CN111552710B - Query optimization method for distributed database - Google Patents

Query optimization method for distributed database Download PDF

Info

Publication number
CN111552710B
CN111552710B CN202010352089.3A CN202010352089A CN111552710B CN 111552710 B CN111552710 B CN 111552710B CN 202010352089 A CN202010352089 A CN 202010352089A CN 111552710 B CN111552710 B CN 111552710B
Authority
CN
China
Prior art keywords
filter
tikv
data
dimension
algorithm
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202010352089.3A
Other languages
Chinese (zh)
Other versions
CN111552710A (en
Inventor
侯孟书
樊敏
何东升
杨键
曾骁阳
周世杰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN202010352089.3A priority Critical patent/CN111552710B/en
Publication of CN111552710A publication Critical patent/CN111552710A/en
Application granted granted Critical
Publication of CN111552710B publication Critical patent/CN111552710B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2453—Query optimisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/28—Databases characterised by their database models, e.g. relational or object models
    • G06F16/284—Relational databases
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses a query optimization method of a distributed database, which comprises the following steps: modifying calculation operation in the LIP algorithm, pushing down part of calculation, establishing a hash table based on the aggregated data after the multi-table aggregation operation is completed, and completing detection to form a distributed LIP algorithm; constructing a filter in a storage layer, constructing a filter by each partition table of a dimension table after the data completes table scanning and operator operation in a TiKV node, and distributing the constructed filter to other nodes through a network; modifying a loss rate calculation formula, detecting a filter, calculating the loss rate of the filter, and arranging the filters in ascending order according to the loss rate; and returning the data to the TiKV node after operator calculation and filter survey of the TiKV node, and returning the aggregated query result to the client after aggregation on the TiDB server. The invention can reduce the influence of sub-optimal plan execution on the query performance, improve the performance of TiDB under star model query, and reduce the data network overhead under the distributed environment.

