JP6956043B2 - Arithmetic logic unit and search method - Google Patents

Description

The present invention relates to an arithmetic unit and a search method.

Conventionally, techniques for retrieving various types of information are known. As an example of such a technique, the multidimensional space in which the information to be searched (hereinafter referred to as "search target information") is arranged is divided into a plurality of areas, and information similar to the query information is queried. A technique for searching from an area corresponding to information is known.

Japanese Unexamined Patent Publication No. 10-124528

"Graph division and eigenvalue problem", Tatsuro Kawamoto, Journal of Japanese Neural Network Society Vol. 21, No. 4 (2014), 162-169 <Internet> https://www.jstage.jst.go.jp/article/jnns/21/4/21_162/_pdf (Searched on May 1, 2018) "Paging method" <Internet> http://mt-net.vis.ne.jp/ADFE_mail/0126.htm (Searched on May 1, 2018) "How FPGA works and software and hardware used for development", Masahiko Tange <Internet> http://toragi.cqpub.co.jp/Portals/0/backnumber/2009/03/p096-097.pdf (May 2018) Search on the 1st of the month)

However, in the above-mentioned technique, there is room for improving the efficiency in the search process.

For example, there is known a neighborhood search technique for searching search target information having a similar concept to query information among search target information having a graph structure. Even in such a neighborhood search, it is considered that the efficiency of the search can be improved by limiting the search target information to be searched. However, in the above-mentioned technique, it is difficult to appropriately divide the graph structure because the multidimensional space is simply divided into a plurality of regions. Therefore, it may not always be possible to improve the efficiency of the search process.

The present application has been made in view of the above, and an object of the present application is to improve the efficiency of search processing.

The arithmetic unit according to the present application is a reception unit that receives query information and a plurality of pages set in the main storage device, and is a divided graph in which a graph of search target information is divided into areas estimated as a range of neighborhood search. Among the pages in which is registered, the divided graph is obtained from a specific part that specifies the page information indicating the page in which the divided graph corresponding to the query information is registered and the page indicated by the page information specified by the specific part. It is characterized by having a reading unit to be read and a search unit for searching information similar to the query information by using the divided graph read by the reading unit.

According to one aspect of the embodiment, the efficiency of the search process can be improved.

FIG. 1 is a diagram illustrating an example of a search process executed by an arithmetic unit. FIG. 2 is a diagram showing an example of information on the divided graph according to the embodiment. FIG. 3 is a diagram showing an example of information registered in the search target database according to the embodiment. FIG. 4 is a diagram showing an example of information registered in the page table. FIG. 5 is a flowchart illustrating an example of the flow of the determination process according to the embodiment.

Hereinafter, the arithmetic unit according to the present application and the embodiment for carrying out the search method (hereinafter, referred to as “the embodiment”) will be described in detail with reference to the drawings. The arithmetic unit and the search method according to the present application are not limited by this embodiment. Further, in each of the following embodiments, the same parts are designated by the same reference numerals, and duplicate description is omitted.

[Embodiment]
[1. Example of arithmetic unit]
First, an example of the search process executed by the arithmetic unit will be described with reference to FIG. In the following description, an example of a process of searching for information similar to query information by neighborhood search will be described. More specifically, in the following description, an example of a process of executing a neighborhood search will be described using a graph in which nodes corresponding to the search target information are connected according to the similarity between the search target information.

FIG. 1 is a diagram showing an example of a functional configuration of the arithmetic unit according to the embodiment. In FIG. 1, the arithmetic unit 100 is an arithmetic unit that executes the search process described below. Here, the arithmetic unit 100 is an integrated circuit in which a user can define or change an internal logic circuit after manufacturing, and is a so-called PLD (Programmable Logic Device). More specifically, the arithmetic unit 100 is realized by an FPGA (Field-Programmable Gate Array). Further, the arithmetic unit 100 is connected to the external device OA and the main storage device 200.

