KR102102307B1 - Method for searching storage device in database management system with multiple storage devices - Google Patents

Abstract

The present invention relates to a technology which, based on a vector, can more quickly search for storage in which search target data is stored by using a bloom filter in a database management system storing data in multiple storage. According to the present invention, a method for searing for the storage based on vectorization by using the bloom filter in the data management system having the multiple storage comprises: a first step of sequentially assigning N index keys corresponding to requested data to a hash vector; a second step of calculating a hash function result value for each index key and inserting the same into an assigned hash vector cell; a third step of generating test vectors for each index key, respectively; a fourth step of generating a bit vector; a fifth step of updating the hash vector; and a sixth step of determining that the data is stored in corresponding storage.

Description

METHOD FOR SEARCHING STORAGE DEVICE IN DATABASE MANAGEMENT SYSTEM WITH MULTIPLE STORAGE DEVICES in a database management system with multiple storages using bloom filters

According to the present invention, a vector based on a storage in which data to be searched is stored using a bloom filter in a database management system that stores data in multiple storages such as DRAM, NVM, or DISK can be more quickly searched based on vectors. It's about technology.

A database management system (DBMS) is a set of software tools that allow multiple users to access data in a database. More specifically, a DBMS is implemented in a database server to systematically handle the needs of a large number of users or programs and respond appropriately to use data.

Meanwhile, when a specific query is input from the outside, the DBMS performs functions such as selecting, inserting, updating, and deleting data in the database according to the entered query. Here, the query means a description of what request for data stored in a table in a database, that is, what operation to perform on the data, is expressed by using a language such as Structured Query Language (SQL).

In addition, as the amount of data increases, the DBMS generally performs data management using an index. Here, the index means a data structure that speeds up a search for a table in the database field.

Meanwhile, recently, hybrid DBMSs having storages having different performances have been proposed and operated. Storage provided in the hybrid DBMS includes DRAM, NVM, and DISK.

DRAM is fast to read and write, but it is still relatively expensive despite a significant price drop.

NVM writes are as fast as DRAM, but read performance is poor compared to write. Also, NVM is more expensive than DISK.

DISK provides low cost and large storage space, but has the lowest read and write performance.

Accordingly, the hybrid DBMS operates storage by storing frequently accessed HOT data in DRAM, normally accessed WARM data in NVM, and longest inaccessible COLD data in DISK. That is, data for one table may be stored in different storages according to its characteristics.

Because of these characteristics, determining which storage the data is stored in when retrieving data has a significant impact on the data retrieval performance of the hybrid DBMS.

Currently, a bloom filter is used as a method for determining storage in which data is stored.

The bit check of the bloom filter calculates all the results for each hash function by applying multiple hash functions to the index key corresponding to the data to be searched, and based on the results of each hash function, the bit value of the bloom filter, that is, the bloom filter The storage search is performed through the process of checking the check value.

That is, the conventional method of determining storage in which data is stored using a bloom filter is a hash count operation that calculates all hash function result values for one index key, and then corresponds to all hash function result values in the bloom filter. This is done by checking the bit value.

However, when a plurality of index keys need to be checked, a problem occurs in data retrieval performance of the hybrid DBMS because the other index keys must be in a standby state while performing a bloom filter check on one index key.

1. Korean Registered Patent No. 1775107 (Name: Hybrid database and how to manage tables in hybrid database)

Accordingly, the present invention was created in view of the above-described circumstances, and a SIMD collection command is applied to a Bloom Filter in a database management system that stores data in a plurality of storages. The technical object of the present invention is to provide a vectorization-based storage search method using a bloom filter in a database management system having a plurality of storages, which enables a faster search for storage in which search target data is stored by acquiring a value.

According to an aspect of the present invention for achieving the above object, data and index key information for the corresponding data are stored, and a plurality of storages equipped with a bloom filter corresponding to the index keys are provided, and storage where query requested data is stored In a storage search method using a bloom filter in a database management system having a plurality of storages having data processing means for searching and performing the corresponding query processing, the data requested to query the hash vector with N cells in the data processing means The first step of sequentially assigning N index keys corresponding to the second step, calculating a hash function result value for each index key, and inserting it into the assigned hash vector cell, based on the hash function result value of each index key in the bloom filter Index consisting of a preset number of bloom filter check values collected randomly A third step of generating a test vector for each key, a fourth step of determining a target bit corresponding to a hash function result value in the test vector for each index key, and generating a bit vector consisting of bloom filter check values of the target bit for each index key. If there is a target bit having an invalid value in the bit vector, remove the corresponding index key from the hash vector, and update the hash vector by inserting the index key of the difference rank into the cell assigned to the index key removed from the hash vector. Step 5 and the operation after the second step of calculating the result of the hash function for the index key assigned to the hash vector and inserting it into the assigned hash vector cell are repeatedly performed, but a predetermined hash function for each index key Sequentially to calculate the result value, and based on the index key where the bloom filter inspection for all the hash function result values has been completed. A vectorization-based storage search method using a bloom filter is provided in a database management system having a plurality of storages, characterized in that it comprises a sixth step of determining that the corresponding data is stored in the corresponding storage.