Description

Query optimization method for distributed database
Technical Field
The invention relates to the field of computer database management, in particular to a query optimization method of a distributed database.
Background
A database management system (RDBMS) is a layer of data management software located between a user and an operating system, and is a core component in a modern computer environment, and provides functions of data storage, management, processing, maintenance, and the like. The relational database ensures the integration and sharing of data, can integrate the data and the relationship thereof together and store the data in a certain structural form, is shared by a plurality of different users at the same time, and provides great convenience for the data management of enterprises and government departments at the moment. But with the continued development of the internet, the amount of data that is generated and needs to be processed has grown dramatically. One problem with the dramatic growth of data is that it is difficult to effectively store and process the data on a single machine, and it is difficult to meet the requirements of efficient processing and storage of data in the new era.
In order to solve the problem that the relational database cannot cope with various challenges in the big data age, the architecture of a database system is greatly changed, the appearance of a distributed relational database (namely newSQL) fuses SQL and NoSQL modes, and the high expansibility of an SQL interface, distributed transactions and clusters is provided for the outside. After technologies such as distributed transaction, mapping from SQL to NoSQL and the like are basically complete, compared with traditional relational data, the application scene of the NewSQL database is not limited to online transaction analysis, and the application scene also comprises a plurality of scenes of large complex analysis query and offline analysis due to the increase of the stored data volume. While the advent of NoSQL can solve many of the problems at the present time, many remain. The most important and common point is that all NoSQL databases do not support the SQL standard, which brings inconvenience to numerous companies and researchers who widely use SQL as an upper layer interface, and also causes migration difficulty of applications specially designed for SQL. How to improve the query performance in large complex analysis query and OLAP scenarios is a key issue for NewSQL databases.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, provides a query optimization method of a distributed database, and provides stable, expected or near-expected query performance for each query of the distributed database by reducing network overhead of data and transmission and a connection reordering algorithm in an execution process under a distributed environment based on a novel NewSQL database TiDB, so that the influence on performance after an optimizer selects a suboptimal query plan is reduced, the robustness of the query plan is improved.
The aim of the invention is realized by the following technical scheme:
in the query optimization process of the distributed database, a step of improving an optimized LIP algorithm is added, and the step of improving the LIP algorithm comprises the following steps:
modifying calculation operation in the LIP algorithm, pushing down part of calculation, not pushing down two operations of constructing a hash table and detecting the hash table, and then establishing the hash table based on the aggregated data and completing detection after multi-table aggregation operation is completed to form a distributed LIP algorithm;
constructing a filter on the storage layer, constructing a filter by each partition table of the dimension table after the data completes table scanning and other operator operations on the TiKV nodes, and distributing the constructed filter to other TiKV nodes in the TiKV cluster through a network;
modifying the loss rate calculation formula, and changing the filter loss rate calculation formula into:
Figure BDA0002472206050000021
wherein i is the number of fragments;
and after all the data are subjected to operator calculation and filter survey of the TiKV node, returning the TiKV node, and after aggregation is carried out on the TiDB server, returning an aggregated query result to the client.
Specifically, the filter constructed by the storage structure is a cuckoo filter, and is used for searching elements in the hash table, dynamically adding and deleting any elements into the table, and further improving the space utilization efficiency.
Specifically, the calculation operations of the distributed LIP algorithm sequentially include: single table scanning and filtering, filter construction, probe filter and ordering algorithm, single table aggregation, multi-table aggregation, join operation, hash table construction, probe hash table.
Specifically, the step of improving the optimized LIP algorithm to achieve query optimization includes:
s1, after receiving a query request, a TIDB server partitions a dimension table and a fact table according to a partitioning strategy to obtain a dimension partition table and a fact partition table, and distributes the dimension partition table and the fact partition table to different TiKV nodes through a network for storage;
s2, the TiKV node scans the table data in the dimension partition table and the fact partition table respectively, filters the scanned table data according to the filtering condition, and outputs the filtering results of the dimension partition table and the fact partition table;
s3, each TiKV node constructs a cuckoo filter by using a dimension partition table, a filtering result of the dimension partition table is inserted into the filter, and the filter is distributed to the fact partition tables of other TiKV nodes in the TiKV cluster through a network;
s4, after the fact partition table of other TiKV nodes receives the filter, circularly detecting the filter by using a sequencing algorithm, calculating the deletion rate of the filter by using an operator according to a modified deletion rate calculation formula, and arranging the filter in an ascending order according to the deletion rate;
s5, after being calculated by a TiKV node operator and detected by a filter, all the table data are returned to a TiKV Server end, and the TiKV Server end returns the table data to a TiDB Server through a network;
and S6, after receiving all the table data, the TIDB server performs single-table aggregation and multi-table aggregation operations to form a new dimension table and a fact table, performs join operation on the dimension table formed by aggregation to construct a hash table, detects the hash table according to the data matching condition, and returns the data meeting the matching condition to the client.
Specifically, gRPC service is adopted between the TiKV node and the TiDB server to complete network communication.
