JPH0370826B2 - - Google Patents

Description

[Detailed description of the invention]

[Technical Field of the Invention] The present invention relates to a data processing device that can perform a join operation at high speed by performing a relational operation based on a merge/sort process on two relations that are being sorted. [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, a problem arises in that the time required to search the database for necessary data becomes longer. As one method for solving 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 the relational operations is join
equal) operation. This merging operation is, for example, 2
Two relations {R1} {R2}

From [Table], combine the records with the same key (attribute A and attribute C) values for the number of combinations to create a new relation {R3}.

[Purpose of the invention]

The present invention has been made in consideration of these circumstances, and its purpose is to provide a data processing device that can easily and quickly execute a merging operation between two relations. It is in. [Summary of the Invention] The present invention involves inputting first and second relations each consisting of a plurality of records that have been sorted according to a predetermined rule.
When storing the records in the record buffer memory, duplicate values of keys indicating record attributes in each relation are detected, and the duplicate detection results are stored in the first and second record buffer memory in correspondence with each record. Store in file memory. Then, when the records necessary for the relational operation are sequentially read from the first and second record buffer memories and compared, and the output of the record is controlled according to the comparison result, the records attached to the record are compared with the comparison result. The next record to be read is controlled according to the duplicate detection result. In particular, the read control of this record is performed.If the comparison results are not equal, the next record is read from the record buffer memory that stores the record with the smaller value in the direction of the sort process, and the next record is read from the other record buffer memory. reads the same record, and when the comparison results are equal and each of these records is not a duplicate record, reads the next record from the first and second record buffer memories, and the comparison results are equal; If only one record is indicated as a duplicate record, the next record is read from the record buffer memory that stored that record, and the first record of the duplicate records is read from the other record buffer memory. reading, and when the comparison results are equal and the records are both indicated as duplicate records, reading the same record from the first record buffer memory and reading the next record from the second record buffer memory; It is characterized in that it reads out. [Effects of the Invention] Thus, according to the present invention, only the duplicate records in the two relations are checked according to the duplicate detection results attached to each record in the two relations.
Since that record is read repeatedly and other records are read in order, all the record comparison processing required for the merge operation can be accomplished by scanning each of the two sorted relations almost once. It becomes possible to do so. Therefore, if there is almost no duplication of records in each relation, the time required for processing is approximately (number of records in {R1}) + (number of records in {R2}), and the processing time is It becomes possible to significantly shorten the time. Furthermore, since the duplication of key values of records in each relation is checked and the record duplication information is used to control the reading of each record, the control algorithm is simple and the configuration does not become complicated. This brings about great practical effects, such as being able to construct a data processing device without having to do so. [Embodiment of the Invention] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. The figure is a schematic configuration diagram of a data processing device according to an embodiment of the present invention. A data processing device basically includes a calculation unit that performs relational calculations based on merge/sort processing, and an output control unit that controls the output of the results of the relational calculations. The input stage of the arithmetic unit receives first and second relations consisting of duplicate records that have been sorted according to predetermined rules. A duplication detection circuit is provided to detect. This duplication detection circuit includes an input register 11 for setting one word of records input sequentially, a record buffer memory 12 having a storage capacity for one record and a FIFO (first-in/first-out) function, and a comparison circuit 13 for comparing the key value of the record read from the record buffer memory 12 with the key value of the next record set in the input register 11; The key value of the record set in the record buffer memory 12 is determined from the data register 14, which sets one word at a time and transfers it to the next stage circuit, and the equal signal E output from the comparison circuit 13. It is configured to include a flag circuit 15 that determines whether or not the key value of the input record is duplicated, and sets a duplicate flag F if it is duplicated. In addition, when each record is composed of a plurality of words, since the comparison circuit 13 performs comparison processing for each word, the values of keys (attributes) of successive records in the input relation are The timing at which it is determined whether or not there are duplicates is when the last word of the record is processed. Therefore, the duplication flag F becomes valid when the last word of the record is transferred to the next stage via the data register 14. Therefore, each record of the first relation input through such a duplicate detection circuit is
The data is input to the 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 second relation record 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) is stored in the second record buffer memory 17 corresponding to that record. These first and second record buffer memories 16 and 17 each have a capacity to store a record of one relation, and are controlled by record readout control circuits 18 and 19 to record stored records and add data to the records. The reading of the duplicate detection results is controlled. Note that the record reading control circuits 18, 19
are address register 2 for read control, respectively.