In addition, in the second step, the data processing means generates a hash function control vector consisting of a hash function sequence that calculates a hash function result value for each index key, and in the sixth step, the data processing means for each index key of the hash function control vector. If the hash function sequence number and the predetermined number of hash functions for the index key are the same, the hash vector is removed by removing the corresponding index key from the hash vector and updating the hash vector by inserting the index key of the second rank into the removed index key. A vectorization-based storage search method using a bloom filter in a database management system having a storage of is provided.

In addition, in the third step, the data processing means generates an offset vector having the same value as the hash function result value for each index key, and checks the bloom filter of the bit corresponding to a multiple of the offset value in the bloom filter based on the offset value for each index key. Database management with a plurality of storages, characterized in that by extracting a predetermined number of values to generate a test vector, in the fourth step, the data processing means determines the first bit as the target bit of the corresponding index key in the test vector for each index key. A system-based storage search method using bloom filters is provided.

In addition, in the fifth or sixth step, when the index key is removed from the hash vector, the data processing means determines whether the number of index keys to be queued is equal to or greater than the number of index keys removed from the hash vector, and If the number of keys is less than the number of index keys removed from the hash vector, a number of features characterized by performing a conventional bloom filter inspection operation to extract the bloom filter check value by applying the hash function result value to each bloom filter for each remaining index key A vectorization-based storage search method using a bloom filter in a database management system having a storage of is provided.

In addition, in the fifth step, the data processing means generates a mask set by ANDing a zero vector and a bit vector consisting of a bloom filter invalid value, and the target bit having the same value as the zero vector in the mask set has an invalid value. A vectorization-based storage search method using a bloom filter is provided in a database management system having multiple storages characterized by determining.

In addition, the data processing means has a plurality of storage, characterized in that each of the storage using a single instruction multiple data (SIMD) collection command that simultaneously calculates a plurality of values for a single command to perform a bloom filter inspection process for each storage. A vectorization-based storage search method using a bloom filter in a database management system is provided.

According to the present invention, in addition to performing a bloom filter check on a plurality of index keys at the same time using the SIMD method, when there is a hash function result value that has an invalid value based on a hash function result, the index key for the corresponding index key is performed. By stopping the bloom filter inspection process, it is possible to determine whether or not the data is stored in the storage more quickly by using the bloom filter.

1 is a view showing a schematic configuration of a database management system having a plurality of storage to which the present invention is applied.
FIG. 2 is a conceptual diagram illustrating a bloom filter inspection process using a SIMD method performed by the data processing means 100 shown in FIG. 1.
3 is a flowchart illustrating a storage search method using a bloom filter in a database management system having multiple storages according to the present invention.
FIG. 4 is a flowchart for explaining a process (ST300) of simultaneously performing a bloom filter test shown in FIG. 3.
FIG. 5 conceptually illustrates the bloom filter inspection process of FIG. 4.

The configurations shown in the embodiments and drawings described in the present invention are only preferred embodiments of the present invention, and do not represent all the technical spirit of the present invention, so the scope of the present invention is the embodiments and drawings described in the text It should not be construed as limited by. That is, since the embodiments can be variously changed and have various forms, it should be understood that the scope of the present invention includes equivalents capable of realizing technical ideas. In addition, the purpose or effect presented in the present invention does not mean that a specific embodiment should include all of them or only such an effect, and the scope of the present invention should not be understood as being limited thereby.

All terms used herein have the same meaning as commonly understood by a person skilled in the art to which the present invention pertains, unless otherwise defined. The terms defined in the commonly used dictionary should be interpreted as being consistent with meanings in the context of related technologies, and cannot be interpreted as having ideal or excessively formal meanings that are not explicitly defined in the present invention.

1 is a view showing a schematic configuration of a database management system having a plurality of storage to which the present invention is applied.

