JPH1049545A - Associative storage device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce partiality of data and a decrease conflicts by generating an address of a table on the basis of a virtual key and making access to an entry possible with the generated address. SOLUTION: This device is equipped with a virtual key generation part 8, an address generation part 3, a table 7, and a control part 5, and in the table 7, an entry consisting of one key and a record corresponding to it is made to correspond to each address. This device registers and retrieve data (composed of a key and a record) to each entry of the table 7 under the control of the control pat 5, but the virtual key generation part 8 generates a virtual key in this case from a given key by a hash function and the address generation part 3 generates an address on the basis of the virtual key. Then, each entry of the table 7 is accessed with the generated address to register and retrieve data.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an associative memory device that can be used for various computers, and more particularly to an associative memory device using a hashing technique. An associative storage device is a device that searches for a record associated with a given key, and is generally used for devices that require management of a symbol table of a computer language processing system and data associative search in a database or the like. ing.

[0002]

2. Description of the Related Art A conventional example will be described below with reference to the drawings. (1): Description of an associative storage device and a hashing device FIG. 15 is an explanatory diagram 1 of a conventional example. FIG. 15A shows an associative storage device, and FIG. FIG. 16 is an explanatory diagram 2 of the conventional example. FIG. 16A shows an open address hash device, and FIG. 16B shows a chain hash device (in the case of a two-branch tree).

[0003] For example, commentary by Seiichi Nishihara, "Hashing Techniques and Applications" (Information Processing Society of Japan, Vol. 21, No.
9, Sep. 1980), a device using a hashing method as a basic technique for realizing an associative storage device such as a symbol table used in a processing system of a computer language (hereinafter, referred to as a “hashing device” or a “hash device”). Is widely used.

As shown in FIG. 15A, an associative storage device uses a key k (usually a character string) given to data, and associates the data (hereinafter referred to as "character k") with the key k. Record). In the search processing, related information is added as needed.
The concept of the associative search device can be represented as a table 7 having a key k and a record.

For example, in a computer language processing system, in order to analyze a language, an identifier such as a reserved word or a variable name appearing in a program is checked whether it is a reserved word or not. It is necessary to store and use various information such as what type is for the name. In such a situation, an associative memory is used to obtain the relevant information from the identifier.

[0006] A hashing device is a type of associative search device. As shown in FIG. 15B, in the hashing device, each entry (a set of one key and a record corresponding thereto) of the content addressable storage device (corresponding to the table 7) has addresses (addresses) 1, 2,. 3, 4, 5, ...
・ Has been shaken. For example, in the case of a table having m entries, it can be considered that addresses (addresses) from 0 to m-1 are assigned.

In the search in the hashing device, first, the address for storing the record is calculated by the address generating unit 3 from the given key k. The value obtained as an address must be greater than or equal to 0 and less than m. A function for calculating an address from a key k is called a “hash function”. Key k
The value that can be taken as is called a “key set” and is represented by K. A set that can be taken as an address is called an “address set” and is represented by A. Accordingly, the hash function h (corresponding to the internal function of the address generation unit 3) is a function of h: K → A.

If the entry of the address calculated by using the hash function is free, the key k and the record associated therewith are stored therein. When searching, the address is calculated in the same way from the given key k, and the key k is calculated.
Look for an entry with. In general, the number of entries in the table is much smaller than the number of keys that can be recognized (the number of elements of the set K, including the case where the number is theoretically infinite), so that the same address is calculated from different keys. Occur. This is called a "collision."

Several methods have been proposed to deal with the collision. As the method, there is a method called an open address method or a chain method. An open-address hash device using the open address method is shown in FIG. 16A, and a chain hash device using the chain method is shown in FIG. 16B.

In the open address hashing apparatus, the address generating unit 3 calculates the address of the table 7 from the given key k, and when a collision occurs, the address recalculating unit 4 instructs the address recalculating unit 4 according to an instruction from the control unit 5. This is a method of calculating a new address based on the assigned address. In calculating this new address,
In many cases, a new address is calculated by adding an increment value to the original address by some method and adding it.

Further, in the above-mentioned chaining hash apparatus, link information is added to each entry of the table 7 and connected to an entry provided inside or outside the table 7. That is, the address generation unit 3 calculates an address from the given key k, and when a collision occurs, further connects the link using the key k in accordance with an instruction from the control unit 5. Conventionally, a method of combining the open address method and the chain method has also been proposed.

§2: Processing explanation of open address hashing device
... See FIG. 17 FIG. 17 is a processing flowchart of the conventional open address hashing apparatus. Hereinafter, the processing of the open address hash device will be described with reference to FIG. In addition, S1 to S6 indicate each processing step.

In this processing, when a key k is given from outside (S1), the address generating unit 3 calculates a variable a by a hash function h (k) using the given key k.
Then, the address generating unit 3 generates an address indicated by the variable a (hereinafter, simply referred to as “address a”) (S2).

Next, the control unit 5 determines that the entry at address a is empty. Is $ table (a) empty? ｝ Is determined (S
3) If it is free, the key k and the record corresponding to it are registered in the entry of the address a (S6), and the registration work is completed. However, if the entry at the address a is not empty, the control unit 5 determines whether the key registered at the address a matches the given key k or {key (a) = k? The judgment is made by comparing Ｓ (S4).

As a result, if they match, the key is already registered, and the process ends. However, if the keys do not match, the address recalculation unit 4 calculates a new address for registration. Then, a registration operation is performed using the new address obtained by the above calculation. This operation is performed until an empty entry is found (S5).