Specifically, in step S3, before inserting the filtering result of the dimension partition table into the filter, the method further includes: the TiKV node keeps the filtering result of the dimension partition table in a localFilter structure of the local node.
The invention has the beneficial effects that:
1. the performance of the TiDB under a star model and similar query can be improved, and the data network overhead under a distributed environment is reduced;
2. the influence of sub-optimal plan execution on the query performance can be reduced, and the robustness of the query plan is ensured.
Drawings
Fig. 1 is a flow chart of the method of the present invention.
Fig. 2 is a sequence diagram of the computational operations of the LIP algorithm of the present invention.
Fig. 3 is a sequence diagram of the calculation operation after the LIP algorithm of the present invention is optimized.
FIG. 4 is a diagram of a preferred embodiment of the filter lookup element of the present invention.
Fig. 5 is a diagram of a preferred embodiment of the detection filter of the present invention.
Fig. 6 is a flow chart of the LIP algorithm after improvement and optimization according to the present invention.
Detailed Description
For a clearer understanding of technical features, objects, and effects of the present invention, a specific embodiment of the present invention will be described with reference to the accompanying drawings.
The invention mainly realizes the query optimization of TiDB distributed relation data in a distributed environment. The query optimization method is mainly based on LIP algorithm improvement optimization, and the original LIP algorithm is mainly divided into four stages and is described as follows:
first, build phase. After each dimension table completes the table scanning and selecting operation, a hash table and a filter are established on the dimension table.
And a second stage, a filter detection stage. And detecting the filters on all dimension tables according to the data in the fact table, recording the hit number and the miss number of the filters, and recording the miss rate.
And a third stage, namely, a self-adaptive sequencing stage. The algorithm can continuously call the next method of the executor to acquire data according to the size of one batch_size, that is, the processing of the data takes the batch_size as a unit, so that the size of the batch_size can be continuously increased in the iterative process in order to speed up the convergence speed of the algorithm. At this stage, all filters are reordered by the probability of absence obtained after the filters screen the data.
Fourth stage, hash table detection stage. And detecting the hash table according to the ordered connection sequence, and returning the data meeting the matching condition.
In this embodiment, as shown in fig. 1, a query optimization method of a distributed database performs improved optimization on an LIP algorithm, where the method includes:
firstly, modifying calculation operation in the LIP algorithm, pushing down part of calculation, not pushing down two operations of constructing a hash table and detecting the hash table, and after the multi-table aggregation operation is completed, establishing the hash table based on the aggregated data and completing detection to form the distributed LIP algorithm. The calculation operation sequence of the push-down part in the original LIP algorithm is as shown in fig. 2, and sequentially includes: constructing a filter, detecting a filter and ordering algorithm, constructing a hash table, detecting a hash table, single-label scanning and filtering, single-table aggregation, multi-table aggregation and join operation. The operations in FIG. 2 for LIP algorithm construction are represented using Greek letters, and the related operations for native in the TiDB database are represented in uppercase letters, the second item of the diagram describing whether the operation has been pushed down. In this embodiment, the hash table is selected not to be pushed down, but is built based on the aggregated data after multi-table aggregation and detection is completed, and the calculation operation sequence of the distributed LIP algorithm is as shown in fig. 3, and the sequence is as follows: single table scanning and filtering, filter construction, probe filter and ordering algorithm, single table aggregation, multi-table aggregation, join operation, hash table construction, probe hash table.
And then, constructing a filter in the storage layer, constructing the filter by each partition table of the dimension table after the data completes table scanning and other operator operations in the TiKV node, and distributing the constructed filter to other nodes in the cluster through a network. The filter adopts a cuckoo filter, 4 hash functions and 2 slots are configured for construction in the filter construction process, and elements can be searched in a hash table, dynamically added and deleted. As shown in fig. 4, is a preferred embodiment of the filter lookup element of the present invention. In fig. 4, when searching element X, fingerprint information f of the element is first calculated, then the address of the hash is solved for two times by the fingerprint information and the value of the element, the fingerprint f is searched in the hash bucket corresponding to the address, and the slot position hung after the bucket is traversed and searched for f, if the fingerprint f is not found, the search fails, otherwise, the search is successful.
Secondly, modifying a loss rate calculation formula, and changing the calculation formula of the loss rate of the filter into:
Figure BDA0002472206050000041
where i is the number of slices. As shown in fig. 5, is a preferred embodiment of the detection filter of the present invention. In fig. 5, taking a TiKV cluster of three nodes as an example, assuming T as a fact table, tables a, B, and C as dimension tables of T, all table data are stored on the three TiKV nodes according to range division. T1, T2 and T3 respectively represent the partial table data after slicing, and the dimension tables A, B and C are the same. After the dimension table is calculated by operators such as scan and select, a filter is built on the screened data, and the built filter is sent to other nodes through a network. The amount of data transmitted over the network is not significant because the filter itself occupies little space. After the nodes of the fact tables T1, T2 and T3 receive the filters of all dimension tables, a filter detection stage and an adaptive ordering stage are started. At this time, the filter loss rate calculation of the dimension table a becomes the sum of three sub-table filters. After the fact table is filtered by the filter, the data is sent to the SQL layer, so that network overhead is saved.
And finally, returning all the data to the TiKV node after operator calculation and filter survey of the TiKV node, and returning the aggregated query result to the client after aggregation on the TiDB server.