Referring to FIG. 1, a database management system having a plurality of storages to which the present invention is applied includes a data processing means 100 and a plurality of storages 200.

At this time, the storage 200 may be composed of different heterogeneous storage, including dynamic random access memory (DRAM) 210, non-volatile memory (NVM) 220, and DISK 230, and a plurality of the same It can also consist of dog storage.

Here, the DRAM 210 stores the HOT data having an access point within a certain period from the present, including the initial data insertion, and the NVM 220 stores WARM data whose access point has passed a certain period from the present, DISK 230 is configured to store COLD data whose access point exceeds a certain period from the present. Of course, if another type of storage is provided, data satisfying a preset access point may be moved to and stored in the corresponding storage.

In addition, each of the storage 200 is basically composed of an index data storage 201 and a bloom filter (Bloom filter, 202), the index object storage 203 is stored in the DRAM 210 where the first inserted data is stored Is additionally provided.

The index data storage 201 stores actual index data for data stored in each storage 200, that is, an index key.

The index object store 203 stores index objects consisting of storage information including the number of storage devices currently in use and index points for each storage.

Meanwhile, in FIG. 1, the data processing means 100 analyzes an input query and performs transaction processing such as inserting, deleting, modifying, and retrieving the requested data in response to the query, and querying the result of the transaction processing Returns as

At this time, the data processing means 100 preferentially searches the storage in which the data is stored for queries related to pre-stored data, such as insertion, deletion, modification, and retrieval of data. Then, the data processing means 100 performs a series of transaction processing on the requested data in the storage 200, where it is determined that the data is stored based on the search result.

The data processing means 100 checks whether or not data is stored in the storage 200 by using the bloom filter 202 provided in each storage 200. At this time, the data processing means 100 simultaneously acquires the bloom filter inspection results for a plurality of index keys using the SIMD method. Here, SIMD (Single Instruction Multiple Data) is a type of parallel processor that calculates multiple values at the same time with one instruction. CDC Star-100 operating on a vector of data with one instruction, Texas Instruments ASC, etc. It was first used in vector supercomputers in the early 1970s. Such SIMDs are frequently used in vector processors, especially in multimedia fields such as video game consoles and graphics cards.

FIG. 2 is a diagram conceptualizing a bloom filter inspection process using a SIMD method performed by the data processing means 100 shown in FIG. 1.

Referring to FIG. 2, a plurality of index keys (A, B, C, and D) are set in a hash vector 10 of a predetermined size having a plurality of cells to correspond to each cell S, and the hash vector 10 is set. Bloom filter check value (20: a [A], a [B], a [C]) of each index key in the bloom filter 202 based on the result of the hash function corresponding to the index key set in each cell (S) of the , a [D]) at the same time.

Here, the bloom filter 202 is a probabilistic data structure used to test whether an element is a member of a set, and is represented by a bit stream having an m bit size. In addition, the bloom filter 202 is an effective value in which bits corresponding to a hash value of an element derived by k different hash functions with respect to each element among bit streams of m bits are represented in black, For example, it may be set to "1" and the remaining bits may be set to an invalid value, for example, "0".

In this case, the data processing means 100 is a bit stream required by the SIMD instruction for each index key in order to apply the vector-based SIMD instruction composed of a plurality of bits, and more specifically, the SIMD collection instruction to the bloom filter 202. The resulting hash function vector is generated, and the bloom filter check value corresponding to the desired hash function result value is extracted from the bloom filter using the result.

Then, the data processing means 100, if the bloom filter check value 20 corresponding to a plurality of hash function result values preset for each index key are all valid values (eg, "1"), the corresponding index key It is determined that the corresponding data is stored in the corresponding storage.

In particular, the data processing means 100 sequentially performs the process of simultaneously extracting the bloom filter check value 20 for each hash function for a plurality of index key sets in the unit of hash function. At this time, the data processing means 100, if the index key of the bloom filter check value invalid value (eg, "0") of the index key of the set, the hash vector 10, the index key corresponding to the invalid value Then, the index key of the next order is assigned to the cell S in which the index key is removed from the hash vector 10.

That is, for the index key having an invalid value as the result of the bloom filter, the result of the bloom filter for all the hash functions previously set for the corresponding index key is not checked, and the bloom filter is inspected only up to the hash function result value where the invalid value exists for the first time. Perform processing and stop.