The main storage device 200 is a storage device that stores various types of data, and is realized by, for example, a storage device such as a semiconductor memory element such as a RAM (Random Access Memory) or a flash memory (Flash Memory).

Here, a plurality of storage areas called pages are set in the main storage device 200, and data is read and written in page units. For example, pages m1 to m4 are set in the main storage device 200, and divided graphs G1 to G4, which are divided graphs used for neighborhood search, are registered in each of the pages m1 to m4. Further, the main storage device 200 holds a search target database 201 in which search target information is registered, in addition to the divided graphs G1 to G4.

For example, FIG. 2 is a diagram showing an example of information on the divided graph according to the embodiment. As shown in FIG. 2, the division graph G1 includes items such as "node ID", "data ID", "edge information", and "connection destination". Here, the "node ID" is an identifier for identifying a node included in the divided graph. Further, the "data ID" is an identifier that identifies the search target information corresponding to the node indicated by the associated "node ID". Further, the edge information is an identifier that identifies an edge that connects the nodes. Further, the "connection destination" is a node ID of another node connected by an edge to the node indicated by the associated "node ID".

For example, in the example shown in FIG. 2, information such as the node ID “N1”, the data ID “D1”, the edge information “E12”, and the connection destination “N2” are registered in association with each other. In such information, the node indicated by the node ID "N1" is a node corresponding to the search target information indicated by the data ID "D1" among the search target information, and the connection destination is provided by the edge indicated by the edge information "E12". Indicates that the node indicated by "N2", that is, the node indicated by the node ID "N2" is connected.

Further, FIG. 3 is a diagram showing an example of information registered in the search target database according to the embodiment. As shown in FIG. 3, the "data ID" and the "search target information" are registered in association with each other in the search target database 201. Here, the "search target information" is data of the search target information. In the example shown in FIG. 3, a conceptual value such as "data # 1" is described as the search target information, but in reality, various data of the search target information are registered.

For example, in the example shown in FIG. 3, the data ID “D1” and the search target information “data # 1” are registered in association with each other. Such information indicates that the search target information indicated by the data ID "D1" is "data # 1".

Returning to FIG. 1, the explanation will be continued. For example, the arithmetic unit 100 includes a processor 110, an input / output device 120, a memory controller 130, a storage unit 140, and an FPGA 150. The processor 110 is a processor included in the arithmetic unit 100, and is, for example, a processor or a microprocessor that employs an ARM architecture or a POWER architecture. Then, the processor 110 executes various arithmetic processes in cooperation with the FPGA 150.

For example, the processor 110 has a processor core 111 and a cache memory 112. The processor core 111 is a so-called core that realizes logical operations and four rules operations, and is realized by an arithmetic logic unit composed of an ALU (Arithmetic Logic Unit). The cache memory 112 is an auxiliary storage device included in the processor 110. More specifically, the cache memory 112 is a storage device that can be accessed by the processor core 111 at a higher speed than the main storage device 200, and is a so-called cache memory.

The input / output device 120 is a device that relays communication between the arithmetic unit 100 and the external device OA, and is a so-called I / O (Input Output) device. For example, the input / output device 120 includes USB (Universal Serial Bus), Ethernet, SD (Secure Digital), UART (Universal Asynchronous Receiver / Transmitter), SPI (Serial Peripheral Interface), I2C, GPIO (General-purpose input / output). ) Etc., which is realized by various input / output devices that control communication with the external device OA according to various communication standards.

The memory controller 130 controls the memory access to the main storage device 200 by the arithmetic unit 100. More specifically, the memory controller 130 reads or writes data stored in the main storage device 200 by a paging method.

The storage unit 140 is realized by a storage device such as a semiconductor memory element such as a RAM (Random Access Memory) or a flash memory (Flash Memory). The storage unit 140 has a page table 141. In the page table 141, the virtual address and the physical address are registered in association with each other.

For example, FIG. 4 is a diagram showing an example of information registered in the page table. As shown in FIG. 4, in the page table 141, the "virtual page number" and the "physical page number" are registered in association with each other. Here, the "virtual page number" is a virtual address used by the processor core 111 when designating a storage area in which data to be subjected to various arithmetic processes are registered. The "physical page number" is a physical address indicating a storage area of the main storage device 200.