The improved and optimized LIP algorithm flow is shown in fig. 6, and the key optimization point is that fact tables are screened in advance, and the screened data are data which are required to be returned by the SQL layer, so that network overhead of data transmission in a distributed environment is greatly saved.
In this embodiment, the data query optimization process implemented by the distributed database TiDB applying the LIP algorithm after improvement optimization includes:
firstly, after receiving a query request, the TIDB server partitions the dimension table and the fact table according to a partitioning strategy to obtain a dimension partition table and a fact partition table, and distributes the dimension partition table and the fact partition table to different TiKV nodes through a network for storage.
And secondly, respectively scanning the table data in the dimension partition table and the fact partition table by the TiKV node, filtering the scanned table data according to the filtering condition, and outputting the filtering results of the dimension partition table and the fact partition table.
Thirdly, each TiKV node constructs a cuckoo filter by using a dimension partition table, a filtering result of the dimension partition table is inserted into the filter, meanwhile, the TiKV node keeps the filtering result of the dimension partition table in a localFilter structure of a local node, and distributes the filter to fact partition tables of other TiKV nodes in the TiKV cluster through a network.
And fourthly, after the fact partition table of other TiKV nodes receives the filter, circularly detecting the filter by using a sequencing algorithm, calculating the deletion rate of the filter by using an operator according to the modified deletion rate calculation formula, and arranging the filter in an ascending order according to the deletion rate. The sorting algorithm used by the TiKV node is a self-sorting algorithm.
And fifthly, after being calculated by a TiKV node operator and detected by a filter, all the table data are returned to a TiKV Server end, and the TiKV Server end returns the table data to the TiDB Server through a network.
And sixthly, after receiving all the table data, the TIDB server performs single-table aggregation and multi-table aggregation operations to form a new dimension table and a new fact table, performs join operation on the dimension table formed by aggregation to construct a hash table, detects the hash table according to the data matching condition, and returns the data meeting the matching condition to the client.
And adopting gRPC service between the TiKV node and the TiDB server in the data query optimization process to complete network communication.
In this embodiment, adding a distLIP operator to the TiKV node may better implement the improved LIP algorithm, where the distLIP operator mainly implements the third step and the fourth step in the data query optimization process of this embodiment. In the distLIP operator, the execution result data of the bottom operator can be obtained without directly calling the interface of the bottom layer, so that some extra processing work is reduced; and the code organization mode and structure of the original executor module can not be damaged by adding operators, only the information of the newly added distLIP operator is needed to be newly constructed in the implementation class of the executor Runner interface, and a run file is modified, so that the new distLIP operator is added in the process of constructing the executor. The distLIP operator, from the result of the operation, filters the data in the fact table, and may be placed after the underlying operators tableScan, indexScan, selection, and computed before the operators limit, topN, aggregation that contain computational effort.
In this embodiment, the improved and optimized LIP algorithm mainly includes a processing flow of a dimension table and a processing flow of a fact table in the process of implementing query optimization.
The processing flow for the dimension table is as follows: firstly, a next method is called to obtain a result executed by a selection operator, then a filter is established on the result, in the establishment process, firstly, decoding operation is carried out on the obtained key (because a mode of lazy-decoding is used in TiKV, the key is decoded only when needed), and then the key is inserted into the filter. The newly created filter will call the send_msg method, where the service layer of TiKV is called to provide network services, and broadcast data through the gRPC service. In addition, the dimension table can be reserved in a local localFilter structure.
The processing flow for the fact table is as follows: and obtaining the execution result of the selection operator by calling a next method, then calling a network layer interface by using a direct_msg method to obtain data sent by other TiKV nodes, obtaining local filter data from a local localFilter, and finally transmitting the two to a get_filters method for analysis. According to the obtained fact table data and filter data, the fact table data and the filter data are transmitted to a self-adaptive ordering algorithm of the distLIP to be processed, a processing result is returned, the processing result is continuously called by the next operator, the processing result is finally returned to the RequestHandler, and the processing result is transmitted back to the TiDB server through a network. After the TiDB receives the returned data, data aggregation operation is carried out, join operation is carried out on the data, the join operation itself can construct a hash table to select the data, and erroneous judgment of the data in a distSQL operator can be eliminated.
In the processing flow of the fact table, the input of the self-ordering algorithm is a filter list generated by all dimension tables and the tuple data of the fact table, and the output is the tuple filtered by the filter. The tuple data takes the batch_size as a unit, detects all filters by the data of the batch_size every cycle, and records the missing number and the record number of each dimension table for the batch of data. After one cycle (processing data of one batch), the total deletion rate of one dimension table is obtained, and all filters are arranged according to the ascending order of the deletion rate to participate in the calculation of the data of the next batch again. The sorting of the filters according to the miss rate may ensure that the subsequent detection process first detects filters that can screen more data, thereby reducing the size of the intermediate result or radix.
The foregoing has shown and described the basic principles and main features of the present invention and the advantages of the present invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, and that the above embodiments and descriptions are merely illustrative of the principles of the present invention, and various changes and modifications may be made without departing from the spirit and scope of the invention, which is defined in the appended claims. The scope of the invention is defined by the appended claims and equivalents thereof.

