JP6291435B2 - Program and cluster system - Google Patents

Description

  The present invention relates to a program and a cluster system.

A distributed database (DB) exists as a database technology whose maximum goal is to obtain a highly scalable effect. A representative method for obtaining scalability is KVS (Key Value Store).
An RDB (Relational Data Base) exists as a database (DB) that is currently widely used. This RDB technology expresses data in a two-dimensional table structure consisting of Rows and Columns, and performs JOIN and normalization according to queries in the SQL (Structured Query Language) language. By executing it, it is possible to realize processing such as search. Various applications are often created using RDB for a variety of reasons, but as a particularly important element, not only a simple key search but also a function that can search by data attribute value (value) It can be said that it has a large place. However, in the conventional distributed system, although it can be applied to an application using a simple key, it cannot be applied to an application that manipulates a value, or even if it can, the scale cannot be changed. Moreover, it has been necessary to reexamine the data structure for the distributed platform (PF).
In the field of research on distributed systems, in view of the above, a lot of studies are being made from NoSQL for simple key search to NewSQL having an operation capability similar to that of RDB including attribute value search.

As a representative method for enabling attribute value search in a distributed system, there is a transposed index method such as Secondary Index, and representative distributed DBs such as Cassandra and Mercury have been adopted (see Non-Patent Documents 1 and 2). ). However, they basically use multiple hash functions, place the master data at the output location obtained by applying the key to the main hash function, and apply the attribute value you want to search to another hash function. This is a mechanism to place replica data at the output location obtained.
There is also a technique called MerDy that uses replica data as a link to improve the efficiency of data capacity.

FIG. 13 is a diagram for explaining MerDy.
MerDy assigns an identifier (hash value) indicating the location of data in a distributed DB that enables search by attribute value using an inverted index (for attribute search). As a method for determining the arrangement of the data arrangement in the distributed DB to the server, the hash value is applied to the attribute tag or attribute value of the data to make an identifier. On the other hand, an identifier area for each server is assigned.

  As shown in FIG. 13, the master data (v0, v1, v2) is placed in the output server obtained by applying h0 (key) to the main hash function (this server functions as a storage) and searched. Replica data (v0, v1, v2) and (v0, v1, v2) are placed at the output location obtained by applying the desired attribute value to another hash function. Note that placing two replica data is an example, and is determined according to the number of redundancy.

The main hash function is h0 (key) searched by key, and is a discontinuous hash function such as MD5 (Message Digest Algorithm 5). Master data (v0, v1, v2) is placed in this h0 (key).
The other hash function is a “hash function for searching an attribute value”, and h1 (v0) searched by an attribute value different from the main hash function applied to the attribute value to be searched. ), h2 (v1), h3 (v2). For example, h1 (v0) is a continuous hash function such as character code conversion, h2 (v1) is a continuous hash function such as the number of characters, and h3 (v2) is a discontinuous hash function such as SHA1 (Secure Hash Algorithm 1). It is. Note that the key of the inverted index (for attribute search) may be represented by k1, k2,.
In Mercury and MerDy described in Non-Patent Documents 1 and 2 described above, a hash function is applied to an attribute value in the same way as a key, and the hash value is placed in a spatial position such as a consistent hash (Consistent Hashing). Store the inverted index in the corresponding server.

Daisuke Kawakami, et al .: “Distributed data store adapted to range search and multi-attribute data processing”, IPSJ SIG [System Software and Operating System] 2010-OS-113 (10), 1-8, 2010 -01-20. Cassandra secondary index, [online], [Search July 21, 2014], Internet <URL: http: //books.google.co.jp/books? Hl = en & lr = & id = MKGSbCbEdg0C & oi = fnd & pg = PR7 & dq = Cassandra + secondary + index & ots = XpPC2yy91A & sig = oCxd9a_Gvrp4cbfpidKEEVVMW6Y # v = onepage & q = Cassandra% 20secondary% 20index & f = false>

  There are several methods for searching by using a product as a condition when searching for data distributed based on KVS using attribute values instead of key values. For example, this is a conditional hash method in which a hash value is calculated by creating a key with a character string including a matching method and a logical expression by collecting indexes that match a single attribute condition and matching them. However, the former matching method has a problem that the amount of calculation at the time of searching is large. In the latter condition hash method, it is necessary to search for hash values for all combinations of logical expressions at the time of index creation. Therefore, if the number of attributes increases, the amount of index increases and the indexing load increases. There was a problem. Furthermore, the latter has a problem in terms of index consistency, for example, when an attribute value is changed, it affects all indexes of logical expressions including the attribute value.