For example, in the example shown in FIG. 4, the virtual page number “page s1” and the physical page number “page m1” are registered in association with each other. Such information means that, for example, the information designated by the processor core 111 using "page s1" is stored in the area designated by "page m1" in the storage area of the main storage device 200. Is shown. In the page table 141, for example, the identifier of the application that executes the arithmetic processing using the storage area indicated by each virtual page number may be registered.

Returning to FIG. 1, the explanation will be continued. For example, when the configuration shown in FIG. 1 is provided, the processor core 111 accesses the cache memory 112 and attempts to acquire data to be calculated. Then, when the data is not stored in the cache memory 112 (that is, a cache miss), the processor core 111 notifies the memory controller 130 of the address of the storage area in which the data to be calculated is stored. .. More specifically, the processor core 111 notifies the memory controller 130 of the virtual address used for the arithmetic processing.

In such a case, the memory controller 130 refers to the page table 141 in which the virtual address is associated with the physical address indicating the storage area in which the information corresponding to the virtual address is stored in the storage area of the main storage device 200. Then, specify the physical address to be accessed. Then, the memory controller 130 reads out the data stored in the storage area indicated by the specified physical address in the storage area of the main storage device 200. After that, the memory controller 130 associates the read data with the virtual address and registers it in the cache memory 112.

When the memory controller 130 writes back the data registered in the cache memory 112 to the main storage device 200 (writeback), the memory controller 130 refers to the page table and associates the virtual address with the information to be written back. And identify the corresponding physical address. Then, the memory controller 130 stores the information to be written back in the storage area indicated by the specified physical address in the storage area of the main storage device 200.

The FPGA 150 is an arithmetic unit whose circuit configuration can be changed according to an instruction from the external device OA. For example, the FPGA 150 includes digital circuit elements and analog circuit elements. For example, the FPGA 150 has a plurality of logical blocks that are programmable logical components, and the logical blocks are interconnected by reconfigurable wiring. Then, the FPGA 150 can realize various processes by hardware by changing the connection of the wiring between the logical components by the hardware description language (HDL) received from the external device OA.

Here, the FPGA 150 has the following functional configuration by configuring the circuit according to the HDL received from the external device OA. Then, the FPGA 150 executes a process of searching the search target information for information similar to the query information received from the external device OA by using the following functional configuration. More specifically, the FPGA 150 realizes efficient neighborhood search execution by using the following functional configuration.

For example, the FPGA 150 has a division unit 161, a generation unit 162, a reception unit 163, a specific unit 164, and a search control unit 165.

The division unit 161 divides the graph of the search target information into a plurality of division graphs based on the graph Laplacian. More specifically, the division unit 161 divides the graph of the search target information into a division graph divided for each area estimated as the range of the neighborhood search.

For example, the division unit 161 receives the search target information to be the target of the search process via the input / output device 120. In such a case, the division unit 161 registers the received search target information in the search target database 201. The search target database 201 may be directly registered in the main storage device 200 from the external device OA via the input / output device 120 and the memory controller 130.

Further, the division unit 161 receives a graph in which a plurality of nodes corresponding to the search target information are connected by edges according to the similarity of the search target information from the external device OA. That is, the division unit 161 accepts a graph showing the similarity of the search target information. It should be noted that such a graph may be generated by an arbitrary graph generation technique for generating a graph used for neighborhood search, such as an approximate k-nearest neighbor graph.

In such a case, the division unit 161 divides the graph into a plurality of graphs. More specifically, the division unit 161 estimates the range that can be the target of the neighborhood search when the neighborhood search using the accepted graph is performed, and generates a division graph in which the graph is divided for each estimated range. ..

