correspondence with each record.Then, records necessary for the relational operation are stored in the first and second record buffer memories. In particular, the read control for this record is as follows: ■ If the comparison results are not equal, the next record is read from the record buffer memory storing the record with the smaller value in the direction of the sorting process. At the same time as reading the record, the same record is read from the other record buffer memory; (i) When the comparison results are equal, and the result of detecting duplicate records in the first relation indicates that the record is not a duplicate record; In addition, if the detection result of the record 11 of the second relation indicates that it is a duplicate record, the record to be read from the first record buffer memory is maintained as it is, and the record read from the second record buffer memory is continued. Advance the record to be output to the next record, and ■ When the comparison results are equal and the duplicate detection result for each record is other than ■, advance the record to be read from each record buffer memory to the next record. It is characterized by the following.

〔Effect of the invention〕

Thus, according to the present invention, according to the duplicate detection results attached to each record of two relations, the records with the same comparison result and the record of the second relation subject to the restriction are duplicated. As a result, records can be read from each relation one after the other, so all the record comparison processing required for constraint calculations can be performed by scanning each of the two sorted relations almost once. becomes possible. In other words, records in each relation can be selected almost simultaneously, making it possible to significantly shorten the time required for the process. Moreover, since the duplication of records in each relation is checked and the readout of each record is controlled using the record m-multiple information, the control algorithm is simple and data processing can be performed without complicating the configuration. There are great practical effects such as being able to construct a device.

[Embodiments of the invention]

Hereinafter, one embodiment of the present invention will be described with reference to the drawings.

The figure shows that the input stage of the arithmetic unit to which the first and second relations each consisting of a plurality of records that have been sorted according to a predetermined rule of the data processing device according to the embodiment of the present invention is inputted. A duplication detection circuit is provided to detect duplication of records in the relation. This duplication detection circuit has an input register 11 for setting one word of a record that is input sequentially, and a storage capacity for the record code.
A record buffer memory 12 having a FIFO (first in/)7 strike out) function compares the value of the record read from this record buffer memory 12 with the value of the next record set in the input register 11. A comparison circuit 13, a data register 14 that sets the record read out from the record buffer memory 12 one word at a time and transfers it to the next stage circuit, and an equal signal E output from the comparison circuit 13 to the record buffer memory. 12, and a flag circuit 15 for determining whether or not the record set in No. 12 overlaps with the current input record, and setting a duplication flag F if the record is overlapping.

In addition, when each record is composed of a plurality of words, since the comparison circuit 13 performs comparison processing on each double voltage,
The timing at which it is determined whether or not the key (attribute) values of successive records in the input relation are duplicated is that the last duplicate flag F of the record indicates that the last edict of the record is in the data register 15. It becomes valid when the data is transferred to the next stage via .

Therefore, each record of the first relation inputted through such a duplication detection circuit is inputted to the first record buffer memory 16 and stored therein. At this time, the duplication detection result (duplication flag F) for each record detected by the duplication detection circuit is stored in the first record buffer memory 16 in correspondence with that record. Similarly, when a record of the second relation is input, the record is stored in the second record buffer memory 17 via the duplication detection circuit, and the duplication detection result (duplication flag F) detected for each record is is stored in the second record buffer memory 1 corresponding to that record.
7 is stored.

These first and second record buffer memories 16
．． 17 each has a capacity capable of storing a record of one relation, and a record read control circuit 18.1
9211.9, the value of the address register 20 is saved, or the value of the address register 20 is saved, or the value of the backer 717-,) is read out 11 (duplication flag F) of the stored record and the duplicate detection result added to the record under the control of 9211.9. : Loads the stored value into the address register 20. The values set in these address registers 20 and backup registers 21 are updated based on the comparison results and the duplicate detection results (Ii double flag F).
ll1 controls the records of each relation read from the first and second record buffer memos January 6.11.

In addition to this, although not shown because it is not directly related to the gist of the present invention, there is also a record readout III circuit 18.
．． It goes without saying that 19 is provided with write registers and a control circuit for controlling these registers.

The record data set in the records read from the first and second JDJO buffer memories 16 and 17, respectively, are compared with each other, and the comparison results are sent to the record reading circuits 18 and 19.
At the same time, it is output to the output III m circuit 25 at the next stage. The record read control circuits 18 and 19 input the record comparison result by the comparison circuit 24 and the information of the duplication detection result set in the data registers 22.23, and read the data in each of the registers 20.21.
Update control of the set data.