  The present invention has been made in view of such a background, and an object of the present invention is to provide a program and a cluster system that can reduce the load during search and the load of indexing.

  In order to solve the above-described problem, the invention described in claim 1 applies a hash function for retrieving attribute values in a distributed database (DB) using KVS (Key Value Store), and hashes obtained. In the server corresponding to the position of the value in the hash space, the computer as each of the servers constituting the cluster system that stores the inverted index, based on the hash function for searching the attribute value when creating the index, A program for functioning as a hash bit string creating means for creating a hash bit string of attributes corresponding to the attribute types of the attribute value, and a combining means for combining the hash bit strings of attributes corresponding to the attribute types in a predetermined connection order; did.

  According to the sixth aspect of the present invention, in the distributed database (DB) using KVS (Key Value Store), the hash function for retrieving the attribute value is applied, and the position of the obtained hash value in the hash space In the cluster system that stores the inverted index in the server corresponding to the above, the server, when creating the index, based on the hash function for searching for the attribute value, the hash of the attribute for the attribute type of the attribute value A cluster system comprising: a hash bit string creating unit that creates a bit string; and a coupling unit that joins the hash bit strings of attributes corresponding to the attribute types in a predetermined connection order.

  In this way, by using a value obtained by combining hash bit strings of attributes corresponding to the attribute types of attribute values, it is possible to reduce the load at the time of search and the load of indexing. That is, the indexing load is small because it does not depend on the number of combinations of product conditions. In addition, data capacity can be saved, consistency can be ensured, and flexible conditional search can be performed without having to worry about what is key when designing data.

  The invention according to claim 2 is characterized in that the combining means arranges and combines all attributes corresponding to the attribute types of the attribute values at least once at the head of the connection order. 1 as a program.

  By doing in this way, since the attribute with high appearance frequency comes first, the search speed of the entire product condition search is improved.

  The invention according to claim 3 is characterized in that the combining means rearranges the attribute types of the attribute values in descending order of appearance frequency, and combines the hash values calculated from the rearranged attribute values as a bit string. The program according to claim 1, which is characterized.

  By doing so, the search speed when an attribute tag that frequently appears is included in the conditional expression is improved.

  The invention according to claim 4 is characterized in that the combination means performs bit concatenation with the hash value of the attribute type of the attribute value having a high combination frequency of product conditions as a higher rank. Program.

  By doing in this way, the search speed in the conditional product of the combination of the attribute tag which appears frequently is improved.

  Further, the invention according to claim 5 is that the server selects an index having the attribute of the searched search condition in the head hash bit column at the time of searching, and executes a range search on the index. The program according to claim 1, which is characterized.

  In this way, when searching, an index having one of the condition attributes as the first bit is selected and a range search is performed on it. Search. Since there is no huge index matching, the search load is small.

  ADVANTAGE OF THE INVENTION According to this invention, the program and cluster system which can make small the load at the time of a search, and the load of indexing can be provided.

It is a figure which shows the whole structure containing the cluster system which concerns on this embodiment. It is a functional block diagram which shows the structure of the server which comprises the cluster system which concerns on this embodiment. (A) is a figure which shows an example of the attribute tag information which the attribute tag information storage means of the cluster system concerning this embodiment stores. (B) is a figure which shows an example of the output bit length of the hash function which the attribute tag information storage means of the cluster system concerning this embodiment stores. It is a figure explaining the structure of the bit string of the hash function for searching the attribute value of the cluster system which concerns on this embodiment. It is a figure explaining the bit connection indexing method of the hash value of the attribute of the cluster system concerning this embodiment. It is a figure explaining the bit connection indexing method at the time of the search of the cluster system which concerns on this embodiment. It is a figure explaining the bit connection indexing method at the time of the search of the cluster system which concerns on this embodiment. It is a figure which shows the example of a combination of the Hash value combined according to the appearance frequency of the product conditions in the search of the cluster system which concerns on this embodiment. It is a figure which shows the example of a combination of the Hash value combined according to the appearance frequency of the product conditions in the search of the cluster system which concerns on this embodiment. It is a figure which shows the example of a combination of the Hash value combined according to the combination frequency of the product conditions in the search of the cluster system which concerns on this embodiment. It is a figure which shows the example of a combination of the Hash value of the cluster system which concerns on this embodiment. It is a figure which shows the example of a combination of the Hash value of the cluster system which concerns on this embodiment. It is a figure explaining MerDy.