For example, in the graph used for the neighborhood search, each node is connected according to the similarity between the search target information corresponding to each node. Therefore, it is considered that a plurality of similar search target information and a plurality of corresponding nodes are easily connected by an edge. On the other hand, in order to divide the graph appropriately, the graph is divided into a plurality of divided graphs so that the number of edges of the divided graph feeling is as small as possible and the size (number of nodes) of each divided graph is as equal as possible. Various methods for dividing have been proposed. When the graph is divided into a plurality of divided graphs in this way, it is considered that search target information similar to each other is likely to be included in the same divided graph. Then, the divided graph is considered to show the search target information that is the search range when the neighborhood search is performed.

Therefore, the division unit 161 divides the graph into a plurality of graphs. For example, the division unit 161 generates an adjacency matrix showing the entire graph, and sets a graph Laplacian based on the generated adjacency matrix. Then, the division unit 161 generates a division graph in which the graph of the search target information is divided for each estimated search range using the graph Laplacian (step S1). For example, the division unit 161 divides the graph into a plurality of division graphs so as to minimize the evaluation function called RioCut and the evaluation function called Ncut (normalized-cut). That is, the division unit 161 divides the graph by a method called spectral clustering.

For example, the division unit 161 divides the graph into a plurality of division graphs by using the method disclosed in Non-Patent Document 1 described above. To give a more specific example, the division unit 161 divides the graph into four division graphs G1 to G4. As a result, when a certain query information is acquired, the division unit 161 can generate a division graph showing a range of search target information for which it is necessary to determine whether or not it is similar to the query information. That is, the division unit 161 can generate a division graph showing the range of the neighborhood search.

The generation unit 162 generates a page table 141 in which the physical page number of the main storage device in which each divided graph is registered and the virtual page number used for the search process are associated with each other. For example, the generation unit 162 instructs the memory controller 130 to set a storage area for registering each divided graph as a storage area used by the application that executes the search process. In such a case, the memory controller 130 refers to the page table 141, identifies the virtual page number and the physical page number that are not used by other applications, and associates the specified virtual page number with the physical page number. And register it in the page table 141.

Further, the generation unit 162 registers the division graph in the main storage device 200 by using the virtual page number assigned from the memory controller 130. For example, the generation unit 162 provides the allocated virtual page number and the data of the divided graph to the memory controller 130, and registers the data of the divided graph in the storage area indicated by the physical page number corresponding to the virtual page number. For example, the generation unit 162 registers the divided graph G1 in the storage area where the physical page number is "page m1", registers the divided graph G2 in the storage area where the physical page number is "page m2", and registers the divided graph G3 in the storage area where the physical page number is "page m2". Is registered in the storage area where the physical page number is "page m3", and the divided graph G4 is registered in the storage area where the physical page number is "page m4".

The generation unit 162 does not need to register the data of one divided graph on one page. For example, the generation unit 162 may register the data of one division graph on a plurality of pages depending on the data length of the data of one division graph. As a result of such processing, when the generation unit 162 indicates the physical address indicating the storage area of the main storage device 200 in which each division graph is registered and the storage area of the main storage device 200 in which each division graph is registered, the generation unit 162 indicates. A page table 141 associated with the virtual address used by the processor core 111 can be created.

Then, the FPGA 150 executes a neighborhood search using the divided graph. For example, the reception unit 163 receives the query information. More specifically, the reception unit 163 receives the query information from the external device OA via the input / output device 120. In such a case, the reception unit 163 provides the query information to the specific unit 164.

The specific unit 164 is a plurality of pages set in the main storage device, and among the pages in which the divided graph in which the graph of the search target information is divided into the areas estimated as the range of the neighborhood search is registered, the query information. Identify the page information indicating the page in which the corresponding split graph is registered. That is, the specifying unit 164 specifies the page information indicating the page in which the divided graph corresponding to the query information is registered, among the pages in which the divided graph divided based on the graph Laplacian is registered. For example, the identification unit 164 specifies the virtual page number of the page in which the division graph corresponding to the query information is registered as the page information.