In practice, it is necessary to perform processing when there is no free space in the table 7. However, in this example, the processing is omitted for simplicity, and it is assumed that there is free space in the table 7. . In the open address method, a collision may occur again at the calculated new address. In such a case, calculation of the new address is continued until a free entry is found. Hereafter, in order to distinguish between collisions that occur in the first address calculation and those that occur in such a recalculation,
The former is called "primary collision" and the latter is called "secondary collision".

§3: Description of the processing of the chain hash device
FIG. 18 is a processing flowchart of a conventional chain hashing apparatus. Hereinafter, the processing of the chain hash device will be described with reference to FIG. In addition, S11 to S21 indicate each processing step.

Hereinafter, for the sake of simplicity, a case of a two-branch tree structure will be described. In this process, when a key k is given from outside (S11), the address generation unit 3 calculates an address a by a hash function h based on the given key k (S12). Next, the control unit 5 determines that the entry at the address a is empty. Is $ table (a) empty? ｝ Is determined (S
13) If it is vacant, the key k and the record corresponding to it are registered in the entry of the address a (S16), and the registration work ends.

However, if the entry at the address a is not empty, the control unit 5 determines whether the key registered at the address a matches the given key k or ｛key (a) = k? The judgment is made by comparing Ｓ (S14). As a result, if they match, the key is already registered, and the process ends. But,
If the keys do not match, that is, if a record having another key is already registered in the entry of the address a to be registered, the key to be registered and the already registered key {k: key (a)} Are compared using some kind of comparison mechanism (S15).

As a result, when the key k is larger, Δk> k
ey (a)} follows the link for the large key. When the key k is smaller, {k <key (a)} follows the link for the smaller key. At this time, when tracing the link for the large key, is there a link that connects to the larger one? Is greater-link (a) empty? ｝
It is determined whether or not the link for the small key is free (S18).
Is ess-link (a) empty? It is determined whether or not (S17)
I do.

As a result, when it is determined in S17 and S18 that the link is empty, it means that the registration with the key has not been performed, and the record of the key k to be registered is registered. I do. That is, a new entry is created and registered (S20).

If there is an entry already registered in the link for tracing (S17, S1)
8) Follow the smaller link or the larger link (S19, S21). And again, that key and k
Compare. By repeating this, the key k to be registered and the record corresponding thereto are registered at an appropriate place in the tree structure.

§4: Description of Experimental Example 1—See FIG. 19 FIG. 19 is an explanatory diagram of Conventional Experimental Example 1. Experimental example 1
17 is an example in which an experiment was performed using the open address method hash device illustrated in FIG. 16. In this experiment, an experiment was performed using the data shown in the figure, and an experimental result was obtained.

As the experimental data, the key k is a,
aa, ab, abc, b, bc, cd, cc, bb,
19 keys (character strings) consisting of d, e,... bcb, and the size of the table (number of addresses) was 25. In this case, the number of primary collisions was nine. Then, the address at the time of collision is recalculated by adding 1 (the original address a +
1) A new address is generated by taking the remainder after dividing by the table size (table size). The experimental results were as shown in the figure, and the number of secondary collisions was 78.

§5: Description of Experimental Example 2—See FIG. 20 FIG. 20 is an explanatory diagram of Conventional Experimental Example 2. Experimental example 2
17 shows an example in which an experiment was performed using the chain hash device shown in FIG. 16B. In this experiment, the same data as in Experimental Example 1 was used, and an experimental result was obtained.

That is, as in the experimental data, the key k is a, aa, ab, ab
c, b, bc, cd, cc, bb, d, e,.
19 keys (character strings) consisting of cb, and the size (number of addresses) of the table was 25. In this case, the number of primary collisions was nine. The depth of the two-branch link was deep as shown (the number of bifurcations is large).

[0027]

The above-mentioned prior art has the following problems. (1): If no collision occurs, the hash device can find the entry for any key by a single calculation of the hash function, which is an extremely efficient method. However, compared to the size of the key set K,
Since the address set A is extremely small, it is inevitable that the efficiency is deteriorated due to the collision. Therefore, how to prevent collisions (primary and secondary collisions) is important in designing an efficient hashing device, but it is difficult to reduce the number of collisions with the conventional device.

(2): Number of table entries (address set A
When the number of elements is m and n entries are empty, an empty entry is found in the calculation of the first hash function with the probability of n / m. But,
It is usually worse because of the offset in the data distribution that actually occurs. In the hashing method, the number of recalculations of addresses is also larger than an ideal value due to a secondary collision that occurs due to data deviation.

The present invention solves such a conventional problem, and effectively uses virtual addresses to reduce data deviation and reduce collisions (particularly, secondary collisions) to achieve an ideal data. It is an object to realize a high-speed hash device close to a state. Further, the present invention, when a key has a certain structure, realizes not only the speed of access of the associative search device but also the flexibility of access by using a virtual key calculation device reflecting the structure. With the goal.

[0030]

FIG. 1 is a diagram illustrating the principle of the present invention. The present invention is configured as follows to achieve the above object. (1): The associative storage device has a plurality of entries each consisting of one key and a record set corresponding to the key, and uses a hash function based on a table 7 in which addresses are assigned to the respective entries and a given key. A virtual key generator 8 for generating a virtual key by using the virtual key generator 8 and an address generator 3 for generating an address of the table based on the virtual key generated by the virtual key generator 8. The access to each entry is enabled by the address.

(2): In the associative memory device of (1),
An address recalculation unit 4 is provided for recalculating an address based on the virtual key generated by the virtual key generation unit 8 when a collision occurs due to the generation of an address by the address generation unit 3.

(3): In the associative memory device of (1),
A plurality of branch links are provided in each entry of the table 7, and when a collision occurs due to the generation of an address by the address generation unit 3, a tree structure is generated using the link by comparing the sizes of the virtual keys. It has a tree structure generation control means for performing the following.

