JPS63143627A - Associative summarizing arithmetic unit - Google Patents

Abstract

PURPOSE:To attain rapid processing by executing expansive summarizing operation highly in parallel by means of an associative memory device about group identifying processing and rapidly executing said operation by means of an exclusive circuit about summarized value calculating processing. CONSTITUTION:Associative processing is constituted of two phases consisting of a registration phase (phase 1) for registering a group in the associative memory device 3 and a group identification and summarized value calculation phase (phase 2). In the phase 1, different data (group) in data with group-by property (data for summarized operation) are found out and registered by retrieving the device 3. In the phase 2, data to be summarized are identified in each group by retrieving the device 3 and the summarized value of an entry to a summarized value table 5 corresponding to the obtained group number is calculated. In the phase 2, the group identification processing and summarized value calculating processing are regularly executed by pipeline processing.

Description

[Detailed Description of the Invention] "Industrial Application Field"
The present invention relates to an associative aggregation arithmetic device that quickly executes various aggregation operations on a relational database that specifies (-by) attributes and aggregation attributes.

"Prior Art" This invention deals with not only aggregation operations for relational databases that have been known in the past, but also extended aggregation operations that have not been proposed in the past, so this will be explained first. Commonly known aggregation operations specify group-by attributes and aggregate attributes in the same relation. For example, when Lane Yon searches for part-time employees (name 1 year old 2 departments/sections), the search for ``Find the number of part-time employees for each department'' is 1.
The department name may be used as a group-by attribute, and the number of part-time employees may be used as an aggregate attribute. In relational algebra, this is written as part-time employee <department name, count>.

On the other hand, the extended aggregation operation introduced in this invention uses the relation for which the aggregate value is obtained and another relation.
Specify the p-by attribute. For example, if the ration is a part-time employee (name: 1 year old, department/section name) and an organization (department/section name: 1 person, section manager name), the same search as above can be performed using the section name of the organization ration as the group-by attribute. The department name of the employee relationship is the key attribute, and the number of part-time employees is the aggregate attribute. This is expressed in relational algebra.

Part-time employee [department/section name, count) (organization [department/section name]). As can be seen from the above example, the difference between the extended aggregation operation and the conventional aggregation operation is whether the group-by attribute is derived from the relation for which the aggregate value is to be obtained. In the example above, this difference is the difference between (department/section name in part-time employee) and (department/section name in organization),
In other words, it differs by the name of the department or section that does not have part-time employees. In traditional aggregation operations.

Department/section names without part-time employees do not appear, but in the extended aggregation calculation, such department/section names also constitute a tuple with the number of employees as 0. In this way, extended aggregation operations can leave data that would be missing in conventional aggregation operations.

Conventional dedicated devices for processing aggregate operations (of course conventional aggregate operations) include 2, for example:
ons on Database System
s” (pp, 214-242. Vol, 9. NO,
2, 1984)” A Parall
el Pipelined Re1ational Q
aggregation pipe (
aggregationpipe). In this conventional aggregation device, as shown in FIG.
Integration (sorting) is performed based on the group-by attribute, and then, focusing on the group-by attribute, aggregate value calculation is performed using a dedicated circuit.

In other words, in the prior art, an efficient sorting mechanism for large amounts of data is required, and the processing load of the sorting process cannot be ignored.

The purpose of this invention is to solve the problem of a special processing mechanism and processing load for sort processing, which was an essential route for conventional aggregation processing devices, and to be able to process not only conventional aggregation operations but also extended aggregation operations at high speed. The object of the present invention is to provide an integrated computing device.

"Means for Solving Problems" According to the present invention, there is an input/output system that exchanges data with an aggregate calculation request source, a processing control system that controls aggregate value calculation processing, and an associative memo! 1. An associative memory device that is capable of linking a large number of ILsI, an aggregated value table that stores aggregated values at intermediate stages, and 1. and an aggregate value calculation circuit that calculates an aggregate value.

As shown in FIG. 8, the conventional aggregation calculation device first sorts the data to be subjected to aggregation calculation (group-by attribute, aggregation attribute) according to the group-by attribute, and then sorts it.