Claims (5)

1. The query optimization method of the distributed database is characterized in that in the query optimization process of the distributed database, a step of improving an optimized LIP algorithm is added, and the step of improving the LIP algorithm comprises the following steps:
modifying the calculation operation of the push-down part in the LIP algorithm, not performing push-down processing on two operations of constructing a hash table and detecting the hash table, and then establishing the hash table based on the aggregated data and completing detection after the multi-table aggregation operation is completed to form the calculation operation of the push-down part of the distributed LIP algorithm;
constructing a filter on the storage layer, constructing a filter by each partition table of the dimension table after the data completes table scanning and other operator operations on the TiKV nodes, and distributing the constructed filter to other TiKV nodes in the TiKV cluster through a network;
modifying the loss rate calculation formula, and changing the filter loss rate calculation formula into:
Figure FDA0004189434710000011
wherein i is the number of fragments;
after all data are subjected to operator calculation and filter survey of the TiKV node, returning to the TiKV node, and after aggregation is carried out on the TiDB server, returning an aggregated query result to the client;
the step of improving the optimized LIP algorithm to achieve query optimization includes:
s1, after receiving a query request, a TIDB server partitions a dimension table and a fact table according to a partitioning strategy to obtain a dimension partition table and a fact partition table, and distributes the dimension partition table and the fact partition table to different TiKV nodes through a network for storage;
s2, the TiKV node scans the table data in the dimension partition table and the fact partition table respectively, filters the scanned table data according to the filtering condition, and outputs the filtering results of the dimension partition table and the fact partition table;
s3, each TiKV node constructs a cuckoo filter by using a dimension partition table, a filtering result of the dimension partition table is inserted into the filter, and the filter is distributed to the fact partition tables of other TiKV nodes in the TiKV cluster through a network;
s4, after the fact partition table of other TiKV nodes receives the filter, circularly detecting the filter by using a sequencing algorithm, calculating the deletion rate of the filter by using an operator according to a modified deletion rate calculation formula, and arranging the filter in an ascending order according to the deletion rate;
s5, after being calculated by a TiKV node operator and detected by a filter, all the table data are returned to a TiKV Server end, and the TiKV Server end returns the table data to a TiDB Server through a network;
and S6, after receiving all the table data, the TIDB server performs single-table aggregation and multi-table aggregation operations to form a new dimension table and a fact table, performs join operation on the dimension table formed by aggregation to construct a hash table, detects the hash table according to the data matching condition, and returns the data meeting the matching condition to the client.
2. The query optimization method of a distributed database according to claim 1, wherein the filter constructed in the storage layer is a cuckoo filter, and is used for searching elements in a hash table, dynamically adding and deleting any element in the table, and further improving space utilization efficiency.
3. The query optimization method of claim 1, wherein the calculation of the push-down part of the distributed LIP algorithm sequentially comprises: single table scanning and filtering, filter construction, probe filter and ordering algorithm, single table aggregation, multi-table aggregation, join operation, hash table construction, probe hash table.
4. The query optimization method of a distributed database according to claim 1, wherein gRPC service is adopted between the TiKV node and the TiDB server to complete network communication.
5. The method according to claim 1, wherein the step S3 of inserting the filtering result of the dimension partition table into the filter further comprises: the TiKV node keeps the filtering result of the dimension partition table in a localFilter structure of the local node.
CN202010352089.3A 2020-04-28 2020-04-28 Query optimization method for distributed database Active CN111552710B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010352089.3A CN111552710B (en) 2020-04-28 2020-04-28 Query optimization method for distributed database