3rd and 4th record buffer memories 26 and 27 that have a storage capacity for 1 code and also have a FIFO function.
are for temporarily storing each record transferred from the record backup memory 16.17 via the data register 22.23. These record backup memories 26 and 27 store all words of the record because the record comparison process is performed word by word and the comparison result is obtained at the time when the last line of the record is processed. This is provided to store multiple copies of that record until the comparison process for that record is completed. And in this way, each of the above records,
This record output is performed by combining the record stored in the third record backup memory 26 and the record stored in the fourth record buffer memory 27; This is done by outputting only the records stored in . When performing the above-mentioned combined output process, for example, after outputting the record stored in the third record buffer memory 926 via the selection circuit 28, the record stored in the fourth record buffer memory 27 is outputted as described above. This is done by outputting through the selection circuit 28. Here, when combining two records and outputting them, the comparison operation described above is determined by the word length of the record with the largest number of words, whereas the output requirement is determined by the word length of both records. The time corresponds to the ``wa'' of . For this reason, record output processing time generally takes longer than record comparison calculation time. In order to cope with this, it goes without saying that, for example, the performance section is provided with a circuit or the like that executes and outputs the comparison operation in the first half of one cycle.

Next, the basic flow of the merging operation process in the embodiment device will be explained.

CI) First, relations (Rx) are sorted in ascending order (smallest to smallest) according to their key values, duplicate records are detected and a duplicate flag F is added, and then stored in the first record buffer memory 16. Similarly, relation (R
y) in ascending order according to the key value, detect duplicate records, add a duplicate flag F, and then
The data is stored in the record buffer memory 17 of.

(II) Next, each of the above relations (Rx) CF? V), a pointer (hereinafter simply referred to as a pointer) and a backup pointer (
Hereinafter, this will be referred to as a backup pointer).

(I) The records indicated by the pointers are read from the record buffer memories 16 and 17, and the values of their keys are compared with each other.

Therefore, if the key values of the above records are equal, the duplicate flag F set is O'', and the duplicate flag F attached to the record of the relation (Ry) is M1'', then the pointer of the above relation (Rx) , and advance the relation (RY) pointer to the next record.

■ And it was attached to each record! ? ! If the relationship of flag F is other than the relationship shown in (2) above, the pointers of each relation (Rx) (Ry) are advanced to the next one.

■ Note that if the key values of the above records are not equal, the record is not output. Then, the pointer of the relation having the smaller key value, that is, the relation having the smaller key value in the sorting direction, is advanced to the next record. In this case, the duplication flag F
is not used to control pointers.

After calculating (mV)L, if there is a record to be compared next in each relation, return to the process of ([[) described above, and −
When there is no record to be compared next in a given relation, the above-described process ends.

By the way, in the process shown in (Ill) above, if we consider the case where the key values of two records are equal, depending on whether the records in each relation are duplicated or not, there are the following four situations, which are actually ineffective. This is the meaning, but considering that case, the relation (R4) and relay 1 for each state above! (1) The relation (R4) (R5) when both records are not duplicated is, for example, as follows. In this case, if at the end of cycle ■, the relation (
R4), or even if the record pointed to by the relation (R5) pointer remains, the key value of the record pointed to by the pointer to which the pointer went earlier may be greater than the key value of the remaining record. As is obvious, the pointers of both relations can be updated simultaneously.

(Records of 2 relations (R4) are duplicated,
An example of a case where records of relation (R5) are not duplicated is shown as follows.

In this case, only the relations that define the constraint conditions are duplicated, so even if the pointers of both relations are updated simultaneously in cycle (2), no problem will occur in record comparison for constraint calculations.

(3) If the records of relation (R4) do not overlap and only the records of relation (R5) overlap, the following will occur.

In this case, in cycle (2), the pointers of both relations cannot be updated simultaneously. In other words, when the pointers are updated simultaneously, the next record [2/ll
is left behind, whereas relation (R4) advances to the next record [5], resulting in record [2/l]
is no longer output. Therefore, if only the record of the relation on the side subject to constraint processing is duplicated, as shown in cycle (2), only the pointer on the relation (R5) side is updated without updating both pointers simultaneously.

Then, in cycle (2), since the duplication flags F of both records become "O", the pointers of both records are updated simultaneously.

(4) When records overlap in both relations, the relation is, for example, as follows.

In cycle ■, the duplicate flags F are both 1'', so even if the pointers are updated at the same time, the condition records will continue to show the same value.Therefore, it is necessary to update the pointers of each relation at the same time. Can be done.