0 and a backup register 21. The backup register 21 saves the value of the address register 20 or stores the value in the backup register 21 according to the record comparison result (described later) and the duplicate detection result (duplicate flag F) attached to each record. The stored value is loaded into the address register 20. The first and second record buffer memories 16 are updated by updating the values set in the address register 20 and the backup register 21 based on the comparison result and the duplication detection result (duplication flag F). , 17 are controlled. In addition to this, the record read control circuits 18 and 19 are provided with write registers and a control circuit that controls these registers, although they are not shown in the drawings as they are not directly related to the gist of the present invention. Needless to say. The record data read from the first and second record buffer memories 16 and 17, respectively, and the information on the duplicate detection results attached to the records are set in data registers 22 and 23, respectively. The comparison circuit 24 includes the data registers 22, 2
The record data set in 3 and 3 are compared with each other, and the comparison results are output to the record read control circuits 18 and 19 as well as to the output control circuit 25 at the next stage. The record read control circuits 18 and 19 are configured by the comparison circuit 2 as described above.
4 and the information on the duplicate detection results set in the data registers 22 and 23 are inputted to update and control the data set in each of the registers 20 and 21. The third and fourth record buffer memories 26 and 27, which have a storage capacity for one record and have a FIFO function, are connected to the data registers 22 and 2.
3 through the record buffer memory 16,1
Each record transferred from 7 to 7 is temporarily stored therein. These record buffer memories 26 and 27 store all of the records in the record buffer memory 26 and 27 because the comparison process of the records is performed word by word and the comparison result is obtained at the time when the last word of the record is processed. It is provided to save each word of the record until the comparison process for the word is completed. In this way, each of the record buffer memories 26, 2
The records stored in each of the records 7 are read out under the control of the output control circuit 25, and are outputted from the selection circuit 28 via the output register 29. Note that the output of this record is the same as the record stored in the third record buffer memory 26 and the record stored in the fourth record buffer memory 26.
This is done by combining the records stored in the record buffer memory 27. This combining process is performed by the selection circuit 28 outputting, for example, the record stored in the third record buffer memory 26 and then outputting the record stored in the fourth record buffer memory 27. It is carried out with When outputting a combination of two records, the comparison operation described above is determined by the word length of the record with the largest number of words, whereas the output time is determined by the word length of the record with the largest number of words. The time corresponds to the long word ``wa''. For this reason, record output processing time generally takes longer than record comparison calculation time. In order to deal with this, for example, the arithmetic unit may execute the comparison operation in the first half of one cycle, and put the second half in a waiting state. In addition, the arithmetic unit is provided with a circuit that outputs whether or not a key field is currently being processed, and a circuit that outputs whether or not the last word of the record, that is, the end of the cycle, is being processed. Of course. Next, the basic flow of the merging operation process in the embodiment device will be explained. () First, the relations {Rx} are sorted in ascending order (smallest first) according to their key values, duplicate records are detected, a duplicate flag F is added, and then the records are stored in the first record buffer memory 16. do. Similarly, the relation {Ry} is sorted in ascending order according to its key value, duplicate records are detected, a duplicate flag F is added, and the records are stored in the second record buffer memory 17. () Next, each of the above relations {Rx} {Ry}
A pointer (hereinafter simply referred to as a pointer) and a backup pointer (hereinafter referred to as a backup pointer) are placed in the first record of the file. () Then, the record indicated by the pointer is read from the record buffer memories 16 and 17, and the key values are compared. Then, if the key values of the records are equal, the merging condition is satisfied and both records are output together, and if the key values are different, output of that record is stopped. Therefore, if the values of the above keys are equal,
Check the duplicate flag F attached to that record,
The following processing according to the duplication flag F is performed. First, if the duplicate flags F of both records are both "0", the pointer of each relation is
Proceed to the next record. At this time, the backup pointer also saves the new value of the pointer. If only the duplication flag F of either record is "1", the pointer of the relation containing the record whose duplication flag F is "1" is advanced to the next record. At this time, the value of the backup pointer is also updated with the value of the above pointer. For a relation that includes a record for which the duplication flag F is "0", the record pointed to by the backup pointer is set as the pointer. In this case, the value of the backup pointer is not updated. If the duplication flags F of both records are both "1", the pointer of the relation {Ry} is advanced to the next record, and the backup pointer is not updated. On the other hand, for relation {Rx}, the record pointed to by the backup pointer is used as a pointer. Also, as mentioned above, if the key values are different, the relation with the smaller key value,
In other words, the pointer of the relation having the key with the smaller value is advanced to the next record in the direction of the sort processing. At this time, the backup pointer is also updated at the same time. In this case, the duplication flag F is not used for pointer control. () Thereafter, if there is a record to be compared next in each of the relations, the process returns to the process of () described above, and if there is no record in either of the relations, the process described above ends. By the way, in the above process (), if we consider the case where the key values of two records are the same, there are four possible situations depending on whether or not the records in each relation are duplicated.
Regarding each of these states, the above-mentioned pointer control will be explained using relation {R4} and relation {R5} as examples. (1) The relation {R4}{R5} when both records do not overlap is shown, for example, as follows.