(4): A table having a plurality of entries each consisting of a set of a key and a standard key and a record corresponding to the key, and using a hash function based on a table in which addresses are assigned to each entry and a given key. A standard key generator for generating a standard key, a virtual key generator for generating a virtual key using a hash function based on the standard key generated by the standard key generator, An address generating unit for generating an address of the table based on the virtual key is provided, and each of the entries can be accessed by an address generated by using a hashing method.

(Operation) The operation of the present invention based on the above configuration will be described with reference to FIG. (1): Action of the above (1) An entry consisting of a key and a record is accessed through a two-step operation as follows. First, in the first stage,
When given a key k, the virtual key generator 8 generates a virtual key by a hash function based on the key k. Next, in a second stage, the address generating unit 3 generates an address reflecting the size of the actual table from the obtained virtual key. When the address is obtained in this way, access to the entry (data registration / search) is performed using the address.

As described above, by dividing the address calculation processing that has been performed at once by one associative storage device into two stages, the virtual key generation unit 8 and the address generation unit 3 meet the purpose. Transformation methods can be used. The virtual key generation unit 8 can reduce collision by generating a virtual key with a small bias for keys that appear with a bias.

(2): Operation of the above (2) In the associative memory device, the address recalculation unit 4 generates an address in the address generation unit 3 and, when a collision occurs, the virtual key generation unit 8 The address is recalculated based on the generated virtual key.

In the recalculation of the address, since the virtual key appears uniformly, the use of the virtual key makes it possible to easily calculate a new address with less bias even in the case of recalculation. It can be expected that secondary collisions will be reduced.

(3): Operation of the above (3) In the associative memory device, when the address is generated by the address generating unit 3 and a collision occurs, the tree structure generation control means compares the virtual keys with each other. , A tree structure is generated using the link.

In this case, only when the virtual keys match, the whole comparison is performed by comparing the magnitudes of the keys. Since the bias of the appearance of the virtual key is eliminated, for example, if the link is a two-branch tree, compared to the case of comparing two keys,
In comparison of virtual keys, the appearance ratio of large and small is almost 1/2.
become.

As a result, it is expected that the created tree structure is close to a well-balanced tree which is an ideal state, and that the depth of the tree with respect to the number of entries becomes almost the minimum value. As a result, the number of searches for the tree structure for registration and search for the entry is reduced, which contributes to an improvement in processing efficiency.

(4): Operation of the above (4) When a key is given, the standard key generation unit generates a standard key using a hash function based on the given key.
The virtual key generation unit generates a virtual key using a hash function based on the standard key generated by the standard key generation unit. Thereafter, the address generating unit generates an address of the table based on the virtual key generated by the virtual key generating unit.

As described above, instead of directly using the original key, the key is converted to a standard form and is used as a key for registration / search. In this way, by using the standard key, not only the search by the original key but also the search by the standard key is possible, and a kind of semantic search becomes possible. In addition, the processing speed can be increased by reducing collisions, and the flexibility of the processing is improved.

(5): Others As described above, by effectively using the virtual key, the deviation of the data is reduced, and the collision (particularly, the secondary collision) is reduced, so that the high speed close to the ideal state is achieved. A simple associative storage device (hash device) can be realized. Further, when a key has a certain structure, by using a virtual key generation unit reflecting the structure, not only the access speed of the associative memory device can be increased but also the access flexibility can be realized.

[0044]

Embodiments of the present invention will be described below in detail with reference to the drawings. In the following description, a device using the hashing technique is referred to as a “hash device”.
Further, in all of the above-described apparatus examples, a control unit for performing control based on the hashing technique is provided, but is not shown except for some of the apparatus examples. Further, a virtual key k ₁ is obtained from a given key k by a hash function h, and a variable a obtained by a certain function (a function of the same kind as the hash function h) ad (k ₁ ) based on the virtual key k ₁ Is simply referred to as “address a”.

§1: Description of Examples of Various Apparatus ... FIGS. 2 to 6
Reference is made to FIG. 2 for explanatory views of the apparatus examples 1 and 2 (FIG. A: apparatus example 1, FIG. B: apparatus example 2), FIG. 3 is an explanatory view of apparatus example 3, and FIG.
FIG. 5 is an explanatory view of an apparatus example 5, and FIG. 6 is an explanatory view of an apparatus example 6. Hereinafter, various examples of the apparatus will be described with reference to FIGS.

(1): Description of Apparatus Example 1 Refer to FIG. 2A. Apparatus Example 1 shown in FIG. 2A is an example of a hash apparatus using a virtual key. This device includes a virtual key generation unit 8
, An address generating unit 3, a table 7, and a control unit 5. The table 7 includes one for each address (address).
An entry consisting of one key and the corresponding record is associated with it.

In this apparatus, registration processing of data (key + record data) to each entry of the table 7 and search processing are performed under the control of the control unit 5. In this case, the virtual key generation unit 8 is provided. A virtual key is generated (generated) by a hash function from the generated key, and the address generation unit 3 generates an address (address) based on the virtual key. Then, each entry of the table 7 is accessed at the generated address, and data registration / search processing is performed.

(2): Description of Apparatus Example 2—See FIG. 2B The apparatus example 2 shown in FIG. 2B is an example of an open address hashing apparatus. This open address hashing apparatus includes a virtual key generation unit 8, an address generation unit 3, an address recalculation unit 4, a control unit 5
And a table 7. The table 7 includes:
One key and an entry composed of a record corresponding to the key are associated with each address (address).

In this apparatus, data (key + record data) is registered and searched for in each entry of the table 7 under the control of the control unit 5. In this case, the virtual key generating unit 8 is provided. A virtual key is generated (generated) from the key, and the address generating unit 3 generates an address (address) based on the virtual key. Then, the table 7 is accessed by the generated address, and the table data is registered / retrieved.