  Next, the cluster system 1 and the like in a mode for carrying out the present invention (hereinafter referred to as “the present embodiment”) will be described.

(System configuration of this embodiment)
The cluster system 1 according to the present embodiment will be specifically described.
FIG. 1 is a diagram showing an overall configuration including a cluster system according to the present embodiment.
As shown in FIG. 1, the cluster system 1 according to the present embodiment is connected to a client 2 that is an external system via a network. Then, input data from the client 2 is received, data is stored, updated, searched, etc. in the cluster system 1, and the result is transmitted to the client 2 as output data.

FIG. 2 is a functional block diagram showing the configuration of the server 11 constituting the cluster system 1 according to this embodiment.
As shown in FIG. 2, the server 11 includes a control unit 110, and the input / output unit 20 includes a memory unit 30 and a storage unit 40.

  The input / output means 20 inputs / outputs information between the client 2 and each server 11. The input / output unit 20 includes a communication interface that transmits and receives information via a communication line, and an input / output interface that performs input / output between an input unit such as a keyboard (not shown) and an output unit such as a monitor. Is done.

  The memory unit 30 includes a primary storage device such as a RAM (Random Access Memory), and temporarily stores information necessary for data processing by the control unit 110.

  The storage unit 40 includes a storage device such as a hard disk or a flash memory, and stores the ID (IP address) and the like of each server 11 in the cluster system 1.

  The control unit 110 controls the entire server 11. The information reception unit 111, the syntax analysis unit 112, the attribute tag information storage unit 113, the hash bit string creation unit 114, the combination unit 115, the search unit 116, and the information transmission unit 117. It is comprised including. The control means 110 is realized by, for example, developing a program stored in the storage means 40 in a RAM (memory means 30) by a CPU (Central Processing Unit).

  The information receiving unit 111 acquires input data from the client 2 and output data from another server 11 or the like via the input / output unit 20.

  The syntax analysis unit 112 receives input data from the information reception unit 111 and parses the contents of the input data. For example, the parsing unit 112 may search the input data for (a) key exact match search, (b) key range search, (c) attribute value (value) complete match search, and (d) value range search. Analyze whether it is (including substring search). Then, the syntax analysis unit 112 passes the analysis result to the hash bit string creation unit 114.

  For each attribute tag, the attribute tag information storage unit 113 stores a hash function applied to the attribute tag, a hash function applied to the attribute value, and a hash function applied to the key as attribute tag information in the storage unit 40 (FIG. 3). (See (a)). Also, each hash function individually stores how many bits of hash values are calculated. For example, for each attribute tag “Location” or “Boolean”, a set of hash functions having a predetermined output bit length is stored (see FIG. 3B).

Based on the hash function and bit length for each attribute tag acquired from the attribute tag information stored in the storage unit 40, the hash bit string creating unit 114 generates a “hash bit string obtained from the attribute value” (reference numeral in FIG. 4). (See (A)), “discontinuous hash bit sequence obtained from attribute tag” (see symbol (B) in FIG. 4), and “continuous hash bit sequence obtained from key” (see symbol (C in FIG. 4)). )) And create. Here, when the hash value is viewed in binary (represented in binary), it becomes a hash bit string.
The hash bit string creating unit 114 creates a hash bit string of attributes corresponding to attribute types of attribute values based on a hash function for retrieving attribute values when creating an index.

  The combining means 115 combines hash bit strings. Specifically, the combining unit 115 combines hash bit strings of attributes corresponding to attribute types in a predetermined connection order. The combining means 115 combines all the attributes corresponding to the attribute types of the attribute values at least once at the head of the connection order. In this case, the combining unit 115 cyclically replaces the attribute connection order.

  At the time of the search, the search means 116 selects an index having the attribute of the search condition that has been multiplied in the first hash bit string, and performs a range search on the index.

  The information transmission unit 117 transmits output data to the client 2 and other servers 11 via the input / output unit 20.

  The present embodiment is characterized in that the “hash function for searching attribute values” is configured by a combination of a plurality of hash bit strings. For example, exclusive functions such as discontinuous functions (use discontinuous hash functions (such as MD5) for one attribute value) and continuous functions (use numerical hash functions continuous hash functions used for character code conversion) Instead of using the alternative, use a hash value combined with a bit string and create an inverted index (for attribute search).