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010352089.3A CN111552710B (en) 2020-04-28 2020-04-28 Query optimization method for distributed database

Publications (2)

Publication Number Publication Date
CN111552710A CN111552710A (en) 2020-08-18
CN111552710B true CN111552710B (en) 2023-07-04

Family

ID=72007741

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010352089.3A Active CN111552710B (en) 2020-04-28 2020-04-28 Query optimization method for distributed database

Country Status (1)

Country Link
CN (1) CN111552710B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115729889A (en) * 2021-08-31 2023-03-03 华为技术有限公司 Data access method, database system and storage device
CN114817262B (en) * 2022-04-27 2023-03-28 电子科技大学 Graph traversal algorithm based on distributed graph database
CN115048402B (en) * 2022-08-16 2022-11-18 成都信息工程大学 Self-adaptive dynamic data set member inserting, deleting and retrieving method with time effect
CN117633024B (en) * 2024-01-23 2024-04-23 天津南大通用数据技术股份有限公司 Database optimization method based on preprocessing optimization join

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103678520A (en) * 2013-11-29 2014-03-26 中国科学院计算技术研究所 Multi-dimensional interval query method and system based on cloud computing
CN107454162A (en) * 2017-08-01 2017-12-08 成都鼎智汇科技有限公司 A kind of system for improving cloud computing environment reliability
CN110175176A (en) * 2019-05-31 2019-08-27 杭州复杂美科技有限公司 A kind of KV configuration method for database, querying method, equipment and storage medium

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8762407B2 (en) * 2012-04-17 2014-06-24 Renmin University Of China Concurrent OLAP-oriented database query processing method

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103678520A (en) * 2013-11-29 2014-03-26 中国科学院计算技术研究所 Multi-dimensional interval query method and system based on cloud computing
CN107454162A (en) * 2017-08-01 2017-12-08 成都鼎智汇科技有限公司 A kind of system for improving cloud computing environment reliability
CN110175176A (en) * 2019-05-31 2019-08-27 杭州复杂美科技有限公司 A kind of KV configuration method for database, querying method, equipment and storage medium

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Looking ahead makes query plans robust：Making the initial case with in-memory star schema data warehouse workloads;J.Zhu,et all;《Proceedings of the VLDB Endowment》;20171031;全文 *
面向大数据的建筑能耗与环境实时管理云平台架构设计;陈勤平等;《绿色建筑》;20190120(第01期);全文 *

Also Published As

Publication number Publication date
CN111552710A (en) 2020-08-18

Similar Documents

Publication Publication Date Title
CN111552710B (en) Query optimization method for distributed database
US10846285B2 (en) Materialization for data edge platform
US5345585A (en) Method for optimizing processing of join queries by determining optimal processing order and assigning optimal join methods to each of the join operations
CN106528773B (en) Map computing system and method based on Spark platform supporting spatial data management
US7240044B2 (en) Query optimization by sub-plan memoization
US8620888B2 (en) Partitioning in virtual columns
US5832475A (en) Database system and method employing data cube operator for group-by operations
CA2388515C (en) System for managing rdbm fragmentations
JP4747094B2 (en) DML statement for densifying data in a relational database system
CN106294374A (en) The method of small documents merging and data query system
US11573987B2 (en) System for detecting data relationships based on sample data
CN114090695A (en) Query optimization method and device for distributed database
CN110032676B (en) SPARQL query optimization method and system based on predicate association
CN113704248B (en) Block chain query optimization method based on external index
US7617189B2 (en) Parallel query processing techniques for minus and intersect operators
CN114490724B (en) Method and device for processing database query statement
CN113448994B (en) Continuous regrettage minimization query method based on core set
CN115048377A (en) Time-space keyword query method under mixed storage block chain environment
CN111159175B (en) Incomplete database Skyline query method based on index
KR100886607B1 (en) Method of effectively performing limited-union based multiple-queries in a database management system
CN108829796A (en) A kind of real time indexing method for the inquiry of electric power big data efficient combination
CN117131069B (en) Database list grouping scanning method
CN115033597A (en) Method and system for deep learning to participate in SQL optimization of HTAP database
CN117807087A (en) Method and device for converting relational model data into document model data
CN115563148A (en) Database query method and device

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant