JP2016015011A - Database device, database management method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To configure a database of small-capacity memories dramatically improved in efficiency of use.SOLUTION: A database device can generate from key values, by using N hash functions f[n], N entry numbers each matching an N index table 2[n] (n=0 ... N-1), it can avoid in an advance collision in which N entry numbers take on the same value, store a key Keach time it is entered from outside while selecting a bucket for storing the key and allocating each key to one or other bucket, and accordingly the efficiency of use of memories by the index table can be dramatically improved over a current level.

Description

  The present invention relates to a database apparatus, a database management method, and a program, and more particularly, to a database apparatus, a database management method, and a program for acquiring a record corresponding to a key from a large-scale database having a large data capacity with high throughput.

  Conventionally, a LAN switch such as Ethernet (registered trademark), which is a node that constructs a communication network, is provided with a database device having a transfer database for determining a transfer destination of a packet received from a line.

  In the LAN switch, with the destination address (for example, MAC address) included in the packet to be transmitted as a key, the transfer destination of the packet (a line ID that specifies a line of a communication network, a transmission port ID that specifies a transmission port, etc.) and the packet The information on whether or not to discard is acquired from the transfer database of the database device. In the LAN switch, the destination address value of the packet and the transfer destination of the packet having the destination address value are registered in the transfer database.

  For example, when transferring a packet between lines of 100 [Gbps], the LAN switch determines the transfer destination (line ID, transmission port ID, etc.) of the packet within the time when a packet with a minimum packet length of 64 bytes is received. Although it needs to be done, the time to decision is only about 6 nanoseconds.

  Therefore, the number of clock cycles allowed for the LAN switch to determine the packet transfer destination is only 3 cycles since one cycle is 2 nanoseconds when the clock frequency is 500 [MHz].

  As a technique for acquiring transfer destination information (hereinafter referred to as “transfer destination information”) for a packet destination address from the transfer database in such a short time, there is a technique using a hash table. In this technique, a packet destination address is converted into a hash table entry number using a hash function.

  Each entry of the hash table stores a destination address value associated with the entry number of the entry and transfer destination information of a packet corresponding to the destination address value.

  When the LAN switch obtains the packet transfer destination information, it converts the packet destination address into the hash table entry number using a hash function, and the destination address value stored in the entry of the hash table indicated by the entry number And the destination address of the packet match, the transfer destination information stored in the entry is acquired. The LAN switch determines a packet transfer destination based on the transfer destination information.

  If the destination address value stored in the entry does not match the destination address of the packet, the LAN switch recognizes that the forwarding destination information of the packet is not registered in the hash table, and all possible packets Broadcast to the forwarding destination.

  In the method using the hash table as described above, when each of two or more different destination addresses is converted by one hash function, the entry number becomes the same value, that is, the hash value (entry number) collides. When (collision) occurs, only one of these destination addresses can be registered in the transfer database.

  When transferring a packet having a destination address that could not be registered, since the transfer destination of the packet is unknown, the LAN switch copies the packet and broadcasts it to all transfer destinations. By copying packets and broadcasting them to all transfer destinations in this way, transfer performance deterioration and line congestion that are unnecessary when the destination address can be registered are caused.

  As a method for avoiding the above-described transfer performance deterioration and line congestion, a maximum of L (L) entries can be obtained for each hash value (each entry number) obtained by converting each of two or more different destination addresses using one hash function. There is a technique in which a bucket capable of storing a destination address value of 1 or more is used as an entry in an index hash table (see, for example, Non-Patent Document 1).

  In this method, when registering the destination address value and its transfer destination information in the transfer database, the destination address value is stored in the empty position of the bucket corresponding to the entry number converted by the hash function, and the destination address value The transfer destination information corresponding to the destination address value is stored at a position in the transfer database determined in accordance with the storage position.

  Also, when determining the packet transfer destination, a bucket in the index hash table is specified by the entry number converted from the destination address of the packet using a hash function, and a plurality of destination addresses stored in the bucket Among the values, a destination address value that matches the destination address of the packet is searched.

  In this search, when a destination address value matching the destination address of the packet is found from the designated bucket, the LAN switch stores the transfer destination information stored in the transfer database according to the storage location of the destination address value. , And the transfer destination of the packet is determined using the transfer destination information.

  As described above, in the above-described method, up to L destination address values having the same hash value (entry number) obtained by converting each of two or more different destination addresses using a hash function may be registered in the transfer database. Therefore, by increasing the bucket size L of the bucket in the index hash table, the probability that the destination address value cannot be registered due to the collision of hash values can be reduced.

Shuichi Okubo, “Design and Implementation of Network Virtualization Using“ Sakura Cloud ”as an Example”, Sakura Internet Laboratory, p.25-26http: //research.sakura.ad.jp/wp-content/uploads/2011 /10/20110927-hbstudy.pdf

  However, in a database device that can register a destination address consisting of a fixed number of random values, the number of destination addresses that have the same hash value (entry number) varies. The bucket size L of the hash table that can be registered must be the maximum value of the variation in the number of destination addresses.

  However, since there is a low probability that the variation in the number of destination addresses having the same hash value (entry number) will be the maximum value, when a predetermined number of destination addresses have been registered in the transfer database, In the majority of buckets, it is considered that the number of destination addresses in which destination address values are stored in the bucket has not reached the bucket size L.

  That is, many unused areas remain in the memory provided for the index hash table. For the same reason, a large number of unused areas remain in the memory provided for the transfer database.

  Therefore, in a database device in which the hash table for index and the transfer database are configured by memory, the memory use efficiency becomes low, and it is necessary to provide a memory having a capacity larger than the data size of the destination address to be registered and the transfer destination information. There was a problem that memory wasted.

  The present invention has been made to solve the above problems, and a database device, a database management method, and a database device configured by a small-capacity memory in which the use efficiency of the memory constituting the database is remarkably improved. The purpose is to propose a program.

  The database device according to the present invention is provided with a plurality of key storage positions capable of storing a plurality of key values, and has a plurality of N (N is an integer of 2 or more) buckets each associated with a different entry number. An index table and N hash functions different from each other corresponding to each of the N index tables, and the N indexes based on a key value of a key input from the outside using the N hash functions An entry number generation unit that generates N entry numbers corresponding to each table; a record table that stores a record value corresponding to the key value in a record storage position corresponding to the key storage position in the bucket; A bucket for storing the key value from N buckets in the N index tables corresponding to N entry numbers. An index table registration unit for storing the key value in the bucket, and storing the record value corresponding to the key value in the record storage position of the record table corresponding to the key storage position storing the key value And a record table registration unit.

  2. The index table search unit according to claim 1, wherein the index table search unit searches for the bucket in which a key value matching the key is stored from N buckets in the N index tables corresponding to the N entry numbers. And a record table search unit for acquiring the record value from the record storage position of the record table corresponding to the key storage position in which the key value is stored in the bucket searched by the index table search unit. It is characterized by.

  2. The index table registration unit according to claim 1, wherein the index table registration unit includes a bucket length that is the number of key values already stored in the N buckets, or a number of key values that can be additionally stored in the N buckets. In the N index tables, a bucket that stores the key value is selected from N buckets corresponding to the N entry numbers in the N index tables.

  2. The N index tables can be simultaneously accessed from both the index table registration unit and the index table search unit, and the record table includes the record table registration unit and the record table search unit. It can be accessed from both sides simultaneously.

  2. The record table according to claim 1, wherein the record table is provided with a plurality of record storage positions capable of storing a plurality of record values and has a plurality of buckets respectively associated with different record entry numbers. It is characterized by.

  The index table registration unit according to claim 1, wherein the index table registration unit refers to a key storage map table that indicates whether or not the key value is stored in all the key storage positions in the N index tables, and In the index table, the bucket serving as the storage destination of the key value is selected from N buckets corresponding to the N entry numbers.

  According to the present invention, since N entry numbers corresponding to the N index tables can be generated from the key values using N hash functions, the probability that the N entry numbers collide is obtained. Since it is possible to select and store a bucket in which a key is to be stored from N index tables each time a key is input from the outside, it is possible to reduce the memory usage efficiency of the index table compared to the conventional case. In addition to the significant improvement, the memory can be configured with a small capacity memory.

It is a block diagram which shows the structure of the database apparatus in 1st Embodiment. It is a figure which shows the structure of the index table in 1st Embodiment. It is a figure which shows the structure of the record table in 1st Embodiment. It is a flowchart which shows the registration processing procedure by the index table control part in 1st Embodiment. It is a flowchart which shows the search processing procedure by the index table control part in 1st Embodiment. It is a graph with which it uses for description of the memory usage efficiency in 1st Embodiment. It is a block diagram which shows the structure of the database apparatus in 2nd Embodiment. It is a block diagram which shows the structure of the database apparatus in 3rd Embodiment. It is a figure which shows the structure of the record table in 3rd Embodiment. It is a figure which shows the structure of the key storage map table in 3rd Embodiment. It is a flowchart which shows the registration processing procedure by the index table registration part in 3rd Embodiment. It is a flowchart which shows the search processing procedure by the index table search part in 3rd Embodiment.

<First Embodiment>
<Overall configuration of database device>
Hereinafter, the configuration of the database apparatus according to the first embodiment of the present invention will be described with reference to FIG.

  The database device 1 is realized by installing a computer program (software) in a computer (hardware) including a CPU (Central Processing Unit), a memory, an interface, and the like. This is realized by the cooperation of the hardware resource and the computer program. Further, the computer program may be provided in a state of being stored in a computer-readable recording medium or a storage device, or may be provided via an electric communication line.