For example, in the FPGA 150, the logic block includes an input / output block that controls the input / output of the FPGA 150 and a logic cell that executes various arithmetic processes. Such a logic cell is composed of a LUT (Look Up Table) and a register, and outputs data obtained by converting the input data based on the LUT (see, for example, Non-Patent Document 3). Therefore, the FPGA 150 specifies the virtual page number of the page in which the input query information and the corresponding divided graph are registered by rewriting the LUT of the logic cell. More specifically, the FPGA 150 rewrites the LUT so as to output the virtual page number in which the divided graph of the range to be searched is registered when the data string of the query information is input, so that the specific unit 164 To realize.

For example, the generation unit 162 specifies the range of query information (hereinafter, referred to as “corresponding query information”) for performing a neighborhood search using the divided graph for each divided graph. For example, the generation unit 162 may set the range of the search target information corresponding to the node included in the node to the range of the corresponding query information. Further, the generation unit 162 selects a node to be the root node from the nodes included in the divided graph, and specifies the range of the corresponding query information in which the similarity between the selected node and the corresponding search target information falls within a predetermined range. You may.

Then, when the range of the corresponding query information of a certain divided graph is input, the generation unit 162 generates a LUT that outputs the virtual page number of the storage area in which the divided graph is registered. That is, the generation unit 162 generates a LUT for decoding the query information data into a virtual page number indicating a storage area in which a divided graph including a search target information similar to the query information and a corresponding node is registered. do. Then, the generation unit 162 is set in the logic cell that operates as the specific unit 164. As a result of such processing, the identification unit 164 can specify the virtual page number of the divided graph corresponding to the input query information.

Subsequently, the search control unit 165 operates the processor core 111 as a search unit for executing the search process to search for search target information similar to the query information. For example, the search control unit 165 provides the processor core 111 with the virtual page number specified by the specific unit 164, and causes the processor core 111 to read the divided graph registered in the storage area indicated by the virtual page number. Then, the search control unit 165 causes the processor core 111 to search for search target information similar to the query information by using the divided graph.

For example, the search control unit 165 provides the processor core 111 with the virtual page number specified by the specific unit 164. In such a case, the processor core 111 provides the virtual page number to the memory controller 130. Then, the memory controller 130 specifies the physical page number associated with the virtual page number from the page table 141, and reads out the divided graph registered in the storage area indicated by the specified physical page number. Then, the memory controller 130 registers the read division graph in the cache memory 112.

Further, the search control unit 165 provides the query information to the processor core 111. The search control unit 165 may provide the query information to the processor core 111 via the cache memory 112. In such a case, the processor core 111 reads the division graph from the cache memory 112, and reads the data of the search target information corresponding to the node included in the read division graph from the search target database 201. Then, the processor core 111 identifies data similar to the query information among the read data, and transmits the identified data to the external device OA via the input / output device 120.

As described above, the arithmetic unit 100 includes an FPGA 150 which is a first arithmetic unit capable of setting a process to be executed by a program, and a processor 110 which is a second arithmetic unit which executes a predetermined instruction set. Then, the FPGA 150 operates as a reception unit 163 and a specific unit 164, and the processor 110 operates as a search unit that executes a search process. Here, the FPGA 150 divides the graph showing the similarity of the search target information into a plurality of divided graphs that can be estimated as the range of the neighborhood search, and registers the divided graphs on individual pages. Then, the FPGA 150 identifies the division graph corresponding to the query information, and reads out the information of the page in which the specified division graph is registered.

As a result of such processing, the arithmetic unit 100 can execute the neighborhood search by using the divided graph showing the search target information estimated to be the search range among all the search target information. As a result, the arithmetic unit 100 can improve the efficiency of the search process.

[2. An example of the flow of processing executed by the arithmetic unit]
Next, an example of the flow of arithmetic processing executed by the arithmetic unit 100 will be described with reference to FIG. FIG. 5 is a flowchart illustrating an example of the flow of the determination process according to the embodiment. First, the arithmetic unit 100 acquires a graph of search target information (step S101). In such a case, the arithmetic unit 100 estimates the range of the neighborhood search and generates a divided graph divided for each estimated area (step S102). Then, the arithmetic unit 100 registers the divided graph on each page of the main storage device 200 (step S103).