[Table] However, in this case, records [2/G] and [2/p] are output in cycles. At the end of this cycle, even if the record pointed to by the pointer in relation {R4} or relation {R5} is left behind, the key value of the record pointed to by the pointer to which the previous pointer moved will be the one left behind. Since it is clear that the key value is greater than the key value of the other record, the pointers of both relations can be updated at the same time. In the next cycle, both relations may advance their pointers to the next records, [3/T] and [5/h], and perform the comparison process. (2) An example where the key values of records in relation {R4} are duplicated and the key values of records in relation {R5} are not duplicated is:
It is shown as follows.

[Table] In this case, at the end of the cycle, the pointer of relation {R4} is advanced from [2/G] to [2/Q], and the pointer of relation {R5} is fixedly recorded. [2/
p]. Then, in the cycle, since the duplication flags F of both records become "0", the pointers of both records are updated at the same time. (3) If the key values of records in relation {R4} are not duplicated, and only the key values of records in relation {R5} are duplicated, the following will occur.

[Table] In this case, in the cycle, the pointer of relation {R4} is fixed to point to record [2/G], and the pointer of relation {R5} is changed from [2/p] to [2 /t]
All you have to do is move forward. Then, in the cycle, since the duplication flags F of both records become "0", it is sufficient to update the pointers of both records at the same time. (4) Next, if the key values of the records of both relations are duplicated, the following will happen, for example.

[Table] In this case, in the cycle, the pointer of relation {R4} is fixed to point to record [2/G], and the pointer of relation {R5} is changed from record [2/p] to [ 2/t]. However, the backup pointer of relation {R5} is
First record of duplicate records [2/
p] and fix it. In the cycle, only the duplication flag F of the record of relation {R4} becomes "1". Therefore, the pointer of the relation {R4} is advanced to the next record [2/Z], and the pointer of the relation {R5} is returned to the record [2/p] pointed to by the backup pointer. In the cycle, the pointer of relation {R4} is fixed, and the pointer of relation {R5} is changed from [2/p] to [2/p].
t]. At this time, since the duplication flag F of the record of relation {R4} is "0", the backup pointer of relation {R5} is updated. In the cycle, the duplicate flag F for both records
both become "0", and both relation pointers advance to the next record. By performing the processes (1) to (4) above, the merging operation process is performed for all combinations of [2/α] and [2/β] in both relations. Therefore, after this process, there is no need to return the pointers of both relations to the beginning. Next, the processing operation of the above-mentioned embodiment apparatus will be explained using relation {R1}{R2} as an example.
Here, it is assumed that the word length of each attribute of relation {R1}{R2} is all one word length, and one record is composed of two word lengths. In addition, the processing time for one word is assumed to be one phase, and the processing time for one record (1
cycle) consists of two phases. First, a plurality of records constituting the relation {R1} are sorted in ascending order by the value of their key (attribute A) to obtain the following relation.