  As shown in FIG. 1, the database apparatus 1 according to the first embodiment has N index tables 2 [N is an integer of 2 or more] that store key values (destination address values) corresponding to a plurality of keys. n] (n = 0... N-1), a record table 3 for storing a plurality of record values (transfer destination information) respectively corresponding to a plurality of key values (destination address values) stored in the index table 2 [n]. , Registration and retrieval of key values (destination address values) for the index table 2 [n], registration of record values corresponding to the key values in the record table 3 and retrieval from the record table 3 in conjunction with the record table control unit 5 Index table control unit 4 to be controlled, record table for controlling record value registration to record table 3 or search of record value from record table 3 in conjunction with index table control unit 4 It is constituted by a control unit 5. In the database apparatus 1, a transfer database is configured by a plurality of index tables 2 [n] (n = 0... N−1) and the record table 3.

<Configuration of index table>
FIG. 2 shows a configuration example of the index table 2 [n] (n = 0... N−1). Since the index tables 2 [0],... 2 [n],... 2 [N-1] all have the same configuration, the index table 2 [n] will be described here for convenience.

  The index table 2 [n] (n = 0... N-1) is a memory, and each index table 2 [n] has M entries 11 [i] (i = 0... M-1). . Here, [n] (n = 0... N-1) is an index table number (an integer greater than or equal to 2) that specifies the index table 2 [n], and [i] (i = 0... M-1) is This is an entry number (an integer greater than or equal to 1) that identifies a plurality of entries 11 [i] in the index table 2 [n].

  This entry 11 [i] (i = 0... M-1) has L key storage positions 12 [i] [j] (i = 0... M-) that can store a plurality of key values (destination address values). 1) It is configured as a bucket 10 [i] (i = 0... M-1) provided with (j = 0... L-1). In this case, it can be said that the entry 11 [i] is the bucket 10 [i], and [i] is the bucket number (an integer of 1 or more) at the same time as the entry number.

  Here, the key storage position 12 [i] [j] (i = 0 ... M-1) (j = 0 ... L-1) of the index table 2 [n] (n = 0 ... N-1) is an entry. It is specified by the number [i] (i = 0... M-1) and the in-bucket key storage position number [j] (j = 0... L-1). The key storage position number [j] (j = 0 ... L-1) in the bucket is a number (an integer greater than or equal to 1) that identifies the key storage position in the bucket 10 [i] (i = 0 ... M-1). is there.

  When reading the key value (destination address value) stored in the index table 2 [n] (n = 0... N-1) having such a configuration, the index table 2 [n] (n = 0... N- By specifying an index table number [n] specifying 1), and further specifying an entry number [i] (bucket number [i]) specifying an entry 11 [i] to be read out, the entry 11 [i] Read access is made to a plurality of key values (destination addresses) stored in (bucket 10 [i]) (details will be described later).

  Further, in the case where a key value (destination address value) corresponding to a new key is stored in the index table 2 [n] (n = 0... N-1), the index table 2 [n] serving as the storage destination is stored. ] Specify the index table number [n] (n = 0 ... N-1) for specifying (n = 0 ... N-1). Then, the entry number [i] (bucket number [i]) specifying the entry 11 [i] (i = 0... M-1) and the key storage position number [j] (j = 0... L-1) in the bucket. Is specified, the key storage position 12 [i] [j] (i = 0 ... M-1) (j = 0 ... L-1) is specified, and the key storage position 12 [i] [j] ( i = 0... M-1) (j = 0... L-1) is written to the key value (destination address value) to be stored (details will be described later).

<Structure of record table>
FIG. 3 shows a configuration example of the record table 3. The record table 3 is a memory, and the key storage positions 12 [i] [j] (i = 0 ... M-1) (j = 0 ...) of all the index tables 2 [n] (n = 0 ... N-1). A record entry 20 [k] (k = 0... K-1) is provided that can store record values (packet transfer destination information) corresponding to each of the key values (destination address values) stored in (L-1). ing.

  Here, [k] (k = 0... K-1) is a record entry number (an integer greater than or equal to 1) that identifies the record entry 20 [k] that is a record storage position in the record table 3, and is an index table. 2 [n] (n = 0 ... N-1) key storage position 12 [i] [j] (i = 0 ... M-1) Corresponds to the key value stored in (j = 0 ... L-1) This is for specifying the storage position of the record value.

  In the first embodiment, key storage positions 12 [i] [j] (i = 0... M-1) (j provided in all index tables 2 [n] (n = 0... N-1) = 0... L−1) is the total number of index tables N × each index table 2 [n] (n = 0... N-1) table size M (number of entries in index table 2 [n] (number of buckets) )) × Bucket size L of each bucket (number of key storage positions 12 [i] [j]).

  Therefore, the record table 3 in the case where it is configured as one memory that stores one record value (packet transfer destination information) in each record entry 20 [k] (k = 0... K-1) of the record table 3. Record table size K is K = N × M × L, and key storage positions 12 [i] [j] (i =) provided in all index tables 2 [n] (n = 0... N−1). 0 ... M-1) (j = 0 ... L-1) is the same as the total number.

  The key storage position 12 [i] [j] (i = 0 ... M-1) (j = 0 ... L-1) of the index table 2 [n] (n = 0 ... N-1) and the record table 3 The record entry 20 [k] (k = 0... K-1) is associated with the record entry number [k] in correspondence with the key storage position 12 [i] [j] (i = 0 ... M-1) (j = 0... L-1), any calculation method that can uniquely calculate the record entry number [k] of the record entry 20 [k] (k = 0... K-1) may be used.

For example, as one calculation method, index table number [n] indicating index table 2 [n] (n = 0... N-1), entry number [i] indicating entry 11 [i] (i = 0... M -1), key storage position 12 [i] [j] (i = 0 ... M-1) In-bucket key storage position number [j] (j = 0 ... L- 1), the record entry number [k] of the record entry 20 [k] (k = 0... K−1) can be calculated by the following (Equation 1).
k = n × M × L + i × L + j ………………………………………………………… (Formula 1)

  When the database device 1 reads the record value (transfer destination information) stored in the record entry 20 [k] (k = 0... K-1) of the record table 3 having such a configuration, the database device 1 reads the destination. Read access is performed by designating the record entry number [k] corresponding to the record entry 20 [k] (k = 0... K-1) to be (details will be described later).

  On the other hand, when a new record value (transfer destination information) is stored in the record table 3, the database device 1 designates a record entry number [k] (k = 0... K-1) as a storage destination. Write access is performed (details will be described later).

<Configuration of index table control unit>
In FIG. 1, an index table control unit 4 serving as an index table registration unit and an index table search unit uses a key value (destination address value) of a registration key K _r (packet destination address) input from the outside as an index table. 2 [n] (n = 0 ... N-1) key storage position 12 [i] [j] (i = 0 ... M-1) (j = 0 ... L-1) is performed in conjunction with the record table controller 5 controls such as storing a record value of the record R _r of registered corresponding to (transfer destination information) in the record table 3.

In addition, the index table control unit 4 and the record table control unit 5 also perform control such as acquiring the record value of the record R _s to be searched corresponding to the search key K _s input from the outside from the record table 3. It is done in conjunction.

The index table control unit 4 assigns N different hash functions f [n] (n = 0... N-1) respectively corresponding to the N index tables 2 [n] (n = 0... N-1). And an entry number [n] (n = 0) consisting of hash values respectively corresponding to the N hash functions f [n] (n = 0... N−1) from the registration key K _r (destination address). ... Have an entry number generation unit 4a for generating (N-1).

Here, the entry number [n] (n = 0... N-1) is converted from the key K _r in correspondence with the hash function f [n] (n = 0... N-1). This corresponds to a plurality of entries 11 [i] (i = 0... M-1) provided in the index table 2 [n] (n = 0... N-1).

When the index table control unit 4 stores the key value (destination address value) of the registration key _{Kr in} the index table 2 [n] (n = 0... N-1), the key table in which the key value is not stored is stored. The position 12 [i] [j] (i = 0... M-1) (j = 0... L-1) needs to be searched and stored therein.

Therefore, the index table control unit 4 converts the key K _r in each index table 2 [n] (n = 0... N-1) by the hash function f [n] (n = 0... N-1). For bucket 10 [n] (n = 0 ... N-1) of entry number [n] (n = 0 ... N-1), it is necessary to acquire a bucket length (described later) or an empty bucket length (described later). is there.

  Here, the bucket length is a plurality of key storage positions 12 [i] [j] in the bucket 10 [n] (n = 0 ... N-1) of the entry number [n] (n = 0 ... N-1). This is the number of key values already stored in (i = 0 ... M-1) (j = 0 ... L-1). The empty bucket length is a plurality of key storage positions 12 [i] [j] in the bucket 10 [n] (n = 0 ... N-1) of the entry number [n] (n = 0 ... N-1). (i = 0 ... M-1) This is the number of key values that can be additionally stored for (j = 0 ... L-1).

  In the index table control unit 4, as a method of acquiring the bucket length or the empty bucket length in the bucket 10 [n] (n = 0 ... N-1) of the entry number [n] (n = 0 ... N-1), There are the following three types.

In the first bucket length acquisition method, all the key storage positions 12 [i] [not stored in the key value (destination address value) in the index table 2 [n] (n = 0... N-1). j] (i = 0... M-1) An invalid key value (an arbitrary value that does not match the stored key value) is stored for (j = 0... L-1). As a result, when the index table control unit 4 acquires the bucket length or the empty bucket length, the bucket 10 [n] (n = 0... N) designated by the entry number [n] (n = 0... N-1). acquired all key values (all valid key values and invalid key values) in -1), among the entire key value, counts the number of key K _r with an invalid key value, the target it Can be acquired as an empty bucket length for the bucket 10 [n] (n = 0... N−1). However, the index table control unit 4 may count the number of keys K _r having a valid key value. In this case, the bucket 10 [n] (n = 0. Can be obtained as the bucket length for.