Therefore, in the present invention, in addition to performing a bloom filter check on a plurality of index keys at the same time using the SIMD method, when a hash function result value with an invalid value exists based on a hash function result, the index key for the corresponding index key is performed. By stopping the bloom filter inspection process, it is possible to more quickly determine whether the corresponding data is stored in the storage using the bloom filter.

Next, a vectorization-based storage search method using a bloom filter in a database management system having multiple storages according to the present invention having the above-described configuration will be described with reference to FIG. 3.

Referring to FIG. 3, first, when a query requesting storage search is received from the outside (ST100), the data processing means 100 generates an index key for each query request data (ST200). At this time, the data processing means 100 generates an index key corresponding to the query request data according to the index key generation algorithm stored in the index data storage 201 of the storage 200.

The data processing means 100 applies the SIMD method to the bloom filters 202 for each storage 210, 220, and 230, and simultaneously collects each bloom filter check value corresponding to a plurality of index keys (ST300). At this time, the data processing means 100 converts information on the index key into a vector structure to apply the SIMD collection command to the bloom filter 202, and applies it to the bloom filter 202 of each storage 210 to index multiple data. The bloom filter check values for the keys are collected at the same time.

Then, the data processing means 100 retrieves query request data corresponding to the index key in which the bloom filter inspection is completed in the corresponding storage (ST400).

That is, the data processing means 100 corresponds to the case where the bit values of the bloom filter 202 corresponding to the result values of all hash functions for a plurality of hash functions preset for the index key are all valid values ("1"). It is determined that the data is stored in the corresponding storage.

In this case, the data processing means 100 sequentially performs a bloom filter check on the hash function result values for a plurality of index keys in a hash function unit, and the index key for which the hash function result value is an invalid value ("0") For the rest of the hash function, the bulroom filter inspection is stopped.

 Accordingly, the data processing means 100 is a query request data corresponding to the corresponding index key when the Bloom Filter check value for the result of the last hash function preset for the corresponding index key is a valid value ("1"). It can be judged as being stored in storage.

Next, the process of simultaneously performing the bloom filter inspection shown in FIG. 3 (ST300) will be described in more detail with reference to FIGS. 4 and 5. FIG. 4 is a flowchart for explaining a process (ST300) of simultaneously performing the bloom filter inspection shown in FIG. 3, and FIG. 5 is a diagram conceptually illustrating the bloom filter inspection process of FIG.

 In step ST200 of FIG. 3, in the state in which Q index keys are generated for the query request data, the data processing means 100 sets a hash vector size composed of a plurality of cells to perform a SIMD collection command, and for each cell The index key is allocated (ST311). 5, the hash vector size is composed of four cells (S), and each cell has an index key (V1, V2, V3, V4) assigned to each.

Subsequently, the data processing means 100 calculates a hash function result value for each index key and inserts it into the assigned hash vector cell (ST312). At this time, the type of the hash function to be performed for each index key and the number of hash functions are set in advance, and a hash function result value is calculated by sequentially using a plurality of hash functions set for the index key. In FIG. 5, “Count hash1” includes the first hash function result value (5,8,3,2) for the first hash function (hash1) for the index keys (V1, V2, V3, V4) to the corresponding index key. The state inserted in the assigned hash vector cell is illustrated.

In addition, the data processing means 100 generates a hash function control vector corresponding to the order of hash functions applied to each index key (ST313). At this time, the hash function control vector is set to the same size as the hash vector, and the corresponding hash function sequence information is inserted corresponding to the position to which the index key is assigned. For example, in FIG. 5, in a state in which the first hash function hash1 is applied to all of the index keys V1, V2, V3, and V4, the hash function control vector is {1,1,1 in which each cell is all "1". , 1}.

Subsequently, the data processing means 100 divides the number of bits corresponding to the hash function result value for each index key into the hash function result value to generate an offset vector (ST314). For example, when the result of the hash function consists of 32 bits, the first hash function result (5,8,3,2) of the index key (V1, V2, V3, V4) in FIG. The offset vector of {5,8,3,2}, which is the same as the hash function result, is generated.

In addition, the data processing means 100 generates a test vector for each index key consisting of the bloom filter check values collected using the offset vector in the corresponding bloom filter 202 for each index key (ST315). For example, when the offset vector for the index key V1 in FIG. 5 is “5”, and the result of the hash function is 32 bits, the bloom filter 202 has 32 bloom filter check values corresponding to multiples of 5 By extracting and arranging them sequentially, a test vector for the index key V1 is generated. That is, the bloom filter check value of the "5,10,15,20, ..." th bit in the bloom filter 202 becomes a test vector for the index key V1.