Further, the arithmetic unit 100 generates a page table 141 in which the physical page number of the main storage device 200 and the virtual page number used for the arithmetic processing are associated with each other (step S104). Then, the arithmetic unit 100 determines whether or not the query information has been accepted (step S105), and if not (step S105: No), waits until the query information is accepted. Further, when the arithmetic unit 100 receives the query information (step S105: Yes), the arithmetic unit 100 specifies the divided graph corresponding to the query information (step S106). For example, the arithmetic unit 100 uses a LUT that associates a virtual page number indicating a storage area in which each divided graph is registered with a range of query information in which the search target information indicated by each divided graph is to be searched. Then, the corresponding virtual page number is specified from the query information.

Further, the arithmetic unit 100 reads out the corresponding divided graph using the page table (step S107). Then, the arithmetic unit 100 registers the read division graph in the cache memory of the processor (step S108), executes the neighborhood search process (step S109), and ends the process.

[3. Modification example]
In the above, an example of the search process by the arithmetic unit 10 has been described. However, the embodiment is not limited to this. Hereinafter, variations of the search process executed by the arithmetic unit 10 will be described.

[3-1. Device configuration〕
In the above-mentioned example, an example of the search process executed by the arithmetic unit 100 including the FPGA 150 has been described. However, the embodiment is not limited to this. For example, the above-mentioned search process may be realized by a processor such as a CPU (Central Processing Unit) or an MPU (Micro Processing Unit) in addition to the FPGA 150. Further, the above-mentioned search process may be realized by an integrated circuit such as an ASIC (Application Specific Integrated Circuit).

That is, the above-mentioned search process is an arithmetic processing unit that does not have a programmable logic circuit, and may be realized by various arithmetic processing units that use a paging method. Such an arithmetic processing unit generates, for example, a graph showing the similarity of search target information by dividing it into a range that can be estimated as a search target, and registers the generated graphs on individual pages. As a result of such processing, the arithmetic processing unit can appropriately limit the range to be searched and speed up the identification of the range to be searched.

Further, the arithmetic unit 100 does not have the division unit 161 and may not divide the graph. For example, when the arithmetic unit 100 receives a divided graph that has been divided using a graph Laplacian or the like from the external device OA, the arithmetic unit 100 registers the received divided graph in each page of the main storage device 200, and such a divided graph. You may perform a search using.

[3-2. others〕
Further, among the processes described in the above-described embodiment, all or a part of the processes described as being automatically performed can be manually performed, or the processes described as being manually performed can be performed. All or part of it can be done automatically by a known method. In addition, the processing procedure, specific name, and information including various data and parameters shown in the above text and drawings can be arbitrarily changed unless otherwise specified. For example, the various information shown in each figure is not limited to the illustrated information.

Further, each component of each of the illustrated devices is a functional concept, and does not necessarily have to be physically configured as shown in the figure. That is, the specific form of distribution / integration of each device is not limited to the one shown in the figure, and all or part of the device is functionally or physically dispersed / physically distributed in any unit according to various loads and usage conditions. Can be integrated and configured.

In addition, the above-described embodiments can be appropriately combined as long as the processing contents do not contradict each other.

[4. effect〕
As described above, when the arithmetic unit 100 receives the query information, the arithmetic unit 100 divides the graph of the search target information into a plurality of pages set in the main storage device 200 for each area estimated as the range of the neighborhood search. Among the pages in which the split graph is registered, the page information indicating the page in which the split graph corresponding to the query information is registered is specified. Then, the arithmetic unit 100 searches for information similar to the query information by using the divided graph registered on the page indicated by the specified page information. Therefore, the arithmetic unit 100 can improve the efficiency in the search process.