  In the second bucket length acquisition method, all key storage positions 12 [i] [j] (i = 0 ... M-1) (j = 0 ... L-1) is provided with a flag indicating whether or not it is stored, and the flag is cleared in the initial state (the state where the registered key value does not yet exist). 12 [i] [j] (i = 0... M-1) The flag provided at (j = 0... L-1) is set. As a result, when the index table control unit 4 acquires the bucket length or the empty bucket length, the bucket 10 [n] (n = Number of flags set (or not set) in the key storage position 12 [i] [j] (i = 0 ... M-1) (j = 0 ... L-1) of 0 ... N-1) Can be obtained as the bucket length (or empty bucket length) for the target bucket 10 [n] (n = 0... N-1).

  In the third bucket length acquisition method, a bucket length (or an empty bucket) is assigned to each entry 11 [n] (n = 0 ... N-1) of the index table 2 [n] (n = 0 ... N-1). Counter 10 representing the length), the counter is initialized in an initial state (a state in which a registered key value does not yet exist), and the bucket 10 [n] ( The counter value corresponding to n = 0... N-1) is updated. Thereby, when acquiring the bucket length or the empty bucket length, the index table control unit 4 acquires the counter value of each entry 11 [n] (n = 0... N−1) and targets the counter value. Can be acquired as a bucket length (or an empty bucket length) for a bucket 10 [n] (n = 0... N−1).

<Configuration of record table control unit>
In FIG. 1, a record table control unit 5 as a record table registration unit and a record table search unit interlocks with the index table control unit 4 and stores the key of the index table 2 [n] (n = 0... N−1). position 12 [i] [j] ( i = 0 ... M-1) registered in the record corresponding to the key value of the key K _r for registration stored in the (j = 0 ... L-1 ) ( destination address value) Control for storing the record value (transfer destination information) of R _{r in} the record table 3 is performed.

The record table control unit 5 acquires the record R _s to be searched corresponding to the search key K _s input to the index table control unit 4 from the record table 3 in conjunction with the index table control unit 4. Control is performed.

Here, it is assumed that the registration key K _r and the search key K _s have different key values, but the same key value key K _r or K _s is used for both registration and search. May be. Similarly, it is assumed that the record R _r to be registered and the record R _s to be searched have different record values, but the same record value is used for the records R _r and R _s to be registered and searched. May be.

<Registration process for registration key and record corresponding to the key>
Next, registration process will be described with respect to the forwarding database of records R _r registration symmetry corresponding to the key K _r and the key K _r for registration in the database device 1. The operation of registering the key _{Kr in} the transfer database by the index table control unit 4 will be described with reference to the flowchart of FIG.

In the database apparatus 1 (FIG. 1), when registering a registration key K _r (destination address) and a record R _r (transfer destination information) corresponding to the key K _r in the transfer database, an index table control unit 4 receives a registration key _Kr and a control signal S1 indicating a registration command from the outside (step SP1).

The index table control unit 4 uses the N hash functions f [n] (n = 0... N-1) corresponding to the index tables 2 [n] (n = 0. N entry numbers [n] (n = 0... N-1) are generated simultaneously from the registration key K _r (destination address) input from (step SP2 [0],..., Step SP2 [n]. , ..., step SP2 [N-1]).

  When the index table control unit 4 has already stored a key value for each bucket 10 [n] (n = 0 ... N-1) of each entry number [n] (n = 0 ... N-1). Each bucket length [n] (n = 0... N-1) representing the number of nodes is acquired (step SP3 [0],..., Step SP3 [n],..., Step SP3 [N-1]). Instead of acquiring the bucket length [n], an empty bucket length [n] that is the number of key values that can be additionally stored may be acquired.

  The index table control unit 4 performs N bucket lengths [n] (n = 0... N) obtained in steps SP3 [0],..., Steps SP3 [n],. -1) is searched for the minimum bucket length [nr], and the index table 2 [nr] corresponding to the minimum bucket length [nr] is selected (step SP4).

  The index table control unit 4 determines that the key value of the key storage position [nr] [j] (j = 0... L-1) of the index table 2 [nr] corresponding to the minimum bucket length [nr] The key storage position 12 [nr] [jr] for storage is searched (step SP5). Here, [nr] at the key storage position 12 [nr] [jr] is the entry number [nr] corresponding to the minimum bucket length [nr] that identifies the key storage position 12, and [jr] is the key This is the key storage position number [jr] in the bucket that identifies the key storage position whose value is not stored.

The index table control unit 4 determines whether or not the key storage position 12 [nr] [jr] in which the key value is not stored is found in the bucket 10 [nr] of the entry 11 [nr] in the index table 2 [nr]. (Step SP6) and if found, the key storage position 12 [nr] [jr] where the key value is not stored in the bucket 10 [nr] of the entry 11 [nr] in the index table 2 [nr] It stores the key value of the key K _r for registration (step SP7).

Since the index table control unit 4 stores the key value of the registration key K _r in the key storage position 12 [nr] [jr] of the index table 2 [nr], the index table 2 [nr] indicates the index table 2 [nr]. The in-bucket key storage position number [jr] indicating the table number [nr], the entry number [nr], and the key storage position 12 [nr] [jr] is substituted into the above-described (Expression 1).

As a result, the index table control unit 4 calculates the record entry number [k] (k = 0... K−1) of the record entry 20 [k] by the following (Equation 2), and sends this to the record table control unit 5. (Step SP8), and the registration process of the index table control unit 4 is terminated.
k = index table number [nr] × M × L + entry number [nr] × L + jr (Equation 2)

As with the index table control unit 4, the record table control unit 5 (FIG. 1) receives a control signal S 1 indicating a registration command from the outside, and the record entry of the record table 3 supplied from the index table control unit 4 Enter the number [k] (k = 0... K-1) and the record _Rr to be registered corresponding to the registration key _Kr .

The record table control unit 5 records the record R _{r in} the record entry 20 [k] of the record table 3 corresponding to the record entry number [k] (k = 0... K−1) according to the control signal S1 meaning a registration command. Stores the record value of.

By the way, the index table control unit 4 cannot find the key storage position 12 [nr] [jr] in which the key value is not stored in the bucket 10 [nr] of the entry 11 [nr] in the index table 2 [nr]. If, that is, the already key values into buckets [nr] entry 11 [nr] are all stored can not be registered key value of the key K _r for registration in the index table 2 [nr] Therefore, processing when registration is impossible is performed (step SP7).

Index table control unit 4, as the processing at the time of registration impossible, for example, the number of times that can not register a key value of the key K _r for registration in the index table 2 [nr] counted by the counter, and updates whenever the counting, the The count value is retained.

Alternatively, the index table control unit 4 registers one key value among the already stored key values for the bucket [nr] of the entry 11 [nr] in the index table 2 [nr] as a process when registration is impossible. Delete. The index table control unit 4 then uses the key K _r for registration for the deleted key storage position 12 [i] [r] (i = 0... M-1) (j = 0... L-1). In addition to registering a value, the record entry number [k] (k) in the record table 3 corresponding to the key storage position 12 [i] [r] (i = 0... M-1) (j = 0... L-1) = 0... K-1) is calculated and supplied to the record table control unit 5.

  The index table control unit 4 selects the index table 2 [nr] corresponding to the entry number [nr] of the bucket length [nr] where the bucket length [n] (n = 0... N-1) is the minimum value. However, the present invention is not limited to this, and there are the following three other index table selection methods for selecting the index table 2 [nr] to be registered.

  In the first index table selection method, the index table number in the bucket 10 [n] (n = 0... N-1) where the empty bucket length [n] (n = 0... N-1) is larger than 0. [n] (n = 0... N-1) is selected as the index table 2 [nr] having the index table number [nr] having the minimum value.

  In the second index table selection method, an index having an index table number [nr] that is the empty bucket length [nr] that takes the maximum value among the empty bucket lengths [n] (n = 0 ... N-1). Select table 2 [nr].

  In the third index table selection method, each index table 2 [n] (n = 0... N-1) has a different table size M (number of entries 11 [i] (i = 0... M-1)). When there is a weighted free bucket length [n] (n = 0 ... N-1) weighted by a weighting factor set for each index table 2 [n] (n = 0 ... N-1) n] (n = 0... N-1), the index table 2 [nr] having the index table number [nr] having the maximum weighted free bucket length [nr] is selected.

  In the first to third index table selection methods, the bucket size L of each index table 2 [n] (n = 0... N−1) (key storage position 12 [i] [r] (i = 0 ... M-1) (number of j = 0 ... L-1)) is also different.

<Record search processing procedure corresponding to the search key>
Next, a description will be given of a search process for searching the transfer database for the record value of the record R _s to be searched corresponding to the key value of the search key K _s in the database device 1. A search operation for searching the transfer database for the record R _s corresponding to the search key K _s by the index table control unit 4 will be described with reference to the flowchart of FIG.

In the database apparatus 1, when the record value (transfer destination information) of the record R _s corresponding to the key value (destination address) corresponding to the search key K _s is acquired from the transfer database, the index table control unit 4 A key K _s (destination address) and a control signal S2 indicating a search command are input (step SP11).

The index table control unit 4 uses the N hash functions f [n] (n = 0... N-1) corresponding to all the index tables 2 [n] (n = 0. N entry numbers [n] (n = 0... N−1) are simultaneously generated from the search key K _s (destination address) input from (step SP12 [0],..., Step SP12 [n]). , ..., step SP12 [N-1]).

  The index table control unit 4 stores each bucket 10 [n] (n = 0...) Of each entry number [n] (n = 0... N−1) in the index table 2 [n] (n = 0... N−1). N maximum key values (destination address values) already stored in N-1) are acquired (step SP13 [0],..., Step SP13 [n],..., Step SP13 [N-1]). ).

The index table control unit 4 includes, among the maximum L key values (destination address values) acquired in step SP13 [0],..., Step SP13 [n],. An index table number [ns] storing a key value (destination address value) matching the search key K _s (destination address) and its key storage position [ns] [js] are searched (step S14). ).

Here, [ns] at the key storage position 12 [ns] [js] is the index table number [ns] in which the key value (destination address value) that matches the search key K _s (destination address) is stored. [Js] is a key in which a key value (destination address value) matching the search key K _s (destination address) is stored in the index table 2 [ns] of the index table number [ns]. This is the key storage position number [js] in the bucket that identifies the storage position.

The index table control unit 4 determines whether or not a key value (destination address value) matching the search key K _s (destination address) has been found in the index table 2 [ns] of the index table number [ns]. If it is determined (step SP15) and found, the index table number [ns] indicating the index table 2 [ns], the entry number [ns], and the key in the bucket indicating the key storage position 12 [ns] [js] The storage position number [js] is assigned to the above (Equation 1).

As a result, the index table control unit 4 calculates the record entry number [k] (k = 0... K−1) of the record entry 20 [k] by the following (Equation 3), and sends this to the record table control unit 5. (Step SP16), and the search process of the index table control unit 4 is terminated.
k = index table number [ns] x M x L + entry number [ns] x L + js (Equation 3)

  As with the index table control unit 4, the record table control unit 5 (FIG. 1) has already received a control signal S2 indicating a search command from the outside, and the record table supplied from the index table control unit 4 in that state. 3 record entry number [k] is entered.

The record table control unit 5 records the record entry 20 [k] (k = 0) in the record table 3 corresponding to the record entry number [k] (k = 0... K-1) according to the control signal S2 meaning a search command. ... the record value of the record R _s to be searched is acquired from K-1) and output to the outside.

By the way, the index table control unit 4 could not find the key value (destination address value) matching the search key K _s (destination address) from the index table 2 [ns] of the index table number [ns]. In this case, since the key value of the search key K _s is not registered in the index table 2 [n], a process when the search is impossible is performed such as outputting a record value indicating that the key was not found (step SP17). .

The index table control unit 4 performs, for example, an unregistered search key K _s (hereinafter referred to as “unregistered search key K _s ”) as another process when search is impossible in step SP17. The number of searches can be counted by a counter, updated every time it is counted, and the count value can be held.

Alternatively, the index table control unit 4 obtains a default record value for the unregistered search key K _s from the record table 3 as another process when the search is not possible in step SP17. ] Can generate a special record entry number [k] (k = 0... K-1), which means that there is no registered record value, and supply it to the record table control unit 5.

<Effect in the first embodiment>
The database apparatus 1 in the first embodiment and a conventional database apparatus using existing technology are compared and examined from the viewpoint of memory usage efficiency.

  For example, the number N of index tables 2 [n] in the database apparatus 1 in the first embodiment is “2”, and the table size M (number of entries, ie buckets) of each index table 2 [n] (n = 1, 2) The number is fixed to “512”. The number of bucket sizes L (key storage positions [i] [j] (i = 0... M-1) (j = 0... L-1) of each index table 2 [n] (n = 1, 2). ) Are all the same value, and the total bucket size is “L × 2”.

Using the total bucket size “L × 2” as a parameter, the bucket size L necessary for registering all the registration keys K _r made up of a fixed number of random values in the transfer database was obtained by simulation.

Specifically, the case of registering the key K _r for registration consisting 16000 random value to the forwarding database attempts multiple times in each trial, determine whether the key K _r which can not be registered is generated did. Further, the number of trials in which the key _Kr that could not be registered is counted by a counter (processing when registration is impossible in the above-described step SP9), and a value obtained by dividing the count value by the total number of trials is obtained as a registration failure occurrence rate. did.

  In the first embodiment, since the number N of index tables 2 [n] is “2”, each entry number [n in two index tables 2 [n] (n = 1, 2) ] To generate two hash functions f [n] (n = 1, 2) are required. As these two hash functions f [n], a function that can generate entry numbers [n] having no correlation with each other from one key value is adopted.

  For example, a bit string obtained by performing a CRC (Cyclic Redundancy Check) operation on the key value is divided into two and used for the entry number [n], or the key value is encrypted using two different IVs (Initial Vectors) There is a method of using the two applied bit strings for the entry number [n].

The database device using the existing technology to be compared has one index table 2 [n], and the table size M of the index table 2 [n] is the same as that of the database device 1 in the first embodiment. Fix to “512”. Also, a registration consisting of a fixed number of random values using as a parameter the total bucket size of one index table 2 [n] (in this case, the total bucket size = bucket size since there is only one index table 2 [n]) It was determined by simulation of the bucket size L required to register in all transfer database of the key K _r of use.

  By comparing the database apparatus 1 according to the first embodiment and the conventional database apparatus using the existing technology under the same bucket size, the memory used by the two database apparatuses becomes the same amount. That is, when there are two index tables 2 [n] (n = 1, 2) in the database apparatus 1 in the first embodiment and each bucket size is “L”, the index in the conventional database apparatus to be compared is to be compared. One table 2 [n] is the bucket size “L × 2”.

  Here, since the table size M of the database device 1 in the first embodiment and the conventional database device is the same value, when the total bucket size is the same value, it is used for the index table 2 [n] and the record table 3. Both memory amounts are M × L × 2 × “the amount of memory required to store one key value in the record table 3”, which is the same value.

Therefore, when comparing between the database device 1 in the first embodiment and the conventional database device, the smaller total bucket size when the above-described registration failure occurrence rate is “0” is the same number. This means that the amount of memory necessary for registering the key value of the key _Kr is reduced, and the memory usage efficiency is high.

  FIG. 6 shows a simulation result that the smaller the total bucket size when the registration failure occurrence rate is “0”, the higher the memory use efficiency. In the database apparatus 1 in the first embodiment, the registration failure occurrence rate becomes 0 when the total bucket size “2 × L” is equal to or larger than 36, whereas in the conventional database apparatus, the total bucket size “L”. The registration failure rate became 0 when the value was 58 or more.

  That is, in order to register 16000 keys Kr, the amount of memory necessary for the database apparatus 1 in the first embodiment is 512 (table size M) × 36 (bucket size) × (key length of key value + In the conventional database apparatus, 512 (table size M) × 58 (bucket size) × (key value key length + record value record length).

Therefore, in the database apparatus 1 in the first embodiment, the amount of memory required for the transfer database can be reduced by 38% (1- (36/58)). In the database apparatus 1 according to the first embodiment, since 16000 keys Kr are registered in a memory capable of registering a maximum of 512 × 36 keys K _r , the memory usage efficiency is 87% (16000/1). On the other hand, in the conventional database apparatus, since 16000 keys Kr are registered in a memory capable of registering a maximum of 512 × 58 keys Kr, the memory usage efficiency is 54% ( 16000 / (512 × 58)).

By the way, in the conventional database apparatus, there is one index table 2 [n], and a registration key K _r is entered in the index table 2 [n] by one type of hash function f [n] (n = 0). Just convert it to the number [n]. For this reason, in the conventional database apparatus, even when different keys K _r are used, the same hash value entry number is converted as compared with the case where a plurality of hash functions f [n]) (n = 0... N−1) are used. The probability of being done is high. Therefore, the conventional database device requires a memory capacity that can store up to L (for example, at least 58) key values in a bucket having the same entry number.

On the other hand, in the database device 1 in the first embodiment, the index table control unit 4 has N hash functions respectively corresponding to the N index tables 2 [n] (n = 0... N−1). f [n] (n = 0... N-1), and the registration key _Kr is indexed by indexing table 2 [n] (N hash functions f [n] (n = 0 ... N-1). Since entry numbers [n] that are not correlated with each other can be generated every n = 0... N−1), the probability that these entry numbers [n] have the same hash value is very low.

  Therefore, in the database apparatus 1, the number of key values stored for the bucket 10 [n] having the same entry number [n] is reduced. The database device 1 stores the key values in order from the bucket [nr] having the smallest bucket length [nr], thereby obtaining the number of key values stored between the index tables 2 [n]. Since the variation is flattened with little variation, the memory capacity of the key value stored in the bucket 10 [n] of the same entry number [n] can be small.

As described above, the database apparatus 1 uses the N hash functions f [n] (n = 0... N-1) corresponding to the N index tables 2 [n] (n = 0... N-1), respectively. The registration key K _r is simultaneously converted into N entry numbers [n] (n = 0... N−1), and each bucket 10 [ registering a key value corresponding to the key K _r with respect to n]. Thereby, the memory usage efficiency of the index table 2 [n] (n = 0... N-1) and the record table 3 can be greatly improved, and a certain number of registration keys _Kr are stored. The index table 2 [n] (n = 0... N-1) required for the record table 3 and the memory amount required for the record table 3 can be greatly reduced.

<Another embodiment corresponding to the first embodiment>
Note that the record table 3 is not limited to the above-described configuration, and similarly to the N index tables 2 [n] as shown in FIG. 1, the record table 3 [n] (n = 0 ... N-1). In this case, the table size k of each record table 3 [n] (n = 0... N-1) is the table size M of the index table 2 [n] (number of entries in the index table (number of buckets)) × of each bucket. Bucket size L (the number of key storage positions [i] [j] (i = 0... M-1) (j = 0... L-1)).

  The record value corresponding to the key value stored in the index table 2 [n] is the index table number [n] (n = 0... N-1) and the entry number [i] (i = 0... M-1). , The record entry specified by the record entry number [k] (k = (i × L + j) in the record table 3 [n] using the in-bucket key storage position number [j] (j = 0... L-1). 20 [k] (k = 0... K-1).