Hereinafter, the configuration of each bit string of the “hash function for retrieving an attribute value” will be described, and then a method for combining the bit strings will be described.
FIG. 4 is a diagram for explaining the configuration of the bit string of “hash function for retrieving attribute values”.
As shown in FIG. 4, the bit string of “hash function for retrieving attribute value” includes “continuous hash bit string obtained from attribute value” (see symbol (A) in FIG. 4) and “attribute tag”. The discontinuous hash bit sequence obtained from (see (B) in FIG. 4) and the “continuous hash bit sequence obtained from the key” (see (C) in FIG. 4).
Note that the hash bit sequence obtained from the attribute value indicated by the symbol (A) in FIG. 4 may be a discontinuous hash bit sequence. The bit length of each hash bit string will be described later with reference to FIG.

<Hash bit string obtained from the attribute value> (see symbol (A) in FIG. 4)
As shown in FIG. 4A, the hash bit string “00000101101101” obtained from the attribute value is the hash value of the transposed index (for attribute search). Note that the continuous hash bit sequence obtained from the attribute value indicated by the symbol (A) in FIG. 4 is also the hash value of the transposed index similar to the conventional example. Specifically, in the present embodiment, a hash function is applied to the attribute value in the same manner as the key, and a space such as a consistent hash of the hash value (the hash bit string indicated by the symbol (A) in FIG. 4) is used. The inverted index (for attribute search) is stored in the server corresponding to the upper position.

<Discontinuous hash bit string obtained from the attribute tag> (see symbol (B) in FIG. 4)
As shown in FIG. 4B, the discontinuous hash bit string “01001100101101” obtained from the attribute tag is a hash value obtained by applying the attribute tag to a discontinuous hash function such as MD5. . Take an example. If the “residence” is “Tokyo”, the attribute value is “Tokyo” and the attribute tag is “residence”. Note that “residence” being “Tokyo” is an “attribute”. This “attribute” is the content of the data associated with the key. Referring to FIG. 12, with respect to the master data v0 in FIG. 12, v0 is an attribute value (attribute), and the relationship between the attribute value v0 and the data is associated with the key. The attribute value (attribute) and the key are associated with each other in a one-to-one relationship, for example.

<Consecutive hash bit sequence obtained from key> (See reference (C) in FIG. 4)
As shown in FIG. 4C, the continuous hash bit string “111010101101000” obtained from the key is a hash value obtained by applying the key to the continuous hash function.

  Here, in general, when creating an inverted index, an attribute value is an attribute value, a key is used as a key, and a key is not combined with an attribute value to create a hash value as in this embodiment. . Naturally, there is nothing to combine the continuous hash bit sequence obtained from the attribute value and the discontinuous hash bit sequence obtained from the attribute tag, and it is obtained from the continuous hash bit sequence and key obtained from the attribute value. There is nothing that combines a continuous hash bit sequence.

Next, a bit connection indexing method will be described.
[Overview]
<When creating an index>
The bit string in this bit concatenated indexing method is applied to the portion indicated by reference numeral (A) in FIG.
In this embodiment, a value obtained by applying a hash function for determining a position in a space such as a consistent hash is combined with an attribute tag, an attribute value, and a key.
In particular, when creating an index, the hash values of all attribute values are calculated and bit-concatenated.
(1) At the time of index creation, hash values of different attribute types are arranged at the beginning between each bit string.
(2) Between each bit string, the value of the bit string is set so that the attribute connection order is cyclically changed. That is, the order of bit connection is cyclically changed so that any attribute is equally assigned to each bit area.

<When searching>
At the time of search, an index having one attribute of the search condition that has been multiplied as the first bit is selected, and a range search is performed on the index. Under any product condition, the region is searched in a small range, so there is no huge index matching and the search load is small.

This will be specifically described below.
[When creating an index]
FIG. 5 is a diagram for explaining a bit concatenation indexing method for the hash value of an attribute.
As shown in FIG. 5, the data 50 is master data (for example, master data v0 in FIG. 12) associated with the key. In FIG. 5, the key is “1234”, the attribute value is “Yamada” (the attribute tag is “last name”), “Taro” (the attribute tag is “first name”), “male” (the attribute tag is “sex”), And “Tokyo” (the attribute tag is “residence”).