Also, the data processing means 100 determines a target bit for each index key in the test vector (S5316). At this time, the data processing means 100 sets the target bit by dividing the total number of bits of the test vector by the number of bits corresponding to the result value of the hash function. At this time, the total number of bits of the test vector is "32", which is the same as the number of bits "32" corresponding to the result of the hash function, and the first bit of the test vector is set as the target bit. And, the target bit is always determined as a bit corresponding to the hash function result value calculated for the corresponding index key in the bloom filter 202. As a result, the target bit corresponds to the result value of the hash function result in the bloom filter 202. It becomes a bit. That is, in FIG. 5, the target bit for the index key V1 becomes the "5" th bit of the bloom filter 202.

Thereafter, the data processing means 100 generates a bit vector consisting of a bloom filter check value of a target bit for each index key (ST317). That is, the bit vector of the index keys (V1, V2, V3, V4) as shown in "gather" in Figure 5 is the hash function result value (5) of each index key (V1, V2, V3, V4) in the bloom filter 202 , 8,3,2), and a bloom filter check value of bits corresponding to "1,1,0,1".

Next, the data processing means 100 determines whether there is a target bit having an invalid value in the bit vector (ST318). The data processing means 100 ANDs a bit vector and a zero vector consisting of only the invalid value "0" to extract a mask set having a value of "1" corresponding to "TRUE" when AND has the same value as the zero vector. Then, it is determined whether there is a target bit determined to be "TRUE" in this mask set. That is, as shown in "Compare" in FIG. 5, the result of the mask set may be represented as "o, o, x, o". Through this, the data processing means 100 is the target bit of the index key corresponding to the third cell of the hash vector. It can be judged that has an invalid value.

Then, when the target bit having an invalid value exists in step ST318, the data processing means 100 determines whether there are only a sufficient number of the difference index index keys, which are the waiting targets, to satisfy the hash vector (ST319). At this time, the data processing means 100 determines whether the number of waiting target index keys is greater than or equal to the number of target bits having an invalid value. In FIG. 5, since the data processing means 100 has one target bit having an invalid value, and five standby target index keys (V5, V6, V7, V8, V9), the target bit having an invalid value It is judged that the index key to be replaced is sufficient.

If the standby index key is sufficient in step ST319, the data processing means 100 removes the index key corresponding to the target bit having an invalid value from the hash vector, and removes the index index of the secondary rank index from the standby index key. The key position is inserted into a hash vector cell (ST320). As in "Exclude failed key" in FIG. 5, the index key V3 of the cell corresponding to the target bit having an invalid value in the hash vector is removed, and the difference from the standby target index key (Remained keys) as in "Put next key". The rank index key V5 is assigned to the cell assigned to the index key V3 in the hash vector. That is, "V1, V2, V5, V4" is updated to be assigned to each cell of the hash vector, such as "Count hash" in FIG.

Here, steps ST319 and ST320 may be performed by changing the order. That is, after removing the index key from the hash vector, the data processing means 100 may determine whether the number of index keys to be queued is greater than or equal to the number of index keys removed from the hash vector.

Meanwhile, if the target bit having an invalid value does not exist in step ST318, the data processing means 100 checks whether a hash function to be applied to each index key remains. Then, if a hash function remains, each index key is applied to the corresponding rank hash function to calculate a hash function result, and step ST312 is performed to insert it into the corresponding hash vector cell. At this time, in step ST313, the data processing means 100 updates the hash function control vector to correspond to the order of the next hash function applied to each index key. For example, if the previous hash function control vector is "1,1,1,1", and the hash function applied to the current index key is the next rank, the hash function control vector is updated to "2,2,2,2".

In addition, in step ST318, when there is no target bit having an invalid value and no hash function to be applied to the index key, it is determined that the query request data corresponding to the index key is stored in the corresponding storage. Step ST400 of step 3 is performed. At this time, when there is no hash function to be applied to some index keys in the current hash vector, the index key without the hash function is removed from the hash vector. Then, the data processing means 100 may determine whether the number of index keys to be queued is equal to or greater than the number of index keys removed from the hash vector.