  By using N record tables 3 [n] (n = 0... N-1) made up of N memories in this way, when accessing each record value corresponding to a plurality of key values, each key is accessed. If the index table 2 [n] storing values is different, the record table 3 [n] storing each record value, that is, the memory is also different. For this reason, the database device 1 can access each record value simultaneously and in parallel, and the processing for accessing each key value can be parallelized to increase the throughput.

<Second Embodiment>
Hereinafter, the configuration of the database apparatus according to the second embodiment of the present invention will be described with reference to FIG. Note that in the database device 100 according to the second embodiment, the same reference numerals are used for the same constituent elements as those of the database device 1 shown in FIG.

  The database apparatus 100 is also realized by installing a computer program (software) in a computer (hardware) including a CPU (Central Processing Unit), a memory, an interface, and the like. This is realized by the cooperation of hardware resources and computer programs. Further, the computer program may be provided in a state of being stored in a computer-readable recording medium or a storage device, or may be provided via an electric communication line.

The database apparatus 100 includes N index tables 2 [n] (n = 0... N-1) for storing a plurality of key values (destination address values) of a registration key _Kr , and the index table 2 [n]. a record table 3 for storing a plurality of record values (transfer destination information) respectively corresponding to a plurality of key values (destination address values) stored in (n = 0... N-1), and a record value corresponding to the key value Search index table control unit 110 that controls search, index table control unit 111 for registration that controls registration of key values and record values, and search that performs control to acquire record values corresponding to key values from the record table 3 The record table control unit 120 for registration and the record table control unit 121 for registration that performs control to store the record value corresponding to the key value in the record table 3 are configured.

  That is, the difference from the database device 1 in the first embodiment is that the database device 1 uses one index table control unit 4 and one record table control unit 5, but the second embodiment The database apparatus 100 according to the embodiment is divided into a search index table control unit 110, a registration index table control unit 111, a search record table control unit 120, and a registration record table control unit 121. is there.

  The N index tables 2 [n] (n = 0... N-1) are the same as the configuration of the index table 2 [n] (n = 0... N-1) in the first embodiment. A memory that can be accessed simultaneously from the index table control unit 110 for search and the index table control unit 111 for registration is used. Here, for the index table control unit 110 for search and the index table control unit 111 for registration, entry number generation having the same function as the entry number generation unit 4a of the index table control unit 4 in the first embodiment. Portions 110a and 111a are provided.

  Similarly, the record table 3 has the same configuration as that of the first embodiment, but similarly to the N index tables 2 [n] (n = 0... N−1), the record table control for search is controlled. A memory that can be simultaneously accessed from the unit 120 and the record table control unit 121 for registration is used.

The index table control unit 111 for registration as the index table registration unit performs the key value of the registration key K _r by the same operation as the registration processing procedure (FIG. 4) of the index table control unit 4 in the first embodiment. Are registered in the index table 2 [n] (n = 0... N-1). The search index table control unit 110 as the index table search unit corresponds to the search key K _s by the same operation as the search processing procedure (FIG. 5) of the index table control unit 4 in the first embodiment. The record value of the record R _s is acquired from the record table 3.

As with the record table control unit 5 in the first embodiment, the registration record table control unit 121 as the record table registration unit uses the key value of the registration key K _r as the index table 2 [n] (n = 0... N−1), the record value of the record R _r corresponding to the key value is registered in the record table 3. The record table control unit 120 for search as the record table search unit is the record value of the record R _s corresponding to the key value of the search key K _s , similarly to the record table control unit 5 in the first embodiment. Is retrieved from the record table 3 and acquired.

In the database apparatus 1 (FIG. 1), the registration operation for the registration key K _r and the search operation based on the search key K _s are performed by the index table control unit 4 and the record table control unit 5 that _{serve both} for registration and search. Since the registration operation and the search operation are performed in order without simultaneously performing the above and the search operations, the search processing throughput may be reduced due to frequent registration processing.

On the other hand, in the database apparatus 100 (FIG. 6), the index table control unit 110 for search and the index table control unit 111 for registration simultaneously apply to the index table 2 [n] (n = 0... N−1). The key value registration process and the search process can be executed, and the record value registration process and the search process can be simultaneously executed for the record table 3 by the record table control unit 120 for search and the record table control unit 121 for registration. Thus, it is possible to obtain an effect that the throughput of the search process is not reduced by the registration process of the key _Kr .

<Third Embodiment>
Hereinafter, the configuration of the database apparatus according to the third embodiment of the present invention will be described with reference to FIG. Note that, in the database device 200 according to the third embodiment, the same components as those in the database device 1 shown in FIG.

  The database apparatus 200 is also realized by installing a computer program (software) in a computer (hardware) including a CPU (Central Processing Unit), a memory, an interface, and the like. This is realized by the cooperation of hardware resources and computer programs. Further, the computer program may be provided in a state of being stored in a computer-readable recording medium or a storage device, or may be provided via an electric communication line.

<Overall configuration of database device>
As shown in FIG. 8, the database apparatus 200 according to the third embodiment has N index tables 2 [n] (n = 0... N that store key values (destination address values) corresponding to a plurality of keys. -1), N for storing a plurality of record values (transfer destination information) respectively corresponding to a plurality of key values (destination address values) stored in the index table 2 [n] (n = 0... N-1) Keys for record table 3 [n] (n = 0 ... N-1), index table 2 [n] (n = 0 ... N-1) and record table 3 [n] (n = 0 ... N-1) N index table search units 210 [n] (n = 0... N-1) for controlling search operations of values (destination address values) and record values, key value index tables 2 [n] (n = 0. The index table registration unit 211 controls the registration operation for (N-1).

  In the database apparatus 200, N index tables 2 [n] (n = 0... N-1) and N record tables 3 [n] (n = 0... N-1) are associated with each other. It is composed of a set. For each set of the index table 2 [n] (n = 0... N-1) and the record table 3 [n] (n = 0... N-1), the index table registration unit 211 and the index table. Access is possible simultaneously from the search unit 210 [n] (n = 0... N-1).

  By the way, also in the database apparatus 200 in the third embodiment, the number N of index tables in the representative implementation is “2” as in the database apparatus 1 in the first embodiment. As described above, it is possible to greatly improve the memory use efficiency as compared with the conventional database apparatus with the minimum circuit scale.

  In the database apparatus 200, the number N of index tables is increased to “3” or more, that is, N hash functions f [n] (n = 0 to 3) corresponding to the number N of index tables are used. Is possible.

As a result, the database device 200 uses a plurality of hash functions f [n] (n = 0 to 3) as compared to a case where one hash function f [n] (n = 1) is used, thereby allowing a plurality of registrations. entry number 11 which is generated from the key K _r of use [n] (n = 0~3) it is possible to reduce the probability of collision. Thus, the database device 200, when the collision of the entry number 11 [i] occurs as in the conventional database device, a plurality of key values greater than or equal to the bucket size L in the bucket 10 [i] of the entry number 11 [i]. It is possible to prevent overflow due to the storage of.

<Configuration of index table>
The configuration of the index table 2 [n] (n = 0... N-1) is the same as that of the first embodiment, and the configuration is as shown in FIG. The index table 2 [n] (n = 0... N-1) is a memory, and each index table 2 [n] has M entries 11 [i] (i = 0... M-1). . This entry 11 [i] (i = 0... M-1) has L key storage positions 12 [i] [j] (i = 0... M-) that can store a plurality of key values (destination address values). 1) It is configured as a bucket 10 [i] (i = 0... M-1) provided with (j = 0... L-1).

  Therefore, the key storage position 12 [i] [j] (i = 0 ... M-1) (j = 0 ... L-1) provided for each index table 2 [n] (n = 0 ... N-1). Is the table size M of each index table 2 [n] (n = 0... N-1) (number of entries in the index table 2 [n] (number of buckets)) × bucket size L of each bucket (key storage location) 12 [i] [j]).

  The table size M and bucket size L of the index table 2 [n] (n = 0... N-1) may be different values for each index table 2 [n] (n = 0... N-1). Although possible, the index table size M and bucket size L in a typical implementation can be the same in all index tables 2 [n] (n = 0... N−1).

Thus, in the database apparatus 200, in the registration operation for the index table 2 [n] (n = 0... N-1) of the registration key _Kr , each index table 2 [n] (n = 0... N-1). index table 2 for storing the key value of the key K _r based on the minimum of the bucket length [nr] of the bucket [n] [n] (n = 0 ... n-1) selecting the processing (step SP5) I do. Thereby, in the database apparatus 200, as in the first embodiment, the memory of the index table 2 [n] (n = 0... N-1) and the record table 3 [n] (n = 0... N-1). The usage efficiency can be greatly improved.

<Structure of record table>
FIG. 9 shows a configuration example of the record table 3 [n] (n = 0... N−1). Here, since the record tables 3 [0],..., 3 [n],..., 3 [N-1] all have the same configuration, the record table 3 [n] will be described here for convenience.

  The record table 3 [n] is a memory and corresponds to the index table 2 [n], and the key storage position 12 [i] [j] (i = 0... M-1) (j) of the index table 2 [n] = 0 ... L-1) is provided with a record entry 20 [k] (k = 0 ... K-1) capable of storing record values (transfer destination information) corresponding to each of the key values (destination address values) stored in It has been.

  Therefore, the key storage position 12 [i] [j] (i = 0 ... M-1) (j = 0 ... L-1) provided for each index table 2 [n] (n = 0 ... N-1). As described above, the number of data is represented by the table size M of each index table 2 [n] (n = 0... N−1) × the bucket size L of each bucket. The record table size K of the record table 3 [n] when configured as one memory for storing one record value (transfer destination information) in the record entry 20 [k] (k = 0... K-1) is , K = M × L.

  Key storage position 12 [i] [j] (i = 0 ... M-1) (j = 0 ... L-1) of index table 2 [n] and record entry 20 [k] of record table 3 [n] (k = 0 ... K-1) is associated with the record entry number [k] in the key storage position 12 [i] [j] (i = 0 ... M-1) (j of the index table 2 [n]. = 0... L-1), any calculation method that can uniquely calculate the record entry number [k] of the record entry 20 [k] (k = 0... K-1) may be used.

For example, as one calculation method, entry number [i] (i = 0... M-1) indicating entry 11 [i] of index table 2 [n], key storage position 12 [i] [j] (i = 0 ... M-1) Using the in-bucket key storage position number [j] (j = 0 ... L-1) indicating (j = 0 ... L-1), the record entry 20 [k] (k = 0 ... The record entry number [k] of K-1) can be calculated by the following (Equation 4).
k = i × L + j ………………………………………………………… (Formula 4)

<Configuration of index table registration unit>
The index table registration unit 211 serving as an index table registration unit and a record table registration unit is configured to enter each index table 2 [n] (n = 0... N-1) based on a registration key K _r input from the outside. 11 hash functions f [n] (n = 0... For generating entry numbers [i] (i = 0... M-1) respectively specifying 11 [i] (i = 0... M-1). N-1).

Further, the index table registration unit 211 includes N pieces of hash values each corresponding to the N pieces of hash functions f [n] (n = 0... N−1) from the registration key K _r (destination address). An entry number generation unit 211a that generates entry number [n] (n = 0... N−1) is included.

The index table registration unit 211 stores the key K in the transfer database (each index table 2 [n] (n = 0... N-1) and record table 3 [n] (n = 0... N-1)) of the database device 1. when registering a record value of the record R _r corresponding to the _r key value and the key K _r, and inputs the corresponding record R _r key K _r and the key K _r for registration from the outside.

Similarly to the index table control unit 4 in the first embodiment, the index table registration unit 211 has the smallest bucket length [nr] of the index tables 2 [n] (n = 0... N−1). key value of the bucket [nr] is in conjunction with the store a key value of the key K _r key storage position of the unstored 12 [nr] [jr], the index table 2 [n] (n = 0 ... n -1) and record entry 20 [k] (k = 0 ... K) corresponding to the key storage position 12 [nr] [jr] in the record table 3 [n] (n = 0 ... N-1). The record value of the record R _r is stored in -1).

Incidentally, in the registration operation of the registration key K _r by the index table registration unit 211, each entry number [n] (n = 0... N-1) of each index table 2 [n] (n = 0... N-1). ) For each bucket 10 [n] (n = 0 ... N-1) specified by), set the bucket length (number of already stored key values) or empty bucket length (number of additional key values that can be stored) Need to get.

  For this reason, the index table registration unit 211 corresponds to each index table 2 [n] (n = 0... N-1) and each key of the index table 2 [n] (n = 0... N-1). Key storage map table MP [n] (n = indicating whether or not the key value is stored for each storage position [i] [j] (i = 0 ... M-1) (j = 0 ... L-1) 0 ... N-1).

  FIG. 10 shows a configuration example of the key storage map table MP [n] (n = 0... N−1). The actual state of the key storage map table MP [n] is a memory. The memory is a table having M entries 220 [i] (i = 0... M-1), and each entry 220 [i] (i = 0... M-1) is constituted by L key storage flags 222 [i] [j] (i = 0... M-1) (j = 0... L-1). Here, the key storage flag 222 [i] [j] (i = 0 ... M-1) (j = 0 ... L-1) is an entry of each entry 220 [i] (i = 0 ... M-1). It is specified by a number [i] (i = 0... M-1) and a key storage position number [j] (j = 0... L-1).

The key storage flag 222 [i] [j] (i = 0 ... M-1) (j = 0 ... L-1) is the key storage flag 222 [i] [j] (i = 0 ... M-1). Key storage position [i] [j] (i = 0 ... M-1) of index table 2 [n] (n = 0 ... N-1) corresponding to (j = 0 ... L-1) (j = 0 ... L-1 key values of key K _r for registration in) is 1-bit information indicating whether or not it is stored.

  The index table registration unit 211 stores the key storage flag 222 [i] [j] (i = 0 ... M-1) (j = 0 ... L) of the key storage map table MP [n] (n = 0 ... N-1). -1) is obtained, the key storage position [i] [j] (i = 0 ... M-1) (j of the index table 2 [n] (n = 0 ... N-1) is obtained. = 0 ... L-1), it is possible to know whether or not the key value is stored.

  In addition, the index table registration unit 211 acquires the bucket length or the empty bucket length of the bucket 10 [i] (i = 0 ... M-1) of the index table 2 [n] (n = 0 ... N-1). Are all key storage flags 222 [i] [j] (i = 0 ... M-1) (j = 0 ... L-) included in the key storage map table MP [n] (n = 0 ... N-1). 1), counting the number of flags indicating that key values are stored in bucket 10 [i] (i = 0... M-1) (or indicating that key values are not stored). Can be recognized by.

In the initial state (the state in which there is no key value registered for the index table 2 [n] (n = 0... N-1)), the index table registration unit 211 stores the key storage map table MP [n] (n = 0 ... N-1) of the previously clear all flags in accordance with the registration of the key value of the key K _r, key storage position of the registration key value [i] [j] (i = 0 ... M- 1) The flag of the key storage flag 222 [i] [j] (i = 0 ... M-1) (j = 0 ... L-1) corresponding to (j = 0 ... L-1) is set.

Therefore, when registering the key value of the key K _r , the index table registration unit 211 acquires one bucket 10 [[] for each index table 2 [n] (n = 0... N−1) in order to obtain the bucket length. i] (i = 0... M-1) is accessed and there is no need to count the number of key values stored in the bucket 10 [i] (i = 0... M-1). Access to the memory, which is the actual state of 2 [n] (n = 0... N−1), does not occur.

  In this case, the index table registration unit 211 accesses the memory that is the actual state of the key storage map table MP [n] (n = 0... N−1), but the index table 2 [n] (n = 0. Each entry 11 [i] (i = 0... M-1) of N-1) requires the number of bits that can store the key value, whereas the key storage map table MP [n] (n = 0 ... (N-1) each entry 220 [i] (i = 0... M-1) requires only one bit per key value. Therefore, the key storage map table MP [n] (n = 0... N-1) is accessed as compared to the case of accessing the memory of the index table 2 [n] (n = 0... N-1). This has the effect of consuming less power.

<Configuration of index table search unit>
Each index table search unit 210 [n] (n = 0... N-1) as an index table search unit and a record table search unit uses a search key K _s input from the outside as an index table 2 [n] (n = 0 ... N-1) to convert to entry number [n] (n = 0 ... N-1), one hash corresponding to each index table 2 [n] (n = 0 ... N-1) A function f [n] (n = 0... N-1) is provided. That is, the index table search unit 210 [0] includes a hash function f [0], the index table search unit 210 [n] includes a hash function f [n], and the index table search unit 210 [N−1] The function f [N-1] is provided.

When acquiring the record value of the record R _s corresponding to the key value of the search key K _s from the record table 3 [n] (n = 0... N−1), the index table search unit 210 [0] , enter the key K _s for the search from the outside, a plurality of keys stored in the bucket 10 [0] of the index table 2 which is designated [0] entry number [0] generated by the hash function f [0] A key value matching the search key K _s is searched from the values.

When the index table search unit 210 [0] finds a key value that matches the search key K _{s in} the bucket 10 [0] of the index table 2 [0], the index table 2 [0] of the found key value is found. Record entry 20 [k] (k = 0 ... K) of record table [0] corresponding to key storage position 12 [i] [j] (i = 0 ... M-1) (j = 0 ... L-1) The record value of the record R _s to be searched is acquired from -1).

Then, the index table search unit 210 [0] includes the record value of the record R _s , and the search effective search result H [that indicates that the record value of the record R _s corresponding to the key K _s has already been acquired. 0] is output to the next index table search unit 210 [1].

On the other hand, the index table search unit 210 [0], when not find the key value that matches the key K _s for the search in the bucket 10 [0] of the index table 2 [0], the next index table search unit 210 [1] by to find the key value that matches the key K _s for searches the index table 2 [1], and the key K _s, record value of the record R _s corresponding to the key K _s is not selected A search invalid search result H [1] indicating acquisition is output to the next index table search unit 210 [2]. Thereafter, the index table search unit 210 [n] (n = 0... N−1) repeats the same processing.

When the index table search unit 210 [n] (n = 0... N-1) receives a search result H [n-1] indicating that the search is valid from the previous index table search unit 210 [n-1], without searching for a key value that matches the key K _s for searches the index table 2 [n], which means it is a search enable the search enable means result H [n-1] result H [n] is output to the subsequent index table search unit 210 [n + 1].

Further, the index table search unit 210 [n] (n = 0... N-1) obtains the search result H [n-1] indicating invalid search and the search key K _s in the preceding index table search unit 210. When received from [n-1], a key value matching the search key K _s is searched from the index table 2 [n].

When the index table search unit 210 [n] (n = 0... N-1) finds a key value that matches the search key K _s from the index table 2 [n], the index table 2 [n] Among the records in the record table 3 [n] corresponding to the key storage position 12 [i] [j] (i = 0... M-1) (j = 0... L-1) in which the key value is stored. The record value of the record R _s corresponding to the key value is acquired from the entry 20 [k] (k = 0... K−1).

The index table search unit 210 [n] includes the record value of the record R _s , and the search effective search result H [n indicating that the record value of the record R _s corresponding to the key K _s has already been acquired. ] Is output to the subsequent stage.

Furthermore, if the index table search unit 210 [n] (n = 0... N-1) cannot find a key value that matches the search key K _s from the index table 2 [n], the key K _s and a search invalid search result H [n] indicating that the record value of the record R _s corresponding to the key K _s has not been acquired are output to the subsequent stage.

In the index table search unit 210 [n] (n = 0... N-1), the search by each index table search unit 210 [n] (n = 0... N-1) for the search key K _s is completed. Until then, there is no need to defer retrieval of the next key K _s . After each of the index table search units 210 [n] (n = 0... N-1) finishes the search operation for the search key K _s input from the outside or the previous stage and outputs the search result H [n]. The index table search unit 210 [n] (n = 0... N-1) accepts the input of the next search key K _s and starts a search operation, so-called pipeline processing is possible.

Also, the index table search unit 210 [n] (n = 0 ... N-1) in each index table search unit 210 [n] (n = 0 ... N-1) is to enter the key K _s for the search The time required to output the search result H [n] from whether or not there is a search (whether the input search result H [n-1] is valid or invalid) and the key value that matches the search key K _s Regardless of whether or not the index table [n] (n = 0... N-1) is found, it is constant. Thus, between the index table search unit 210 and the [n-1] and the index table search unit 210 [n], key K _s and search results for the search H [n] of the (n = 0 ... N-1 ) Pipeline processing without stagnation (ie, no bottlenecks) can be enabled.

<Registration process by index table registration unit>
Next, registration processing of the key K _r for registration index table registering unit 211 performs will be described with reference to a flowchart of FIG. 11.

In the database unit 200, to register a record R _r of the registration object corresponding to the key K _r and the key K _r for registration, the index table control unit 211, the key K _r and records R _r for registration from the outside Input (step SP21).

The index table registration unit 211 uses the N hash functions f [n] (n = 0... N-1) corresponding to the index tables 2 [n] (n = 0. N entry numbers [n] (n = 0... N−1) are generated simultaneously from the registration key K _r input from (step SP22 [0],..., Step SP22 [n],. Step SP22 [N-1]).

  The index table registration unit 211 stores each bucket 10 [n] (n) of each index table 2 [n] (n = 0 ... N-1) corresponding to each entry number [n] (n = 0 ... N-1). = 0 ... N-1) bucket length [n] (n = 0 ... N-1), the entry 220 [n] of each key storage map table MP [n] (n = 0 ... N-1) ] By counting the number of flags set in the key storage flag 222 [n] [j] (j = 0 ... L-1) included in (n = 0 ... N-1), each index table 2 [ n] The bucket length [n] (n = 0 ... N-1) of the bucket 10 [n] (n = 0 ... N-1) of (n = 0 ... N-1) is acquired (step SP23 [0] , ..., step SP23 [n], ..., step SP23 [N-1]). Here, [n] in the key storage flag 222 [n] [j] (j = 0... L−1) is the entry number [n].

  The index table registration unit 211 obtains the bucket length [n] (n = 0... N−1) instead of obtaining the bucket length [n] (n = 0 (n = 0) that is the number of key values that can be additionally stored. ... N-1) may be acquired.

The index table registration unit 211 _assigns the key value of the key K _{r to} any bucket 10 [n] (n = 0 ... N-1) of the index table 2 [n] (n = 0 ... N-1). In order to store, the bucket length [nr] that takes the minimum value is searched from each bucket length [n] (n = 0... N-1), and the index table 2 corresponding to the minimum bucket length [nr] [nr] is selected (step SP24).

  The index table registration unit 211 stores the key storage position 12 [nr] [n] in which the key value is not stored among the key storage positions 12 [nr] [j] (j = 0... L-1) of the index table 2 [nr]. jr] is searched (step SP25). Here, [nr] at the key storage position 12 [nr] [jr] is the entry number [nr] corresponding to the minimum bucket length [nr] that identifies the key storage position 12, and [jr] is the key This is the key storage position number [jr] in the bucket that identifies the key storage position whose value is not stored.

  Specifically, the index table registration unit 211 sets the key storage flag 222 [nr] [j] (j = 0... L-1) of the key storage map table MP [nr] corresponding to the index table 2 [nr]. All are read, and the key storage position 12 [nr] [jr] of the index table 2 [nr] corresponding to the key storage flag 222 [nr] [jr] indicating that the key value is not stored is searched for ( Step SP25).

The index table registration unit 211 determines whether or not the key storage position 12 [nr] [jr] of the index table 2 [nr] whose key value is not stored has been found (step SP26). entry 11 [nr] index table 2 [nr] stores the key value of the key K _r for registration to (bucket 10 [nr]) key storage location in the 12 [nr] [jr] (step SP27) .

The index table registration unit 211 assigns the key value of the registration key K _{r to} the key storage position 12 [nr] [jr] in the entry 11 [nr] (bucket 10 [nr]) of the index table 2 [nr]. Since it is stored, the in-bucket key storage position number [jr] of the entry number [nr] and the key storage position 12 [nr] [jr] is substituted into the above-described (Equation 4).

Accordingly, the index table registration unit 211 calculates the record entry number [kr] of the record entry 20 [kr] of the record table 3 [nr] by the following (Equation 5), and records the record entry 20 of the entry number [kr]. The record value of the record R _r is stored for [kr], and the registration result Tk indicating that the registration is valid is output to complete the registration process (step SP28).
kr = entry number [nr] x L + jr ………………………………………………………… (Formula 5)

On the other hand, the index table registration unit 211 does not find the key storage position 12 [nr] [jr] of the index table 2 [nr] in which the key value is not stored (step SP26: No), that is, for registration. if you can not register the key value of the key K _r, as a process at the time of registration impossible, and terminates the registration process by outputting the registration result Tf indicating an invalid registration (step SP29).

Incidentally index table registration unit 211, along with outputting the registration result Tf indicating an invalid registration, the number of times that can not register a key value of the key K _r for registration counted by the counter, and updates whenever the counting, the A count value can be held.

In addition, the index table registration unit 211 outputs a registration result Tf indicating registration invalidity and, at the same time, stores the key already stored in the entry 11 [nr] (bucket 10 [nr]) of the index table 2 [nr]. among the values, unregister the one key value, and registers the key value of the key K _r for registration for the deletion to the key storage location 12, record table 3 corresponding to the key storage location 12 The record value of the record R _r can also be stored in the record entry 20 [k] of [n].

  The index table registration unit 211 takes the minimum value from the bucket lengths [n] (n = 0... N-1) based on the key storage map table MP [n] (n = 0... N-1). The bucket length [nr] is searched, and the index table 2 [nr] corresponding to the minimum bucket length [nr] is selected (step SP24). As the index table selection method for selecting the power index table 2 [nr], there are the following three other methods.

  In the first index table selection method, the empty bucket length [n] (n = 0... N-1) is larger than 0 based on the key storage map table MP [n] (n = 0... N-1). An index table 2 [nr] having an index table number [nr] having the smallest index table number [n] (n = 0 ... N-1) is selected from the bucket 10 [n] (n = 0 ... N-1). select.

  The second index table selection method is based on the key storage map table MP [n] (n = 0 ... N-1), and the maximum value is selected from the free bucket lengths [n] (n = 0 ... N-1). The index table 2 [nr] having the index table number [nr] that becomes the empty bucket length [nr] is selected.

  The third index table selection method is such that when each index table 2 [n] (n = 0... N-1) has a different table size M (number of entries 11 [i]), the key storage map table MP. [n] The empty bucket length [n] (n = 0 ... N-1) acquired based on (n = 0 ... N-1) is obtained for each index table 2 [n] (n = 0 ... N-1). Index with index table number [nr] that has become the maximum weighted free bucket length [nr] out of weighted free bucket length [n] (n = 0 ... N-1) weighted with the set weighting factor Select table 2 [nr].

  In the first to third index table selection methods, the bucket size L of each index table 2 [n] (n = 0... N−1) (key storage position 12 [nr] [jr] (i = 0 ... M-1) (number of j = 0 ... L-1)) is also different.

<Search processing by index table search unit>
Next, with reference to the flowchart of FIG. 12, the search processing of the record R _s to be searched based on the search key K _s performed by each index table search unit 210 [n] (n = 0... N−1). While explaining.

In the database apparatus 200, when searching for a search key K _s and a search target record R _r registered corresponding to the key K _s , an index table search unit 210 [n] (n = 0... N− 1) The search key K _s from the external or previous index table search unit 210 [n-1] and the search result H [n-1] from the previous index table search unit 210 [n-1] Input (step SP31).

Here, in the index table search unit 210 [n] (n = 0... N−1), when the index table search unit 210 [n] is [n] = 0, the index table search unit 210 [ Since n-1] does not exist, only the search key K _s is input. However, when the index table search unit 210 [n] is [n]> 0, since the previous index table search unit 210 [n-1] exists, the search key k _s and the search result from the previous step are present. Input both H [n-1].

  The index table search unit 210 [n] (n = 0... N-1) itself is the index table search unit 210 [0] with [n] = 0, or the previous index table search unit 210 [n]. -1] is determined whether a search invalid search result H [n-1] is received (step SP32).

  The index table search unit 210 [n] (n = 0... N-1) is not the index table search unit 210 [0] with [n] = 0 (step SP32: NO), or the previous index table When no search invalid search result H [n-1] is received from the search unit 210 [n-1], that is, when a search valid search result H [n-1] is received (step SP32: NO), The search valid search result H [n-1] is output to the subsequent stage as the search result H [n] as it is (step SP33), and the process is terminated.