<Create hash bit sequence>
When creating an index, the hash value of all attribute values is calculated.
Each attribute value “Yamada”, “Taro”, “Men”, and “Tokyo” are described in this order as attribute 0 (last name), attribute 1 (first name), attribute 2 (gender), and attribute 3 (residence). To do. A numerical value in a rectangle below the attribute 0 (last name), attribute 1 (first name), attribute 2 (gender), and attribute 3 (residence) is a hash bit string obtained from the attribute value. In this example, 8 bits are used, but any number of bits may be used. For example, a hash bit string “11010111” of attribute 0 (last name), a hash bit string “01110001” of attribute 1 (first name), a hash bit string “00010100” of attribute 2 (gender), and a hash of attribute 3 (residence) The bit string is “01011000”.

<bit concatenation of hash bit sequence>
As shown by a symbol a in FIG. 5, a hash value is generated by bit-connecting hash values of attribute values. In the case of FIG. 5A, “11010111”, “01110001”, “00010100”, and “01011000” are bit-connected.
Then, based on a value obtained by combining hash values calculated from a plurality of attribute values as a bit string, the transposed index is stored in the server 11 in the cluster.

<Change order of bit concatenation>
By switching the bit connection order cyclically, hash values of different attribute types are arranged at the beginning between each bit string.
For example, as shown in FIG. 5B, in the next order, the order of bit concatenation is changed so that the hash bit string “01110001” of attribute 1 (name) comes first. In this example, with this replacement, the hash bit string “11010111” of attribute 0 (last name), which has been the top so far, becomes the last. In the case of general forward matching, only the top is considered important.
By cyclically changing the bit connection order, as shown in FIG. 5C, in the next order, the hash bit string starting with the hash bit string “00010100” of attribute 2 (gender) is connected, Further, in the next order, as shown in FIG. 5D, the hash bit string starting with the hash bit string “01011000” of attribute 3 (residence) is concatenated.
In this way, when creating an index, the order of bit connection is cyclically changed so that any attribute is evenly assigned to each bit area. Since it does not depend on the number of combinations of product conditions, there is an effect that the indexing load is small.

[When searching]
6 and 7 are diagrams for explaining a bit concatenation indexing method at the time of search.
At the time of search, an index having one attribute of the search condition in the first bit is selected, and a range search is performed on it.
<Example 1>
FIG. 6 shows an example in which the attribute 2 (gender) is male and the attribute 3 (residence) is Tokyo for the inverted index of the server 11 in the cluster. Attribute 0 (last name) and attribute 1 (first name) are “*” (wildcard) because they are not search targets.
The conditional expression is shown below.
Attribute 2 (gender) = male ∧ Attribute 3 (residence) = Tokyo ∧ is a product symbol. It is known that if ∧ (product) can be realized, all can be realized using ∨ (sum), ¬ (negation), ⇒ (if).

  In the case of Example 1, the first 8 bits of the hash bit sequence are determined by the hash value of “m”, and the 8 bits of the next hash bit sequence are determined by the hash value of “Tokyo”. Therefore, as shown by the symbol “a” in FIG. 6, all indexes in charge of the hash area in the hash space are acquired. In this example, an index is acquired from the server 11 in the cluster having keys “9876”, “5432”, “1234”, and “5678”. Then, an index existing in the areas of all “11111111” and “11111111” from “00000000” and “00000000” is acquired. By acquiring the indexes existing in the areas “00010100”, “01011000”, “11111111”, “11111111” shown in the code c of FIG. 6 from “00010100” “01011000” “00000000” “00000000” shown in the code b of FIG. , "Men" ∧ "Tokyo" search is realized.

<Example 2>
FIG. 7 shows an example in which the attribute 0 (last name) is Yamada and the attribute 2 (gender) is male with respect to the inverted index of the server 11 in the cluster. Attribute 1 (name) and attribute 3 (residence) are “*” (wild card) because they are not search targets.
The conditional expression is shown below.
Attribute 0 (surname) = Yamada Atsushi Attribute 2 (gender) = male In this example 2, the first 8 bits of the hash bit string is the hash value of “Yamada”, and the next hash bit string is 8 bits of “male”. Determined by value. Therefore, as shown by the symbol a in FIG. 7, all indexes that are in charge of the hash area in the hash space are acquired. In this example, an index is acquired from the server 11 in the cluster having keys “9876”, “5432”, “1234”, and “5678”. Then, an index in which all 8 bits of “*” (wild card) are in the area from “11010111” “00000000” “000010100” “00000000” to “11010111” “11111111” “00010100” “11111111” is acquired. By retrieving an index that exists in the area of the hash bit string indicated by reference symbols b and c and reference symbols d and e in FIG.