Further, the arithmetic unit 100 specifies the page information indicating the page in which the division graph corresponding to the query information is registered, among the pages in which the division graph divided based on the graph Laplacian is registered. Therefore, the arithmetic unit 100 can execute the search process by using the divided graph showing the range of the search target information suitable as the comparison target with the query information.

Further, the arithmetic unit 100 includes an FPGA 150, which is a first arithmetic unit capable of setting a process to be executed by a program, and a processor 110, which is a second arithmetic unit that executes a predetermined instruction set. , The reception unit 163 and the specific unit 164, and the second arithmetic unit operates as a search unit that executes a search process. Therefore, the arithmetic unit 100 can improve the efficiency in the search process.

Further, the arithmetic unit 100 divides the graph of the search target information into a plurality of divided graphs based on the graph Laplacian. Further, the arithmetic unit 100 generates a page table 141 in which the physical page number of the main storage device 200 in which each divided graph is registered and the virtual page number used for the search process are associated with each other. Then, the arithmetic unit 100 specifies the virtual page number of the page in which the division graph corresponding to the query information is registered, and uses the page table 141 to specify the physical page number associated with the specified virtual page number. Then, information similar to the query information is searched for using the divided graph read from the page indicated by the specified physical page number. Therefore, the arithmetic unit 100 can improve the efficiency in the search process.

Although some of the embodiments of the present application have been described in detail with reference to the drawings, these are examples, and various modifications are made based on the knowledge of those skilled in the art, including the embodiments described in the disclosure column of the invention. It is possible to practice the present invention in other improved forms.

Further, the above-mentioned "section, module, unit" can be read as "means" or "circuit". For example, the generation unit can be read as a generation means or a generation circuit.

100 Arithmetic logic unit 110 Processor 111 Processor core 112 Cache memory 120 Input / output device 130 Memory controller 140 Storage unit 141 Page table 150 FPGA
161 Division unit 162 Generation unit 163 Reception unit 164 Specific unit 165 Search control unit 200 Main storage device 201 Search target database G1 to G4 Divided graph OA External device

Claims (5)

The reception department that accepts query information and
Of the pages in which the divided graphs are registered, which are a plurality of pages set in the main memory and the graph of the search target information is divided for each area estimated as the range of the neighborhood search, the divided pages corresponding to the query information. A specific part that identifies the page information indicating the page where the graph is registered,
An arithmetic unit including a search unit that searches for information similar to the query information by using the divided graph registered on the page indicated by the page information specified by the specific unit. The specific unit is characterized in that, among the pages in which the divided graph divided based on the graph Laplacian is registered, the page information indicating the page in which the divided graph corresponding to the query information is registered is specified. The arithmetic unit according to 1. The arithmetic unit includes a first arithmetic unit that can set a process to be executed by a program and a second arithmetic unit that executes a predetermined instruction set.
The first arithmetic unit operates as the reception unit and the specific unit, and operates as the reception unit.
The arithmetic unit according to claim 1 or 2, wherein the second arithmetic unit operates as the search unit. A division unit that divides the graph of the search target information into a plurality of division graphs based on the graph Laplacian, and
It has a generation unit that generates a page table in which the physical page number of the main storage device in which each division graph divided by the division unit is registered and the virtual page number used by the search unit for search processing are associated with each other. death,
The specific unit specifies the virtual page number of the page in which the division graph corresponding to the query information is registered as the page information.
The search unit uses the page table to specify the physical page number associated with the virtual page number specified by the specific unit, and uses the divided graph read from the page indicated by the specified physical page number. The arithmetic unit according to any one of claims 1 to 3, further comprising searching for information similar to the query information. It is a search method executed by the arithmetic unit.
The reception process for accepting query information and
Of the pages in which the divided graphs are registered, which are a plurality of pages set in the main memory and the graph of the search target information is divided for each area estimated as the range of the neighborhood search, the divided pages corresponding to the query information. A specific process to identify page information indicating the page where the graph is registered, and
A search method comprising a search step of searching for information similar to the query information by using the divided graph registered on the page indicated by the page information specified by the specific step.

Images