The index table search unit 210 [n] (n = 0... N-1) is the index table search unit 210 [0] with [n] = 0 (step SP32: YES) or the previous index. When an invalid search result H [n-1] is received from the table search unit 210 [n-1] (step SP32: YES), one hash function f [n] (n = 0... N-1) possessed by itself ) to generate the entry number [is] to point the search the index table 2 [n] from the key K _s for search using (step SP34).

  The index table search unit 210 [n] (n = 0... N-1) is an index table 2 [n] (n = 0) corresponding to the index table search unit 210 [n] (n = 0 ... N-1). ... (N-1) The maximum L key values stored in the bucket [is] of the entry number [is] are acquired (step SP35).

The index table search unit 210 [n] (n = 0... N-1) obtains the maximum L obtained from the bucket [is] of the entry number [is] in the index table 2 [n] (n = 0 ... N-1). Among the key values, the key storage position 12 [is] [js] in which the key value that matches the search key K _s is stored is searched (step SP36).

The index table search unit 210 [n] (n = 0... N-1) determines whether or not a key value matching the search key K _s has been found among the maximum L key values (step S1). SP37), if not found (step SP37: NO), and outputs the search result indicating the invalid search key K _s for the search H [n] to the subsequent stage (step SP38), and terminates the process.

The index table search unit 210 [n] (n = 0... N−1) finds a key value that matches the search key K _s among the maximum L key values (step SP37: YES). , The record entry number [ks] of the record table 3 [n] corresponding to the key storage position 12 [is] [js] is stored in the bucket key of the entry number [is] and the key storage position 12 [is] [js]. based on the position number [js] calculated by (equation 6), to obtain the record value of the search target record R _s from the record entry 20 [ks] of the record entry number [ks] (step SP39).
ks = entry number [is] × L + js ………………………………………………………… (Formula 6)

Since the index table search unit 210 [n] (n = 0... N-1) has acquired the record value of the record R _s to be searched, it includes the record value of the record R _s and includes the search key K. Search effective search results indicating that the search target record R _s corresponding to _s is already obtained the H [n] is output to the subsequent stage (step SP40), and terminates the process.

<Effect in the third embodiment>
Index table search unit 210 [n] is the key that matches the key K _s for retrieved from the index table 2 corresponding to the index table search unit 210 [n] [n] ( n = 0 ... N-1) If the value can be found, the search effective search result H [n] can be output to the index table search unit 210 [n + 1] at the subsequent stage.

Further, when the index table search unit 210 [n] (n = 0... N-1) receives the search valid search result H [n-1] from the previous index table search unit 210 [n-1], index table 2 [n] without performing the search key value that matches the key K _s for the search from among the (n = 0 ... n-1 ), the search result H [n-1] as it results H [n] can be output to the subsequent index table search unit 210 [n + 1].

Thus the index table search unit 210 [n] (n = 0 ... N-1) is an index table 2 [n] key value that matches the key K _s for the search from among the (n = 0 ... N-1 ) If the index table search unit 210 [n + 1], the index table search unit 210 [n + 2],... The index table 2 [n + 1], the index table 2 [n + 2],... For searching for a key value that matches the key K _s of the index key 2 is not accessed.

Therefore, when the search key K _s is input, the index table 2 [n] (n = 0... N-1) respectively corresponding to the index table search unit 210 [n] (n = 0... N-1). Compared to the case where the key value matching the search key K _s is searched, the index table search unit 210 [n] (n = 0... N−1) is compared with the index table 2 [ n] (n = 0 ... N-1) can be reduced in frequency of access, and as a result, power consumption caused by accessing index table 2 [n] (n = 0 ... N-1) can be reduced. Can do.

Note that the registration operation of the registration key K _r by the index table registration unit 211 is performed by using the key storage map table MP [n] (n = 0... N−1) to register the key K _r for registration. Only the index table 2 [n] to store the values is accessed, and the other index tables 2 [n] are not accessed. Thereby, each index table search unit 210 [n] (n = 0... N-1) does not need to stop the pipeline operation of the search process, and the search invalid search result H [n-1] is obtained from the previous stage. it is not necessary to stop the access to the index table 2 [n] for carrying out the search for the key value that matches the key K _s for search when given.

  Further, in the index table registration unit 211, all the key storage maps as in step SP23 [0],..., Step SP23 [n],..., Step SP [N-1] in the above-described registration process (FIG. 11). The table MP [n] (n = 0... N-1) is accessed simultaneously. On the other hand, when a configuration is used in which each key storage map table MP [n] (n = 0... N−1) is sequentially accessed by pipeline processing, all the key storage map tables MP [n] ( If access to n = 0... N-1) is not completed, the index table 2 [n] (n = 0... N-1) with the smallest bucket length cannot be selected. For this reason, the access to the invalid key storage map table MP [n] (n = 0... N-1) that does not need to be selected cannot be stopped, and the power consumption cannot be reduced accordingly. Will occur.

Further, in the configuration in which each key storage map table MP [n] (n = 0... N-1) is accessed in order by the pipeline processing as described above, from the registration start of the registration key _Kr to the registration completion. Therefore, there is a problem that the search process for the key _Kr cannot be executed until the registration key _Kr is completely registered.

  However, in the database apparatus 200 in the third embodiment, the above-described problems do not occur, and the memory use efficiency can be enhanced as in the first embodiment, and at the same time, the power consumption effect Can be improved.

<Other embodiments>
In the above-described second embodiment, the case where one record table 3 (FIG. 7) is used has been described. However, the present invention is not limited to this and corresponds to the first embodiment. As in the other embodiments, a record table 3 [n] (n = 0... N-1) made up of N memories may be used.

  DESCRIPTION OF SYMBOLS 1,100,200 ... Database apparatus, 2 [n] ... Index table, 3, 3 [n] ... Record table, 4 ... Index table control part (index table registration part, index table search part), 5 ... Record table control (Record table registration unit, record table search unit), 10 [i] ... bucket, 11 [i] ... entry, 12 [i] [j] ... key storage location, 20 [k] ... record entry, 110 ... search Index table control unit (index table search unit), 111 ... registration index table control unit (index table registration unit), 120 ... search record table control unit (record table search unit), 121 ... registration Record table control unit (record table registration unit), 210 [n] ... index table search unit (index table search unit, record table search unit), 211 ... index table registration unit Index table registration unit, record table registration unit), 222 [i] [j] ... key storage flag, MP ... key storage map table.

Claims (8)

A plurality of key storage locations capable of storing a plurality of key values, and N (N is an integer of 2 or more) index tables each having a plurality of buckets respectively associated with different entry numbers;
N hash functions different from each other corresponding to each of the N index tables are provided, and each of the N index tables corresponds to each of the N index tables based on a key value of a key input from the outside using the N hash functions. An entry number generation unit for generating N entry numbers,
A record table storing a record value corresponding to the key value in a record storage position corresponding to the key storage position in the bucket;
An index table registration unit that selects a bucket for storing the key value from N buckets in the N index tables corresponding to the N entry numbers, and stores the key value in the bucket;
A database apparatus comprising: a record table registration unit that stores a record value corresponding to the key value at the record storage position of the record table corresponding to the key storage position where the key value is stored. The database apparatus according to claim 1, wherein
An index table search unit for searching for the bucket in which a key value matching the key is stored from N buckets in the N index tables corresponding to the N entry numbers;
A record table search unit for acquiring the record value from the record storage position of the record table corresponding to the key storage position in which the key value is stored in the bucket searched by the index table search unit; A database device. The database apparatus according to claim 1, wherein
The index table registration unit includes a bucket length that is the number of key values that are already stored among the N buckets, or an empty bucket length that is the number of key values that can be additionally stored among the N buckets. Based on the above, a database device is selected from the N buckets corresponding to the N entry numbers in the N index tables. The database apparatus according to claim 1, wherein
The N index tables are accessible simultaneously from both the index table registration unit and the index table search unit,
The database apparatus, wherein the record table is accessible simultaneously from both the record table registration unit and the record table search unit. The database apparatus according to claim 1, wherein
The record table is N record tables having a plurality of record storage positions capable of storing a plurality of record values and having a plurality of buckets respectively associated with different record entry numbers. apparatus. The database apparatus according to claim 1, wherein
The index table registration unit refers to a key storage map table indicating whether or not the key value is stored in all the key storage positions in the N index tables, and the N index tables store the N The database apparatus, wherein the bucket that stores the key value is selected from N buckets corresponding to the entry numbers. A plurality of key storage positions capable of storing a plurality of key values are provided, and N (N is an integer of 2 or more) index tables each having a plurality of buckets respectively associated with different entry numbers. The entry number generation unit having different N hash functions has N entry numbers respectively corresponding to the N index tables based on key values of keys input from the outside using the N hash functions. An entry number generation step for generating
An index table in which the index table registration unit selects a bucket storing the key value from N buckets corresponding to the N entry numbers in the N index tables, and stores the key value in the bucket. Registration step;
The record table registration unit includes a record table registration step of storing a record value corresponding to the key value at a record storage position of the record table corresponding to the key storage position where the key value is stored in the bucket. Database management method. Against the computer
A plurality of key storage positions capable of storing a plurality of key values are provided, and N (N is an integer of 2 or more) index tables each having a plurality of buckets respectively associated with different entry numbers. An entry number generating step for generating N entry numbers respectively corresponding to the N index tables based on key values of keys input from the outside using different N hash functions;
An index table registration step of selecting a bucket for storing the key value from N buckets corresponding to the N entry numbers in the N index tables, and storing the key value in the bucket;
A record table registration step of executing a record table registration step of storing a record value corresponding to the key value at a record storage position of a record table corresponding to the key storage position storing the key value in the bucket.

Images