[Table] The record data of this relation {R1} is input sequentially, and first it is checked whether there is a duplicate record in the relation, and the check result is output in synchronization with the record. Detection of this duplicate record is performed as follows. That is, each record of relation {R1} is
The records are input in the sorted order described above, and therefore record [2/F] is input in the first cycle. In the first phase of this first cycle, the first word [2] of the first record is input, set in the input register 11, and stored in the record buffer memory 12. In the next second phase, the next second word [F] is set in the input register 11 and stored in the record buffer memory 12. The input of the first record is completed by the processing of the first and second phases. In the subsequent second cycle, the following three processes are performed in parallel. One is inputting the second record [3/U] and storing it in the record buffer memory 12, and the second is reading out the second record and the record buffer memory 12. 1
This is an operation of comparing the second record [2/F]. And the third is the record buffer memory 1.
This is an operation for storing the first record [3/U] read from No. 2 into the first record buffer memory 16. At this time, the value of the duplication flag F determined for the first record is also stored in the first record buffer memory 16. That is, the first process in the first phase is to input [3], which is the first word of the second record, and set it in the input register 11, as in the first cycle. This is done by storing it in 12. Then, the first word [2] of the record stored one cycle ago is read from the record buffer memory 12 having a FIFO function, and transferred to the first record buffer memory 16 via the data register 14. At the same time, it is applied to the comparison circuit 13 and compared with the first word [3] of the second record set in the input register 11. In this case, it can be seen that the two words are not equal, that is, the conditions are not met. In the subsequent second phase, the value [U] of the second word of the second record is stored in the record buffer memory 12 from the input register 11, and the next word of the first record is stored from the record buffer memory 12. [F] is read out and transferred to the first data record buffer memory 16 via the data register 14, and at the same time, the input register 1
Compare with word [U] set to 1. This phase is processing of the last word of the record, and since the key value of the first record does not overlap with the key value of the next record, the equal signal E (=0) of the comparison circuit 13 is In response, the flag circuit 15 outputs a duplication flag F (=0). Thereafter, similar processing is performed in each cycle, but in the third cycle, record [6/D] is input, and in the fourth cycle, record [6/P] is input.
is input, a duplicate flag F (=1) is attached to the record [6/D] input in the third cycle. Through this process of detecting duplicate records,
The relation {R1} is stored in the first record buffer memory 16 in the following format.

[Table] After that, relation {R2} that has been subjected to the same sorting process as relation {R1} is input in sequence, duplicate records are detected in the same way, and the following is stored in the second record buffer memory 17. obtain relationships.