In the above processing, when the address is calculated by the address generator 3 and a primary collision occurs, the address recalculator 4 is controlled by the controller 5 so that the virtual key generator 8 generates the address. The address is calculated again based on the key. And
If the address recalculation unit 4 recalculates the address and a secondary collision occurs again, the address recalculation unit 4 calculates the address again. In this way, a new address is calculated from the virtual key until the collision is eliminated, and the new address is used to access the table 7 to register / search for data.

(3): Description of Apparatus Example 3—See FIG. 3 The apparatus example 3 shown in FIG. 3 is a virtual key hash apparatus having a collision processing mechanism by a chain method (hereinafter, this apparatus is referred to as a “chain method hash”). Device "). In this case, in order to simplify the description, the apparatus is shown as an example of a two-branch tree.

This chain hash device (in the case of a two-branch tree) includes a virtual key generation unit 8, an address generation unit 3, a control unit (not shown), and a table 7. Is associated with one key, a virtual key corresponding to the key, a record, and an entry composed of a two-branch link for each address (address).

In this apparatus, registration of data (key + virtual key + record + data of two branch links) in the table 7 and search processing are performed under the control of the control unit. In this case, the virtual key generation unit 8 uses the hash function. Then, a virtual key is generated (generated) from the given key, and the address generating unit 3 generates an address (address) based on the virtual key. Then, the table 7 is accessed at the generated address, and data registration / search processing is performed.

When a primary collision occurs in the above processing, a tree structure is generated without calculating a new address. Therefore, each entry of the table 7 is provided with a two-branch link. One is an entry link that registers a record with a key greater than the entry's key, and the other is
A link to a small key entry. The feature of this device is that the comparison of the keys is not based on the conventional comparison of the original keys, but rather by using a virtual key generated from the keys.

(4): Description of the device example 4—see FIG. 4 The device example 4 shown in FIG. 4 combines the device example 2 (open address method hash device) and the device example 3 (chain method hash device). FIG. This apparatus includes a virtual key generation unit 8, an address generation unit 3, an address recalculation unit 4, and a control unit (not shown).
And a table 7. The table 7 includes:
One key, a corresponding virtual key, a record, and an entry composed of a two-branch link are associated with each address (address).

In this device, when a primary collision occurs,
Up to a certain number of times, the above-mentioned device example 2 (open address method hash device)
Using the recalculation mechanism (address recalculation unit 4), which is a feature of
Find a free entry by calculating the new address. If the number exceeds that, an entry is secured by using a link mechanism which is a feature of the device of the device example 3 (chain hash device),
It accesses (registers / searches) there. The number of times to move to the link from recalculation is various methods such as a case where a certain number is predetermined, a case where it is determined by calculation using keys, virtual keys, and addresses, and a case where a specific condition is satisfied. It is possible.

(5): Explanation of Example 5 of Apparatus—See FIG. 5 In Example 5 of apparatus, a given key is converted to a standard form without directly using an original key, and the converted key is used as a key for registration / retrieval. As an associative storage device (associative storage device with standard key conversion). This device includes a standard key generator 12, a virtual key generator 8, an address generator 3, and a controller (not shown).
And a table 7. And the table 7
Has a key, a standard key, and a record.

In the device example 5, the standard key generator 12 generates a standard key from a given key, the virtual key generator 8 generates a virtual key from the standard key, and the address generator 3 generates the virtual key from the virtual key. Generates an address from. Then, the table is accessed using the address, and data registration /
Perform search processing.

(6): Description of Apparatus Example 6—See FIG. 6 Apparatus Example 6 is an apparatus example in which the standard key generator 12 of Apparatus Example 5 and Apparatus Example 4 are combined. This device includes a standard key generation unit 12, a virtual key generation unit 8, an address generation unit 3,
A control unit (not shown) and a table 7 are provided. The table 7 is provided with keys and corresponding standard keys, virtual keys, records, and two-branch links, so that data can be registered in these.

In the device example 6, the standard key generator 12 generates a standard key from a given key, the virtual key generator 8 generates a virtual key from the standard key, and the address generator 3 generates the virtual key from the standard key. Generates an address from. Then, the table is accessed using the address, and data registration /
Perform search processing.

When a collision occurs in the above processing, the address recalculation unit 4 recalculates the address from the virtual key, generates a new address, and accesses the table 7. Have been.

§2: Description of Processing of Apparatus Example 1 See FIG. 7 FIG. 7 is a processing flowchart of Apparatus Example 1. Hereinafter, an example of the registration process of the device example 1 will be described with reference to FIG. In addition, S51 to S54 indicate each processing step. The given key is k, the virtual key is k ₁ , and the address is a.

In the apparatus example 1, an entry made up of a given key and a given record is accessed through a two-step operation as follows. First, in the first stage,
The virtual key generator 8 receives the key k (S5).
1), based on the given key k, a hash function h
To generate a virtual key k ₁ by the (k) (S52). Next, in the second stage, the address generating unit 3 uses the obtained virtual key k ₁ to generate an address a reflecting the size of the actual table.
Is generated {a ← ad (k ₁ )} (S53). In this case, the function ad (k ₁ ) is the same kind of function as the hash function h, and based on the virtual key k ₁ , the function ad (k ₁ )
_An address indicated by the variable a obtained by ₁ ) (this is described as "address a") is generated.

When the address a is generated in this way, table processing (data registration / retrieval processing to the corresponding entry) is performed using the address a (S54). in this way,
Conventionally, a single hash device divides an address calculation process performed at one time into two stages, and each device (the virtual key generation unit 8 and the address generation unit 3) uses a conversion method suitable for the purpose. be able to. Virtual key generating unit 8, thereby reducing the collision by generating the unpolarized (small) virtual key k ₁ for the key k appearing with bias. Note that the size of the virtual key space may be infinite in principle or a sufficiently large finite number.

§3: Description of processing of apparatus example 2—see FIG. 8 FIG. 8 is a processing flowchart of apparatus example 2. Hereinafter, an example of the registration process of the device example 2 will be described with reference to FIG. In addition, S61 to S67 indicate each processing step. The given key is k, the virtual key is k ₁ , and the address is a.

In the device example 2, similarly to the device example 1, access to an entry consisting of a given key and a record is performed through a two-step operation as follows. First, in a first step, the virtual key generating unit 8, the key k is given (S61), based on the given key k, and generates a virtual key k ₁ by the hash function h (S
62). Next, in the second stage, the address generating unit 3 generates {a ← ad (k ₁ )} based on the obtained virtual key k ₁ , reflecting the size of the actual table (S 6).
3).

Thereafter, the control unit determines whether or not there is a free space in the table of the address a (S64). If there is a free space, the data (key k + record data) is registered in the entry (S66). ), And terminate the process. But,
If there is no space in the table in the process of S64, the control unit determines whether the given key k is the same as a registered key {key (a) = k? ｝ Is determined (S65).

As a result, if they are the same, the process is terminated, but if they are not the same (primary collision occurs), the address recalculating unit 4 uses the address a based on the calculated virtual key k _1.
{K ₁ ← next k ₁ , a ← ad (k ₁ )}
Is performed (S67). Then, the process is repeated from the process of S64. In the process of S67, the virtual key k
_When seeking ₁ , the virtual key k is calculated from the given key k again.
₁ is calculated {k ₁ ← next (k ₁ , k)}, and then
It is also possible to recalculate the addresses a, based on the virtual key k ₁ mentioned above calculation.

As described above, in the device example 2, the address a is calculated in the above-described processing, and if a primary collision occurs there, the address recalculating unit 4 calculates a new address and generates a new address. Let it. When a secondary collision occurs even at a new address, the address recalculation unit 4 generates a new address until a free entry is found.

The feature of this apparatus is that the address is recalculated (S67).
), The virtual key k₁The point is to use. This
In the recalculation processing of the address of the virtual key k₁Calculated with
It is also possible to use the address a₁← next
(K₁, K)｝. Virtual key k ₁Appear uniformly, so
By using this, even when recalculating, a new
The address is easily calculated, resulting in a secondary collision
Is expected to decrease.

In the conventional hashing device,
Since the new address is calculated based on the address directly calculated from the key, there is a high possibility that a secondary collision will occur again even if recalculation is performed. Even if recalculation is performed in consideration of keys and addresses similar to those of the apparatus of the present invention, there is an increase in the occurrence of secondary collision due to uneven appearance of keys. The device of the present invention reduces the occurrence of secondary collision by using uniformly appearing virtual keys instead of keys.

§4: Description of processing of apparatus example 3—see FIG. 9 FIG. 9 is a processing flowchart of apparatus example 3. Hereinafter, the processing of the device example 3 will be described with reference to FIG. Note that S71
S82 indicate each processing step.

Although there are various possible chaining methods in the apparatus example 3, the processing of the apparatus will be described using an example of a relatively simple and efficient chaining with a two-branch tree structure. In this apparatus, the calculation of the address is performed in the same manner as in the processing of the apparatus example 1. Then, when a primary collision occurs, unlike the processing of the device example 2, a tree structure is generated instead of calculating a new address.

Therefore, the entry of the hashing device is provided with a two-branch link. One is a link of an entry for registering a record having a key larger than that of the entry, and the other is a link to an entry of a smaller key. The feature of this device is that the comparison of the keys is not based on the conventional comparison of the original keys, but rather by using a virtual key generated from the keys.

In this case, only when the virtual keys match, the whole comparison is performed by comparing the magnitudes of the keys. Since the bias of the appearance of the virtual key is eliminated, the appearance ratio of large and small is almost １ ／ in comparison of the virtual keys as compared with the case of comparing two keys. As a result, it is expected that the created tree structure is close to an ideal state of a well-balanced tree, and that the depth of the tree with respect to the number of entries becomes almost the minimum value. As a result, even in the chaining method, the number of tree structure searches for registration and search is reduced, which contributes to an improvement in processing efficiency.

In the process of the device example 3, access to an entry consisting of a given one key and a record is performed through a two-stage operation as follows. First, in the first stage,
The virtual key generator 8 receives the key k (S7).
1), based on the given key k, it generates a virtual key k ₁ by the hash function _{h {k 1 ← h (k} )} to (S7
2). Next, in the second stage, the address generation unit 3 generates {a ← ad (k ₁ )} based on the obtained virtual key k ₁ and reflects the size of the actual table (S 73). .

Next, the control unit 5 determines that the entry of the address a is empty. Is $ table (a) empty? ｝ Is determined (S
74) If it is vacant, the key k and the record corresponding to it are registered in the entry of the address a (S76), and the registration work ends. However, when the entry of the address a is not empty, the control unit 5 sets the key {key} registered in the address a.
(A) Check if {} matches the given key k, or {key
(A) = k? The determination is made by comparing｝ (S75).

As a result, if they match, the key is already registered and the process ends. However, when the keys do not match, that is, in the entry of the address a to be registered,
If a record with another key is already registered,
The control unit compares the virtual key k ₁ with the key {k ₁ : key (a)} already registered in the table 7 using some kind of comparison mechanism (S77).

As a result, when the virtual key k ₁ is larger, {k ₁ > key (a)} follows the link for the larger key. When the virtual key k ₁ is smaller, ｛k ₁ <k
ey (a)} follows the link for the small key.

At this time, when tracing a link for a large key, is there a link available to the larger one of table 7 or is greater-link (a) empty? It is determined whether or not the link for the smaller key is to be traced (S81).
-Is link (a) empty? It is determined whether or not｝ (S78).

As a result, when it is determined that the link is empty in the processing of S81 and S78, it means that the registration with the key has not been performed, so that the record of the key k to be registered at present is Register That is, a new entry is created and registered (S80).

If there is an entry already registered in the link for tracing (S81, S7
8) Follow the smaller link or the larger link (S82, S79). By repeating the process from S75 again, the key k to be registered and the record corresponding thereto are registered at an appropriate place in the tree structure.

§5: Description of processing of apparatus example 4—see FIG. 10 FIG. 10 is a processing flowchart of apparatus example 4. Less than,
The processing of the device example 4 will be described based on FIG. Note that S
Reference numerals 91 to S104 denote processing steps. Apparatus example 4
This is a device that combines the features of the open address hash device shown in device example 2 and the features of the chain hash device shown in device example 3.

In the processing of the device example 4, when a primary collision occurs, a new address is calculated up to a certain number of times by using the recalculation mechanism (address recalculation unit 4) which is a feature of the device example 2. By searching for empty entries. If it exceeds this, an entry is secured using the link mechanism which is a feature of the device example 3 and registered there.

In this case, the number of times to shift to the link from the recalculation is as follows: When a certain fixed number is determined in advance: When determined by calculation using keys, virtual keys, and addresses,
: Various methods are possible, including shifting to when a specific condition is satisfied. Specifically, it is as follows.

In the device example 4 as well, access to an entry consisting of a given key and a given record is as follows.
It is performed through the operation of steps. First, in a first step, the virtual key generating unit 8, the key k is given (S91), based on the given key k, and generates a virtual key k ₁ by the hash function h (S92). Next, in a second stage, the address generation unit 3 generates an address a reflecting the actual size of the table from the obtained virtual key k ₁ ｛a ← ad (k
₁ ) Perform the operation (S93).

Thereafter, the control section determines whether or not there is a free space in the table of the address a (S94).
The data (key k + record) is registered in the entry (S96), and the process ends. However, if there is no free space in the table in the process of S94, the control unit determines whether the given key k is the same as the registered key or not.
(A) = k? ｝ Is determined (S95).

As a result, if they are the same, the process is terminated, but if they are not the same, the control unit determines whether or not to link (S97). As a result, when it is determined that the link should not be made, the address recalculation unit 4 recalculates the address a based on the calculated virtual key k ₁ ｛k ₁ ← next
_{k 1, a ← ad (k} 1) carry out} (S98). Then, the process is repeated from the process of S94.

In the process of S97, the virtual key k
_When seeking ₁ , the virtual key k is calculated from the given key k again.
₁ is calculated {k ₁ ← next (k ₁ , k)}, and then
It is also possible to recalculate the addresses a, based on the virtual key k ₁ mentioned above calculation.

On the other hand, if it is determined in the processing of S97 that a link should be made, the control unit 5 determines the virtual key k ₁ and the key #k ₁ : key (a) already registered in the table 7. ｝ Are compared using some kind of comparison mechanism (S99). As a result, when virtual key k ₁ is larger, {k ₁ > key (a)} follows the link for the larger key. When the virtual key k ₁ is smaller, ｛k ₁
<Key (a)} follows the link for the small key.

At this time, when tracing the link for the large key, is there a link available to the larger one of the table 7? Is greater-link (a) empty? It is determined (S103) whether or not the link for the small key is to be traced.
Is s-link (a) empty? It is determined whether or not｝ (S100).

As a result, when it is determined that the link is empty in the processing of S103 and S100, it means that the registration with the key has not been performed, so that the record of the key k to be registered at present is recorded. Register That is, a new entry is created and registered (S102).

If there is an entry already registered in the link for tracing (S103, S10
0), follow the smaller link or the larger link (S101, S104). And again, the S9
By repeating the process of No. 9, the key k to be registered and
The corresponding record is registered at an appropriate place in the tree structure.

§6: Description of processing of apparatus example 5—see FIG. 11 FIG. 11 is a processing flowchart of apparatus example 5. Less than,
The processing of the device example 5 will be described based on FIG. In addition,
S111 to S115 indicate each processing step. In the processing of the device example 5, a method of using an associative storage device using a key having a certain structure is described in a Prolog language (Program Logic).
Language) and theorem search will be described as examples. Prolog
The basic structure of data in a language is a term. The term is
For example, 'f (a)', 'g (a, b)', 'h (a,
g (b, a)) ′. Variables can be used in the expression.

For example, 'f (X)', 'h (X, g
(B, X)) 'is a variable. Arbitrary terms can be assigned to variables. For example, 'f
Substituting 'g (a, Y)' for the variable 'X' of (X) 'yields the term' f (g, (a, Y)) '. When an assignment is made, the same expression is assigned to the same variable.

For example, when 'a' is substituted for a variable 'X' of 'h (X, g (b, X))', 'h (a, g (b,
a)) ′ is obtained. Thus, the variable serves as a place for the assignment, and it does not matter what name is used as the variable name. For example, 'f (X)'
And 'f (Y)' and 'h (X, g (b, X))'
And 'h (Z, g (b, Z))' have different appearances but have the same function. In addition, a,
Lowercase letters such as b and c are constants, and uppercase letters such as X and Y are variables.

By the way, a logical formula proved to be correct is called a theorem. Once proved, theorem is expected to be stored in a database and used to prove other theorems. In such a situation, when a search is performed to determine whether a certain logical expression expected to be a theorem is stored in the theorem database, the appearance differs as in the case of using the same variables as in the above-described Prolog example. However, if the contents are substantially the same, we would like to treat them as the same rather than treating them differently.

As one means for determining that expressions having different appearances are substantially the same, there is a method of converting the expressions into a standard form and comparing them. For example, '(a∧b) ∨c'becomes' (a∨
c) By converting to ∧ (b∨c) ',
・ A standard expression of ∧ ... 'is obtained (∧: AN
D, Δ: OR).

In this processing, the key is converted to a standard form without directly using the original key, and the converted key is used as a key for registration / retrieval. First, when a key k is given (S111), the standard key generation unit 12 generates a standard key s by a hash function h based on the key k (S1).
12).

Next, the virtual key generator 8 sets the standard key s
Based on, to generate a virtual key k ₁ by the hash function h (S113). Thereafter, the address generation unit 3, based on the virtual key k _1, to generate an address a by the function ad (S11
4). Then, the control unit performs the processing of the table 7 (registration / search processing for the entry) based on the generated address a (S115).

As described above, instead of directly using the original key, the key is converted to a standard form and is used as a key for registration / search. In this way, by using the standard key, not only the search by the original key but also the search by the standard key is possible, and a kind of semantic search becomes possible. Also,
Collisions can be reduced and processing can be speeded up. Further, when the key has a certain structure, by using a virtual key reflecting the structure, not only the access speed of the associative memory device can be increased but also the access flexibility can be improved.

§7: Description of processing of apparatus example 6—see FIG. 12 FIG. 12 is a processing flowchart of apparatus example 6. Less than,
The process of the device example 6 will be described based on FIG. In addition,
S121 to S135 indicate each processing step. In this process, first, when a key k is given (S121), the standard key generation unit 12 uses the hash function s based on the key k.
A standard key s is generated using k (S122).

Next, the virtual key generator 8 sets the standard key s
Based on, it generates a virtual key k ₁ by the hash function h (S123). Thereafter, the address generation unit 3, based on the virtual key k _1, to generate an address a by the hash function ad (S124). Thereafter, the control unit determines whether or not there is a free space in the table at the address a (S125). If there is a free space, the data (key k + record data) is registered in the entry (S127), and the process is performed. finish. In addition,
The standard key s may also be registered depending on the use.

However, if there is no space in the table in the process of S125, the control unit determines whether or not the given key k is the same as the registered key {key (a) = k? ｝ Is determined (S126). As a result, if they are the same, the process ends, but if they are not the same, the control unit determines whether or not to link (S128).

As a result, if it is determined that the link should not be made, the address recalculating section 4 recalculates the address a based on the calculated virtual key k ₁ ｛k ₁ ← next
k ₁ , a ← ad (k ₁ )} is performed (S129). Then, the process is repeated from the process of S125. Note that the S
In process 129, when obtaining the virtual key k _1, to calculate the virtual key k ₁ from the key k given again {k ₁ ←
Next (k ₁ , k)}, and then the address a can be recalculated based on the calculated virtual key k ₁ .

On the other hand, if it is determined in step S128 that a link should be made, the control unit 5 determines that the virtual key k ₁ and the key #k already registered in the table 7
₁ : key (a)} is compared using some kind of comparison mechanism (S130). As a result, when virtual key k ₁ is larger, {k ₁ > key (a)} follows the link for the larger key. Also, when the virtual key k ₁ is smaller, ｛k ₁ <
key (a)} follows the link for the small key.

At this time, when tracing the link for the large key, is the link connecting to the larger one of the table 7 empty? Is greater-link (a) empty? It is determined whether or not there is a link (S134), and if the link for the small key is to be followed, it is determined whether or not the link connecting to the smaller key is free.
Is s-link (a) empty? It is determined whether or not｝ (S131).

As a result, when it is determined that the link is empty in the processing of S131 and S134, it means that the registration with the key is not performed, and the record of the key k which is to be registered at present is recorded. Register That is, a new entry is created and registered (S133).

If there is an entry already registered in the link for tracing (S131, S13
4) Follow the smaller link or the larger link (S132, S135). And again, the S1
By repeating the process of step 30, the key k to be registered
And the corresponding record is registered at an appropriate place in the tree structure.

§8: Description of Experimental Example 1—See FIG. 13 FIG. 13 is an explanatory diagram of Experimental Example 1. Experimental Example 1 is an example in which an experiment was performed using the open address hash device of Device Example 2 shown in FIG. In this experiment, an experiment was performed using the data shown in the figure, and an experiment result was obtained.

As the experimental data, the key k is a,
aa, ab, abc, b, bc, cd, cc, bb,
19 keys (character strings) consisting of d, e,... bcb, and the size of the table (number of addresses) was 25. In this case, the number of primary collisions was nine. However, the number of secondary collisions was 18, which was significantly reduced as compared with the conventional example (the number of secondary collisions in the conventional example was 78). As a result, it was proved that the number of collisions was reduced and that the processing could be speeded up.

§5: Description of Experimental Example 2—See FIG. 14 FIG. 14 is an explanatory diagram of Experimental Example 2. Experimental Example 2 is an example in which an experiment was performed using the chain method hash device of Device Example 3 shown in FIG. In this experiment, the same data as in Experimental Example 1 was used, and an experimental result was obtained.

That is, as the experimental data, the key k is a, aa, ab, ab
c, b, bc, cd, cc, bb, d, e,.
19 keys (character strings) consisting of cb, and the size (number of addresses) of the table was 25. In this case, the number of primary collisions was nine, but the depth of the two-branch link was shallower than that of the conventional example as shown in the figure (the number of two-branch was smaller). As a result, it was proved that the number of collisions was reduced and that the processing could be speeded up.

[0114]

As described above, the present invention has the following effects. (1): A high-speed associative memory device (hash device) close to an ideal state can be achieved by effectively using a virtual key to reduce data deviation and reduce collisions (especially secondary collisions). Can be realized. Further, when the key has a certain structure, by using a virtual key reflecting the structure, not only the access speed of the associative storage device can be increased but also the access flexibility can be improved.

(2): By using a virtual key, the number of collisions in a hash device (a device using a hashing technique), which is a type of associative storage device, can be reduced. Therefore, it is possible to achieve high-speed data registration and search processing for each entry.

(3) By using the standard key, not only the search by the original key but also the search by the standard key is possible, and a kind of semantic search becomes possible. In addition to the above effects, the following effects are provided corresponding to each claim.

(4): In the first aspect, the virtual key generation unit generates a virtual key by a hash function based on the given key in the first stage, and the address generation unit generates the virtual key from the virtual key in the second stage. Then, an address reflecting the size of the actual table is generated. As described above, conventionally, the address calculation processing performed once by one associative storage device is divided into two stages,
The virtual key generator and the address generator can use a conversion method suitable for the purpose. The virtual key generation unit can reduce collision by generating a virtual key with a small bias for keys that appear with a bias.

(5): In claim 2, the address recalculating section recalculates the address based on the virtual key when the address is generated by the address generating section and a collision occurs. In this case, since the virtual key appears uniformly in the recalculation of the address, the use of the virtual key makes it possible to easily calculate a new address with less bias even in the recalculation. Collisions can be reduced.

(6): In the third aspect, the tree structure generation control means generates a tree structure using a link by comparing the magnitude of virtual keys when an address is generated by the address generation unit and a collision occurs. I do. In this case, only when the virtual keys match, the whole comparison is performed by comparing the magnitudes of the keys. Since the bias of the appearance of the virtual keys is eliminated, for example, if the link is a two-branch tree, the appearance ratio of the large and small is almost 1 in the comparison of the virtual keys as compared with the case of comparing the two keys.
/ 2.

As a result, it is expected that the created tree structure is close to a well-balanced tree which is an ideal state, and that the depth of the tree with respect to the number of entries becomes substantially the minimum value. As a result, the number of searches for the tree structure for registration and search for the entry is reduced, which contributes to an improvement in processing efficiency.

(7): According to claim 4, the standard key generation section comprises:
When a key is given, a standard key is generated using a hash function based on the given key.
Generate a virtual key using a hash function based on a standard key. Thereafter, the address generation unit generates an address based on the virtual key.

As described above, instead of directly using the original key, the key is converted to a standard form and is used as a key for registration / search. In this way, by using the standard key, not only the search by the original key but also the search by the standard key is possible, and a kind of semantic search becomes possible. Also,
Collisions can be reduced and processing can be speeded up. Further, when the key has a certain structure, by using a virtual key reflecting the structure, not only the access speed of the associative memory device can be increased but also the access flexibility can be improved.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating the principle of the present invention.

FIG. 2 is an explanatory diagram of device examples 1 and 2 in the embodiment.

FIG. 3 is an explanatory diagram of an apparatus example 3 in the embodiment.

FIG. 4 is an explanatory diagram of an apparatus example 4 in the embodiment.

FIG. 5 is an explanatory diagram of an apparatus example 5 in the embodiment.

FIG. 6 is an explanatory diagram of an apparatus example 6 in the embodiment.

FIG. 7 is a processing flowchart of an apparatus example 1 in the embodiment.

FIG. 8 is a processing flowchart of an apparatus example 2 in the embodiment.

FIG. 9 is a processing flowchart of an apparatus example 3 in the embodiment.

FIG. 10 is a processing flowchart of an apparatus example 4 in the embodiment.

FIG. 11 is a processing flowchart of an apparatus example 5 in the embodiment.

FIG. 12 is a processing flowchart of an example 6 of the apparatus according to the embodiment.

FIG. 13 is an explanatory diagram of Experimental Example 1 in the embodiment.

FIG. 14 is an explanatory diagram of Experimental Example 2 in the embodiment.

FIG. 15 is an explanatory diagram 1 of a conventional example.

FIG. 16 is an explanatory diagram 2 of the conventional example.

FIG. 17 is a processing flowchart of a conventional open address hashing device.

FIG. 18 is a processing flowchart of a conventional chain hash device.

FIG. 19 is an explanatory diagram of Conventional Experimental Example 1.

FIG. 20 is an explanatory diagram of Conventional Experimental Example 2.

[Explanation of symbols]

 3 address generator 4 address recalculator 7 table 8 virtual key generator 12 standard key generator

Claims (4)

[Claims] 1. A virtual key is generated by using a hash function based on a table having a plurality of entries each comprising a set of one key and a record corresponding to the key, and assigning an address to each entry, and a given key. A virtual key generation unit for generating an address of the table based on a virtual key generated by the virtual key generation unit, and an address generated by using a hashing method to each entry. An associative storage device characterized in that access is enabled. 2. An address recalculating section for recalculating an address based on a virtual key generated by the virtual key generating section when an address is generated by the address generating section and a collision occurs. 2. The associative memory device according to claim 1, wherein: 3. A multi-branch link is provided in each entry of the table, and when a collision occurs by generating an address in the address generating section, a tree is generated by using the link by comparing the virtual keys in magnitude. 2. The associative storage device according to claim 1, further comprising tree structure generation control means for generating a structure. 4. A table having a plurality of entries each consisting of one key and a set of a standard key and a record corresponding to the key, a table in which addresses are assigned to each entry, and a standard using a hash function based on a given key. A standard key generator for generating a key, a virtual key generator for generating a virtual key using a hash function based on the standard key generated by the standard key generator, and a virtual key generator for generating the virtual key. An associative storage device, comprising: an address generation unit that generates an address of the table based on a virtual key, wherein each of the entries can be accessed by an address generated by using a hashing method.