Next, variations of combining Hash values will be described.
<Join example 1>
FIG. 8 is a diagram illustrating an example of combining Hash values. FIG. 8 is an example 1 of combining according to the appearance frequency of the product condition in the search.
As attribute types of attribute values, there are attribute 0 (last name), attribute 1 (first name), attribute 2 (gender), and attribute 3 (residence) shown in FIG. Further, it is assumed that the appearance frequency of these attribute types is found to be attribute 2 (gender)> attribute 0 (surname)> attribute 3 (residence)> attribute 1 (first name).
As shown in FIG. 8A, the attribute types are rearranged in the descending order of appearance frequency, and hash values calculated from the rearranged attribute values are combined as a bit string.
Further, as shown in FIGS. 8B to 8D, the bits are connected in a cyclic manner in the order of appearance frequency. Since attributes having a high appearance frequency come to the top, the search speed of the entire product condition search is improved.

<Combination example 2>
FIG. 9 is a diagram illustrating an example of combining Hash values. FIG. 9 is an example 2 of combining according to the appearance frequency of the product condition in the search.
As attribute types of attribute values, there are attribute 0 (last name), attribute 1 (first name), attribute 2 (gender), and attribute 3 (residence) shown in FIG. The frequency of appearance of these attribute types is attribute 2 (gender)> attribute 0 (surname)> attribute 3 (residence)> attribute 1 (first name) in order, and the appearance probability P is P (gender) = 0. 5, P (last name) = 0.4, P (residence) = 0.3, and P (first name) = 0.2.
As shown by reference symbol a in FIG. 9, each attribute always comes to the head once, and as shown by reference symbol b in FIG. 9, a weighted sum of probabilities is taken according to the bit positions to be combined. As shown by the symbol c in FIG. 9, the combination order that maximizes the weighted sum of the probabilities is determined. By doing in this way, there is an effect of improving the search speed when an attribute tag that frequently appears is included in the conditional expression.

<Combination example 3>
FIG. 10 is a diagram illustrating an example of combining Hash values. FIG. 10 is an example of combining according to the combination frequency of product conditions in the search.
As attribute types of attribute values, there are attribute 0 (last name), attribute 1 (first name), attribute 2 (gender), and attribute 3 (residence) shown in FIG. Further, it is assumed that the combination frequency of product conditions is known.
For example, attribute 2 (gender) ∧ attribute 3 (residence), attribute 0 (surname) ∧ attribute 3 (residence), attribute 0 (surname) ∧ attribute 2 (gender), and attribute 0 (surname) ∧ attribute 1 ( Name) Assume that the combination frequency of the product condition of ∧ attribute 3 (residence) is high.
As shown in FIG. 10A, bit concatenation is performed with the hash values of attribute 0 (last name) and attribute 2 (gender) having a high combination frequency of product conditions set higher.

Similarly, as shown in FIG. 10 (b), bit connection is performed with attribute 0 (surname) and attribute 3 (residence) having higher combination frequency of product conditions being higher. Also, as shown in FIG. 8C, bit concatenation is performed with the hash values of attribute 2 (gender) and attribute 0 (surname) having a high combination frequency of product conditions being higher. Further, as shown in FIG. 10 (d), bit connection is performed with attribute 3 (residence), attribute 0 (surname), and attribute 1 (first name) having higher frequency of combination of product conditions.
Here, when there are many searches for the combination frequency of product conditions, the cyclic replacement is not performed.
In order to use the appearance frequency (see FIG. 9) and the combination frequency (see FIG. 10), the attribute type included in the search condition is recorded as a search history when searching. To do. By doing so, there is an effect of improving the search speed in the conditional product of combinations of attribute tags that often appear.

Next, increase / decrease in the number of indexes will be described.
In the examples so far, the index for the number of attributes is created, but an index greater than the number of attributes may be created if the number is fixed. For the same reason, the number of attributes may be less than the number of attributes.
<Combination example 4>
FIG. 11 is a diagram illustrating an example of combining Hash values. FIG. 11 shows an example of combining by increasing or decreasing the number of indexes.
In the case of FIG. 5, the number of attributes is four, ie, attribute 0 (last name), attribute 1 (first name), attribute 2 (gender), and attribute 3 (residence), and FIGS. As shown in the figure, only the number of attributes were created. If the attribute value is a fixed number, more indexes than the number of attributes may be created.
For example, as shown in FIGS. 11A to 11D, as in FIGS. 5A to 5D, the number of attributes is attribute 0 (last name), attribute 1 (first name), attribute 2 (gender). ) And attribute 3 (residence). In addition, about FIG. 11 (a)-(d), the example similar to the case of FIG. 10 (a)-(d) was applied mutatis mutandis.
As shown in FIGS. 11A to 11D, the original number of attributes is four, and those starting from attribute 1 (name) do not appear. Therefore, as shown by the symbol a in FIG. 11 (e), the one starting from the attribute 1 (name) is added.

<Combination Example 5>
FIG. 12 is a diagram illustrating an example of combining Hash values. FIG. 12 shows an example in which there is no combination if there is an attribute that is not used as a product condition.
In Examples 1 to 3 so far, hash values corresponding to the number of attributes are combined. If there is an attribute that is not used as a product condition, it does not have to be combined.
In the case of FIG. 5, the number of attributes is four, ie, attribute 0 (last name), attribute 1 (first name), attribute 2 (gender), and attribute 3 (residence), and FIGS. As shown in the figure, only the number of attributes were created.
If it is known that attribute 1 (name) is an attribute that is not used as a product condition, attribute 1 (name) may not be combined.
For example, as shown in FIGS. 12A to 12C, when attribute 1 (name) is an attribute that is not used as a product condition, attribute 1 (name) is not combined. The number of attributes is three. As a result, in this example, the number of indexes is three in order to avoid duplication.

  As described above, the server 11 of the present embodiment creates a hash bit string that creates a hash bit string of attributes corresponding to the attribute types of attribute values based on the hash function for retrieving attribute values when creating an index. Means 114, combining means 115 for combining the hash bit strings of attributes corresponding to the attribute types in a predetermined concatenation order, and selecting an index having the attribute of the search condition that has been multiplied in the head hash bit string at the time of the search, Search means 116 for performing a range search on the index.

  As a result, by using a value obtained by combining the hash bit strings of attributes corresponding to the attribute types of the attribute values, it is possible to reduce the load at the time of searching and the load of indexing. That is, the indexing load is small because it does not depend on the number of combinations of product conditions. In addition, data capacity can be saved, consistency can be ensured, and flexible conditional search can be performed without having to worry about what is key when designing data.

DESCRIPTION OF SYMBOLS 1 Cluster system 2 Client 11 Server 20 Input / output means 30 Memory means 40 Storage means 110 Control means 111 Information reception means 112 Syntax analysis means 113 Attribute tag information storage means 114 Hash bit string creation means 115 Combination means 116 Search means 117 Information transmission means

Claims (6)

In a distributed database (DB) using KVS (Key Value Store), a hash function for searching attribute values is applied, and the transposed index is stored in the server corresponding to the position in the hash space of the obtained hash value A computer as each of the servers constituting the cluster system,
When creating an index,
A hash bit string creating means for creating a hash bit string of attributes corresponding to the attribute types of the attribute value, based on a hash function for retrieving the attribute value;
A program for causing a hash bit string of attributes corresponding to the attribute types to function as a coupling means for coupling in a predetermined connection order. The coupling means includes
The program according to claim 1, wherein all attributes corresponding to attribute types of the attribute value are arranged and combined at least once at the head of the connection order. The coupling means includes
The program according to claim 1, wherein the attribute types of the attribute values are rearranged in descending order of appearance frequency, and hash values calculated from the rearranged attribute values are combined as a bit string. The coupling means includes
The program according to claim 1, wherein bit concatenation is performed with a hash value of the attribute type of the attribute value having a high combination frequency of product conditions as a higher rank. The server
2. The program according to claim 1, wherein at the time of searching, an index having the attribute of the searched search condition in the head hash bit string is selected, and a range search is executed on the index. In a distributed database (DB) using KVS (Key Value Store), a hash function for searching attribute values is applied, and the transposed index is stored in the server corresponding to the position in the hash space of the obtained hash value A cluster system that
The server
When creating an index,
A hash bit string creating means for creating a hash bit string of attributes corresponding to the attribute types of the attribute value, based on a hash function for retrieving the attribute value;
A cluster system comprising: coupling means for coupling hash bit strings of attributes corresponding to the attribute types in a predetermined connection order.

Images