[Table] After the above processing, the merging calculation process between the relations stored in each record buffer memory 16, 17 is performed as follows. First, pointers are placed at the first records of both relations. In other words, record read control circuit 1
The first word of the first record of each relation, that is, [2] of record [2/F], and record [1/F] are stored in each address register 20 of 8 and 19.
Each pointer is set to point to the address where [1] of [e] is stored. At the same time, the values set in each address register 20 are saved in the backup register 21, respectively. After completing the above preparations, merge operation processing is started. First, in the first cycle, the following three processes are performed in parallel. However, the first cycle in the duplicate record detection process described above and the first cycle here are at different timings. That is, first, the first record [2/F] of the relation {R4} read from the first record buffer memory 16 and the relation {R5} read from the second record buffer memory 17.
Compare with the first record [1/e] of Then, the first record [2/F] read from the first record buffer memory 16 is stored in the third record buffer memory 26, and the first record [2/F] read from the second record buffer memory 17 is stored. 1/e] to the fourth record buffer memory 2
Store in 7. Then, the pointer is controlled to select a record to be compared in the next cycle. Specifically, in the first phase, the word [2] pointed to by each address register 20 is extracted from the first and second record buffer memories 16 and 17.
[1] is read out at the same time, and the roller is set to data register 2.
After setting 2 and 23, the comparator circuit 24 compares them. In this case, since the values of each word in both records do not match, that is, the key value of relation {R1} is larger, it is decided not to output the record. Therefore, the output control circuit 25 is instructed to record [2/F] in the next second cycle.
Notify that [1/r] should not be combined and output.
Then, each address register 20 is counted up. In the next cycle, the pointer of the relation {R2} with the smaller compared key value is advanced, and the pointer of the relation {R1} is returned to its original value. At the same time as this processing, each of the data memories 22, 2
The data [2] and [1] stored in the record buffer memories 26 and 27 are respectively read out.
Store each in In the subsequent second phase, the second word [F][e] of each record is read from each record buffer memory 16, 17, and the second word [F][e] of each record is read out from the data register 22,
23, the record buffer memory 2
6 and 27, respectively. Since the key value of the record in relation {R2} is smaller, the pointer in relation {R2} is advanced, and the pointer in relation {R1} is left unchanged. That is, when moving from the second phase to the next phase, the first phase of the second cycle, the record read control circuit 18 loads the address register 20 with the value of the backup register 21, and reads the first record [2]. /F] so that it can be read again. On the other hand, the record read control circuit 19 counts up the address register 20 so that the next record [2/k] can be read. Then, at the beginning of the next second cycle, the value of the address register 20 is saved in the backup register 21. In the subsequent second cycle, the relation {R4} read from the first record buffer memory 16 is
The first record [2/F] of the relation {R5} read from the second record buffer memory 17 is compared with the second record [2/q] of the relation {R5}. And each of the above record buffer memories 16,1
Records read from 7 [2/F]
[2/q] is stored in the record buffer memory 26,
27, and select the records to be compared in the next cycle. Each of the above processes is performed in the same way as the process in the first cycle, but in this case, since the read record is [2/F][2/k], the key values match. Therefore, in the next third cycle, records [2/F] [2/F] are read out from the third and fourth record buffer memories 26 and 27 by the output control of the output control circuit 25.
k] are combined and output. At this time, at the end of the second phase, since the comparison results are the same, the duplication flag F attached to each record is input to the record read control circuits 18 and 19. in this case,
Duplicate flag F attached to both records is both “0”
Therefore, the pointers for both relations are advanced. In other words, each address register 20 is counted up,
At the beginning of the next third cycle, the values are saved in each of the backup registers 21, respectively. In the next third cycle, in addition to the three processes performed in the first and second cycles described above, a combined output operation of records is performed. That is, in the third cycle, relation {R4}
The second record [3/U] of relation {R5} is compared with the second record [2/k] of relation {R5}, and at the same time each of the above records [3/U] [2/k]
are transferred and stored in each of the record buffer memories 26 and 27, and pointer control is performed to select a record to be compared in the next cycle. In addition to this, the record buffer memory 2
The first record [2/F] of the relation {R1} and the second record [2/F] of the relation {R2}, which are stored in 6 and 27, respectively, and whose key values are determined to be equal. k] are read out via the selection circuit 28, combined and output. To explain this combined output operation of records in more detail, in the first phase, the first word [2] of the record is read from the third record buffer memory 26, and then sent to the output register 2 via the selection circuit 28.
Set it to 9 and output this. In the subsequent second phase, the next word [F] is read out and output in the same manner.
Thereafter, in the third and fourth phases, the words [2] and [q] of the record of relation {R2} are sequentially read out from the second record buffer memory 27 and sent to the output register via the selection circuit 28. 2
Set to 9 and output. As a result, the output of the output register 29 is the first record [2/
F/2/q] is obtained. In subsequent cycles, the same processing as in the third cycle is performed. However, in the fourth cycle, the records compared in the third cycle are [3/U] and [2/k], and since their key values are different, the combined output of these records is not performed. When such a cycle is repeated in the same manner and records [6/D] [6/s] are compared in the fourth cycle, it is detected that the key values are equal. In this case, the above record [6/D]
[6/s] are combined and output. However, at this time, since the duplication flags F attached to each of the above records are both "1", the address register 20 of the record read control circuit 18 that controls the read of the relation {R1}, as described above, does not contain that value. The value of the backup register 21 is loaded,
Pointer control is performed so that the record [6/D] is read out again. On the other hand, at this time, the address register 20 of the record read control circuit 19 is counted up and the pointer is controlled to read the next record [6/w]. However, the value of the backup register 21 of the record read control circuit 19 is maintained as is, so that it is possible to return to the beginning of the duplicate records in that relation, that is, record [6/s]. In the fifth cycle as well, the key values of both records match as [6]. Therefore, records [6/D] [6/w] in this case are combined and output.
However, in this case, since only the duplication flag F of relation {R1} is "1", at the end of this cycle, the value of the address register 20 of the record read control circuit 18 is counted up and the next record [6 /P] is read out,
Moreover, the value of the backup register 21 is updated to the value of the address register 20. In other words,
The backup pointer is controlled so that this record [6/P] can be read again. Then, the address register 20 of the record read control circuit 19 is loaded with the value stored in the backup register 21, and the record [6/s] which is the head of the duplicate records of that relation is read out. Pointer control is performed so that In the next 6th cycle, record [6/P]
[6/s] are compared, and since the key values match, the records are combined and output. Since only the duplication flag F on the relation {R2} side is "1", the pointer of the record read control circuit 18 is not updated, and the value of the address register 20 of the record read control circuit 19 is counted up. Moreover, the value of the backup register 21 is updated to the value of the address register 20. And in the seventh cycle, record [6/P]
[6/w] are compared and combined and output. In this case, since the duplication flags F of the respective records are both "0", the relation pointers are controlled to point to the next record. However, in this case, since there is no next record in both relations, the record comparison process ends in this cycle. As explained above, this device performs a merge operation on relation {R1}{R2} from the first record to the last record by one scan, excluding duplicate records. It is possible to perform the necessary record comparison processing. Therefore, if there are no duplicate records in the relation, or if the amount of duplicate records is negligible, the execution time required for the merge operation is equal to the number of records in the relation {R1} and the number of records in the relation {R2}. The cycle time corresponds to the sum of the number of records. In other words, in the conventional example, the execution time was determined by the "product" of both numbers of records, but according to the present invention, it is determined by the "sum". Therefore, it becomes possible to efficiently execute the merge operation process in a short time. In addition, the above process is performed by comparing the key values between records that are input one after the other, and by attaching a duplicate flag F to each record according to the comparison result, making it efficient with a simple algorithm. Can perform merge operations well. Note that the present invention is not limited to the embodiments described above. For example, the number of keys (attributes) constituting each record, the word length of the keys (attributes), the number of records constituting a relation, etc. may be determined according to specifications. Also regarding the annexation conditions. It is not particularly limited. In short, the present invention can be implemented with various modifications without departing from the gist thereof.

[Brief explanation of drawings]

The figure is a schematic configuration diagram of main parts of a data processing device according to an embodiment of the present invention. 11...Input register, 12...Record buffer memory, 13...Comparison circuit, 14...Data register, 15...Flag circuit, 16...First record buffer memory, 17...Second record Buffer memory, 18, 19...Record reading control circuit, 20...Address register, 21...
... Backup register, 22, 23 ... Data register, 24 ... Comparison circuit, 25 ... Output control circuit, 26 ... Third record buffer memory,
27...Fourth record buffer memory, 28...
...Selection circuit, 29...Output register.

Claims (1)

[Claims] 1. means for detecting duplication of keys indicating attributes of records in first and second relations consisting of a plurality of records sorted according to predetermined rules; The first one stores each record of
and means for storing the duplication detection result in a second record buffer memory in correspondence with each record, and mutually comparing the keys of the records respectively read from the first and second record buffer memories. means for combining and outputting a record in the second relation that is equal to the key in the first relation with a record in the first relation according to the comparison result; and and means for controlling the next record to be read from the first and second record buffer memories according to the duplication detection results attached to each of the records subjected to the comparison process, and controlling the reading of this record. If the comparison results are not equal, the means reads the next record from the record buffer memory storing the record having the smaller value in the direction of the sorting process, and reads the same record from the other record buffer memory. reading, and if the comparison results are equal, check the duplication detection results attached to each of the records, and if each of these records is not a duplicate record, read the next record from the first and second record buffer memories, respectively. , if only one record is indicated as a duplicate record, the next record is read from the record buffer memory that stored that record, and the same record is read from the other record buffer memory, so that each record is duplicated. When indicated as a record, the record from the second record buffer memory is backed up and stored, the same record is read from the first record buffer memory, and the next record is read from the second record buffer memory. The subsequent duplicate records are sequentially read, and then the next record is read from the first record buffer memory, and the backed up record is read from the second record buffer memory, and then the first record buffer is read. A data processing device characterized in that the same record is read from the buffer memory and the next record is read from the second record buffer memory.