Focusing on the group-by attribute, aggregate value calculation is performed using a dedicated circuit. On the other hand, since the aggregation processing device of the present invention does not presuppose data sorting processing, a sorting processing mechanism is not required, and the aggregation processing mechanism is also significantly different from that of the conventional aggregation processing device as follows. That is, to identify a group, the conventional technology uses a comparator of the group-by attribute value, whereas the present invention has a built-in associative memory device that can link a large number of associative memory LSIs. To calculate the aggregate value, the prior art uses an aggregation unit and a duplication remover.
In contrast, this invention uses a dedicated simple aggregate value calculation circuit.

Conventional technology uses a register to store the current aggregated value to store the aggregated value at an intermediate stage.
The present invention uses an aggregate value table made up of random access memory.

In addition, the present invention not only performs aggregation operations on relational databases that have been known in the past, such as specifying group-by attributes and aggregate attributes in the same relation, but also performs group-by
For extended aggregation operations that specify the -by attribute in a separate relation, the group identification processing is executed in high parallelism using an associative memory device, and the aggregation operation mechanism is executed at high speed using a dedicated circuit. Pipeline processing allows high-speed processing.

Embodiment FIG. 1 is a diagram showing the entire embodiment of the present invention, in which the input/output system 1 receives data to be aggregated from an aggregation calculation request source and sends aggregation results to the aggregation calculation request source. , the processing control system 2 controls the aggregate value calculation processing. In addition to the input/output system 1 of the processing control system 2, an associative memory device 3 configured to be able to link a large number of associative memory LSIs, and a dedicated aggregate value calculation process 4 for calculating aggregate values one after another. An aggregate value table 5 consisting of a random access memory for storing the results of aggregation [direct calculations] is connected.

FIG. 2 shows a specific example of the aggregate value calculation circuit 4 in FIG.

Input data to the circuit is stored in areas 11-15, and the contents of area 111 are combined with the contents of areas 12, 13 and comparator 16.
17, the contents of area 11 are added to the contents of area 14 by adder 18i, and the contents of area 15 are added to the constant value II 131 of area 19 by adder 18b. In the comparators 16 and 17, each output from the adders 18a and 18b is stored in areas 20 to 23, respectively, as output data from the circuit.

Hereinafter, the operation of the embodiment of the present invention will be explained by dividing it into a conventional aggregation operation and an extended aggregation operation, using specific examples. The relations used in the specific example are shown in FIG.

(a) Operation in conventional aggregation calculations An aggregation calculation request arrives at the input/output system 1 together with data to be aggregated from an aggregation calculation request source. For example, when the above-mentioned example of part-time employee (department name, count) is applied to the data in FIG. 3, the data of the "department name" attribute of the "part-time employee" relation is entered in order from the top. ``Clothing'', ``food'', ``food''
,... and so on.

The processing control system 2 receives the group-by attribute data one by one from the input/output system 1 and searches the associative memory device 3. If there is no match as a result of the search, the gr
The data of the oup-by attribute is registered in the associative memory device 3. Group numbers are sequentially assigned to the groups to be registered in the associative memory device 3 starting from 1. Assume that the group-by attribute data has a fixed length.

The group number is the address (sequence number) of the associative memory.
It can be easily calculated by dividing the data of the roup-by attribute by the number of words of the associative memory used. In parallel with the registration process, the entry in the aggregate value table 5 corresponding to the group number to be registered is initialized (zero cleared) and sent to the aggregate value calculation circuit 4, where the aggregate value is calculated and output. The result (new intermediate stage aggregate value for the group) is written again to the same entry in the aggregate value table 5.

If there is a match as a result of the search, the group number is known, the corresponding entry in the aggregated value table 5 (containing the intermediate aggregated value of the group) is read out, and the aggregated value calculation circuit 4 is loaded. Then, the aggregate value is calculated and the output result (new aggregate value at the intermediate stage of the group) is written again into the same entry in the aggregate value table 5.

The operation of the aggregate value calculation circuit 4 will be explained with reference to FIG.

Aggregate attribute data is input into area 11. The entry data of the aggregate value table 5 is input into the input areas 12, 13, 14, and 15 in the order of minimum value, maximum value, sum, and count. The input data is sent to comparators 16, 17 and adder 18a.
, 18b, and the calculation results are the arithmetic value, the maximum value, the sum,
Output area 20, 21, 22, 23 respectively in order of count
is output from.

In the above example, "clothing" is first processed, and after being searched, it is registered in the associative memory device 3, and the count value is 1".
'become. Next, 6 foods "" are processed, and after being searched, they are also registered in the associative memory device 3, and the count value becomes I I+.