In cycle (2), since both duplication flags F become "0", the pointers of both relations can be updated simultaneously.

Next, the processing operation of the above-described embodiment apparatus will be explained using relations (R1) and (R2) as an example. In addition, here, the word length of each attribute of relation (R1) (R2) is all 1 word length, and the lou code of relation (R1) is constructed with 1 word length, and the lou code of relation (R2) is constructed with 2Ii length, respectively. shall be It is also assumed that it takes 17 aces to process one word and one cycle to process a lou code. Therefore, here we use the relation (R2
), one cycle is 27 azes.

First of all, enter the condition data (records) of this relation (R1) in order by ascending the multiple records that make up the relation (R1) using the value of the key (attribute A).
First, it is checked whether there is a duplicate record in the relation, and the check result is output in synchronization with the record.

In this case, there are no duplicate records, and the duplicate flags F for each record are all "0". This is stored in the first record buffer memory 16 in correspondence with each of the above records as follows.

Next, after the relation (R2) and the relation (R1) are subjected to various sorting processes, it is checked whether records in the relation are duplicated or not. Then, the test result is output in synchronization with the record and stored in the second record buffer memory γ. Therefore, the relation (R2) is, for example, as follows.

To explain this severity record detection process, the relations (R2) are given in the sorted order described above.

In the first cycle, the first record N/r] is input. That is, in the 17th aid, the first word [1] of the record is input and set in the input register 11. The value (1) set in this input register 11 is stored in the record buffer memory 1-2.

At the next 27th aid, the next edict [r] of the record is set in the input register 11 and stored in the record buffer memory 12 at the same time. This completes the input of the first record.

In the subsequent second cycle, the following three processes are performed in parallel. That is, the process of inputting the second record [5/s] and storing it in the record buffer memory 12, reading out the first record [1/rl stored in this buffer memory 12 one cycle before, and storing it in the data register. 14 to the next stage circuit (record buffer memory), and the first record [1/r] read from the buffer 7 memory 12 and the second record [5 A process is performed to mutually compare the key values with /sl.

Specifically, in the first phase, [5], which is the first 1R of the second record, is set in the input register 11 and stored in the record buffer memory 12, and at the same time, the FI
Record buff with FOII ability! 14 to the next stage. Furthermore, the first 1 of the first record above
1[1] to the comparator circuit 13, and the input register 1
Compare with value [5] which is set to 1. As a result of this comparison, it can be seen that the above-mentioned numbers are not equal. Therefore, the flag circuit 15 sets the duplicate flag F at the end of the cycle.
O''.

In the subsequent second phase, [S], which is the second word of the second record, is set in the input register 11 and then stored in the record buffer memory 12. Then, from the record buffer memory 12, the second word [r] of the first record
is read out and output via the data register 14. At this time, since the processing is for the last word of the record, the flag circuit 15 outputs the duplication flag F(-0) in accordance with the equal signal ε of the comparison circuit 13 in synchronization with the output of the second word.

Similar processing is performed in the subsequent third cycle.

Through such processing, the above-mentioned relation (R2) is stored in the second record buffer memory 17.

Thereafter, constraint calculations are performed between the relations stored in each record buffer memory 16, 17. This constraint calculation is started after placing a pointer at the first record of each of the above relations. In other words, each address register 2 of the record read control circuit 18, 19
0, the first record of each relation and the first data, that is, [5 in relation (R1)]
] and the record [1/r
] Set pointers to point to the first word [1] of each. Incidentally, each backup register 21 saves the data set in each address register 20 described above.

No output.

That is, in this first cycle, the following two processes are performed in parallel. However, the first cycle in the duplicate record detection process described above and the first cycle here are critical timings.

Here, first, each of the record buffer memories 16 and 17
The values (words) indicated by each address register 20 of the record read control circuit 18, 19 are read out from the record read control circuit 18, 19 and set in the data registers 22, 23, and then the values (words) are compared in the comparator circuit 24. At the same time, the values set in the data registers 22 and 23 are stored in the third and fourteenth record buffer memories 26 and 27.

At the end of the cycle, the next record to be read from each record buffer memory 16, 17 is determined.

Specifically, in the 17th aid, the first and second
The first words [5] and [11] of each record are simultaneously read from the record buffer memories 16 and 17 of the record buffer memory 16 and 17, set in the data registers 22 and 23, and compared by the comparison circuit 24. At this time, it is notified that the two values do not match, and the address register 20 of the record read control circuit 19 is counted up.

At this time, the value [5]H] set in each of the data registers 22.23 is transferred to the M3, jll, and fourth record buffer memories 26, 27, respectively, in the second phase. The next word I[rl read from the second record buffer memory 11 is sent to the fourth record buffer memory 2 via the data register 23.
Store in 7.

Then, when moving from this second phase to the next phase, the first phase of the second cycle, the relation (R1
) is larger, so as described above, only the pointer of relation (R2) is advanced. In other words,
The address register 20 of the record read control circuit 18 is loaded with the value stored in the backup register 21, so that the leading address [5] can be read out again. On the other hand, the address register 20 of the record read II t1 circuit 19 continues to count up so that the next record [5/S] can be read. Then, at the beginning of the next second cycle, the counted up value of the address register 20 is saved in the backup register 21.

At the end of the phase, the comparison results are the same, and the duplicate flag F of relation (R1) is "O" and the duplicate flag F of relation (R2) is "1", so only the pointer of relation (R2) is used. Proceed. In other words, the address register 2 of the record read control circuit 19
0 is counted up and the value is saved in the backup register 21 at the beginning of the next third cycle. Then, the record read control circuit 18 loads the value stored in the backup register 21 into the address register 20.

In the next third cycle, in addition to the above three processes, record output processing is performed under the control of the output control circuit 25.

That is, in the 17th aid, the first word [[
2] is read out via the selection circuit 28 and output to the output register 2.
After being set to 9, it is output.

As a result, the first record [2/q] of the relation (R31), which is the output result of the constraint calculation between the relations (R1) and (R2), is obtained.

Further, in this cycle, the key values of the records read from the first and second record buffer memories 16 and 17 are [5] and [5/k], and a matching result is obtained. Therefore, in the next fourth cycle, the above record [
5/kl is output.

Also, at the end of the 27th Aze of this third cycle,
The comparison result of the key values of the above records is the same value, and both relations! ! ? ! Both values of flag F are “0゛′”
Therefore, the pointers of both relations are advanced.

In other words, each address register 20 of the record read control circuit 18 and 19 is counted up, and at the beginning of the next fourth cycle, the value is updated to each backup register 20.
1 is saved.

In the fourth cycle, the key values of each record are different, and the constraint calculation ends when the value of the relation [R1] is larger.

As explained above, in this device, when the comparison results of the key values of the records are equal, except when the relation (R1) is not duplicated and the relation (R2) is duplicated, both records pointers can be advanced simultaneously. Therefore, it is possible to improve the efficiency of reading records of each relation, and it is possible to significantly shorten the time required for constraint calculation processing. Furthermore, it becomes possible to easily execute constraint calculation processing.

Note that the present invention is not limited to the embodiments described above. For example, the number of records that make up each relation, the number of attributes that make up each record, the word length of each attribute, etc. can be determined according to the specifications of the database. ! stomach. Moreover, the constraint conditions can also be set according to various purposes. In short, the present invention can be implemented with various modifications without departing from the gist thereof.

[Brief explanation of drawings]

Claims (1)

[Claims] The first and second relations obtained by sorting a plurality of records according to a predetermined rule are inputted to the calculation unit, and a relational operation is performed between the records of each relation to obtain the first relation specified by the record of the first relation. In the data processing device that selects and outputs records in two relations, the arithmetic unit includes means for detecting duplication of records in the first and second relations, respectively, and a means for detecting duplication of records in the first and second relations, and a means for detecting duplicate records in the first and second relations. first and second record buffer memories that respectively store records in correspondence with the records of the second relation, and a comparison circuit that mutually compares the records read from the first and second record buffer memories, respectively; means for controlling the output of the record according to the comparison result;
means for controlling records of the first and second relations to be read next from the first and second record buffer memories according to the duplication detection results attached to each of the compared records and the comparison results; and means for controlling reading of each record from the first and second record buffer memories, the comparison result of each record read from the first and second record buffer memories is equal, and When only the record read from the second record buffer memory is indicated as a duplicate record, only the record to be read next from the second record buffer memory is updated; otherwise, the first and second records are updated. The next record to be read from the record buffer memory of the record buffer memory is updated, and when the comparison results of the records read from the first and second record buffer memories are not equal, the record is set to a smaller value in the direction of the sorting process. A data processing device characterized in that the data processing device updates a record of a relation on the side that indicates.