On the other hand, after performing step ST319, the data processing means 100 checks whether there is a hash function to be applied to each index key of the currently updated hash vector, and if the hash function remains, corresponds to each index key. Applying to the difference rank hash function, the result of the hash function is calculated and inserted into the corresponding hash vector cell, step ST312 is performed. That is, in FIG. 5, the hash vector is "V1, V2, V5, V4" as in "Count hash". In the updated state as ", a step ST313 of checking a hash function sequence to be applied to each index key and updating a hash function control vector consisting of the hash function sequence are performed. At this time, in FIG. 5, the hash vector index key (V1, V2, V4) is applied to the difference hash function "hash2", the index key (V5) is applied to the first hash function "hash1", the hash function control vector Is updated to "2,2,1,2".

On the other hand, if the index key to be queued is not sufficient in step ST319, the data processing means 100 applies a hash function result value to each bloom filter for each remaining index key to extract a bloom filter check value. And performs (ST321). For example, since the size of the hash vector is set to "4", when the maximum remaining index key is "3" or less including the index key currently assigned to the hash vector, a conventional bloom filter inspection operation is performed on the remaining index key. .

100: data processing means, 200: storage,
201: index data storage, 202: bloom filter,
203: index object storage, 210: DRAM,
220: NVM, 230: DISK.

Claims (6)

In addition to storing data and index key information for the data, a plurality of storages equipped with bloom filters corresponding to the index keys are provided, and data processing means for searching the storage storing the requested data and performing the query processing In a storage search method using a bloom filter in a database management system having a large number of storage,
A first step of sequentially assigning N index keys corresponding to the requested data to the hash vector having N cells in the data processing means;
A second step of calculating a hash function result value for each index key and inserting it into the assigned hash vector cell,
A third step of generating a test vector for each index key consisting of a preset number of bloom filter check values randomly collected based on the result of the hash function for each index key in the bloom filter,
A fourth step of determining a target bit corresponding to a hash function result value in a test vector for each index key, and generating a bit vector consisting of bloom filter check values of the target bit for each index key.
If there is a target bit having an invalid value in the bit vector, remove the corresponding index key from the hash vector, and update the hash vector by inserting the index key of the difference rank into the cell assigned to the index key removed from the hash vector. Step 5 and,
The operation after the second step of calculating the result of the hash function for the index key assigned to the hash vector and inserting it into the assigned hash vector cell is repeatedly performed, but a predetermined hash function for each index key is sequentially applied. And calculating a result value, and comprising a sixth step of determining that the corresponding data is stored in the corresponding storage based on the index key where the bloom filter inspection for all the hash function result values has been completed. A vectorization-based storage search method using bloom filters in a database management system with storage.
According to claim 1,
In the second step, the data processing means generates a hash function control vector consisting of a hash function sequence that calculates a hash function result value for each index key,
In the sixth step, the data processing means removes the corresponding index key from the hash vector and removes the index key from the removed index key when the hash function sequence number for each index key of the hash function control vector and the preset number of hash functions are the same. A vectorization-based storage search method using a bloom filter in a database management system having a plurality of storages, characterized in that a hash vector is updated by inserting a ranking index key.
According to claim 1,
In the third step, the data processing means generates an offset vector having the same value as the result of the hash function for each index key, and based on the offset value for each index key, checks the bloom filter check value of a bit corresponding to a multiple of the offset value in the bloom filter. Generate a test vector by extracting a preset number,
In the fourth step, the data processing means determines a first bit as a target bit of a corresponding index key in a test vector for each index key. A vectorization-based storage search method using a bloom filter in a database management system having multiple storages.
The method according to claim 1 or 2,
In the fifth or sixth step, when the index key is removed from the hash vector, the data processing means determines whether the number of index keys to be queued is equal to or greater than the number of index keys removed from the hash vector, and the number of index keys to be queued. If is less than the number of index keys removed from the hash vector, a plurality of storages characterized by performing a conventional bloom filter inspection operation to extract a bloom filter check value by applying a hash function result value to each bloom filter for each remaining index key A vectorization-based storage search method using a bloom filter in a database management system having a.
According to claim 1,
In the fifth step, the data processing means generates a mask set by ANDing a zero vector and a bit vector consisting of a bloom filter invalid value, and determines that the target bit having the same value as the zero vector in the mask set has an invalid value. A vectorization-based storage search method using a bloom filter in a database management system having multiple storages characterized by the above-mentioned.
According to claim 1,
The data processing means uses a single instruction multiple data (SIMD) collection command that simultaneously calculates multiple values for a single command to perform bloom filter inspection processing for each storage, respectively, and manages a database with multiple storages. A vectorization-based storage search method using bloom filters in the system.

Images