Next, "Food" is processed, and after searching, group number = 2 is found, and the count value becomes 2''.Next, "Food" is processed, and after searching, group number = 2 is also found, and the count value is becomes 3”.

As a result of the aggregate value calculation process progressing in this manner, the content of the associative memory device 3 and aggregate value table 5 becomes as shown in FIG. Further, a time chart of the progress of the aggregate value calculation process is shown in FIG. The registration process in the associative memory device 3 involves the initial group-by attribute data ("Clothing" 5 "Food"
It can be seen that this is done when processing (1st, 2nd, and 9th data processing). It can also be seen that group identification/registration processing and aggregate value calculation processing are pipelined.

After the aggregate value calculation process is completed, the processing control system 2 returns the aggregate calculation result to the aggregate calculation request source via the input/output system 1.

(b) Operation in extended aggregation operation The operation in extended aggregation operation differs from the operation in conventional aggregation operation only in the operation of aggregate value calculation processing, so only that will be described. It consists of two phases: a registration phase in the memory device 3 (phase 1) and a group identification + aggregation calculation phase (phase 2).

In phase 1, the associative memory device 3 is searched for different data (groups) among the group-by attribute data, and the headings are registered. If we apply the previous example of part-time employees [department name, count] (organization [department name]) to the data in Figure 3, in Phase 1, we will have ``food'', ``clothing'', 6 personnel, and ``accounting''. Data is the 6th
As shown in the figure, it is registered in the associative memory device 3.

In phase 2, for the data to be aggregated, the associative memory device 3 in which groups have already been registered is searched for group identification, and an aggregate value is calculated for the entry in the aggregate value table 5 corresponding to the group number found there. . In the above example, data on the "department/department" attribute of "part-time employee" Relayon, "clothing", "food 71% food", etc.
. . are processed in order to obtain the contents of the aggregated value table shown in FIG.

FIG. 7 shows a time chart of the progress of processing in both phases. It can be seen that in Phase 2, group identification processing and aggregate value calculation processing are regularly pipelined.

Furthermore, by comparing FIG. 4 and FIG. 6, it can be seen that the extended aggregation operation generates the tuple of "Human Resources Department" that was missing in the conventional aggregation operation.

"Effects of the Invention" As explained above, according to the present invention, the aggregation (aggregation) of conventionally known relational databases that specifies group-by attributes and aggregate attributes in the same relation
ate) Extended aggregation operations specified by different lanes for operations and group-by attributes are executed in high parallelism using an associative memory device in the case of group identification processing, and a dedicated circuit is used for aggregate value calculation processing. By executing at high speed and pipeline processing both, it is possible to process at high speed, and solves the problem of special processing mechanism and processing load for sort processing, which was an essential route for conventional aggregation processing units. It becomes possible to do so.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the invention. FIG. 2 is a block diagram showing a specific example of the aggregation result 4 in FIG. 1, FIG. 3 is a diagram showing a specific example of the calculation used in the explanation, and FIG. FIG. 5 is a diagram showing the operation time chart in the conventional aggregation operation according to the present invention. FIG. 6 is a diagram showing the operation result in the extended aggregation operation according to the present invention. FIG. 7 is a diagram showing the operation result in the extended aggregation operation according to the present invention. FIG. 8 is a block diagram showing a conventional aggregation operation device. Patent applicant: Nippon Telegraph and Telephone Corporation Agent: Takuki Kusano 1 (
) Aggregate calculation @ Suemoto 10 2 figures aggregation @ Calculation same magazine 4;

Claims (1)

[Claims] (1) An input/output system that exchanges data with the aggregate calculation request source, a processing control system that controls aggregate calculation processing, an associative memory device that can link a large number of associative memory LSIs, and stores aggregated values at intermediate stages. It consists of an aggregate value table that calculates the aggregate value, and an aggregate value calculation circuit that calculates the aggregate value, and performs an aggregate operation that specifies a group-by attribute and an aggregate attribute in the same relation, and an extended aggregation operation that specifies the group-by attribute in a separate relation. The processing control system executes the group identification processing in parallel using the associative memory device, the aggregate value calculation processing using the aggregate value calculation circuit, and pipeline processing of both. Associative aggregation computing device configured.