CN109033301A - A kind of db transaction execution method based on graphics processor - Google Patents
A kind of db transaction execution method based on graphics processor Download PDFInfo
- Publication number
- CN109033301A CN109033301A CN201810778840.9A CN201810778840A CN109033301A CN 109033301 A CN109033301 A CN 109033301A CN 201810778840 A CN201810778840 A CN 201810778840A CN 109033301 A CN109033301 A CN 109033301A
- Authority
- CN
- China
- Prior art keywords
- affairs
- group
- transaction
- vertex
- depth
- 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.)
- Granted
Links
Landscapes
- Processing Or Creating Images (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of, and the db transaction based on graphics processor executes method, is characterized in the step of taking affairs fractionation, building transaction dependency figure, the division of Lothrus apterus affairs set, basic operation grouping and affairs to execute；By the way that affairs are split as basic operation, the dependence between affairs is obtained according to the conflict between basic operation, constructs transaction dependency figure；When executing affairs, each basic operation is executed by a GPU thread, compared with existing each GPU thread executes the execution method of an affairs, under the premise of guaranteeing that affairs execute correctly, the concurrency for improving affairs execution shortens affairs and executes the time, improves throughput of system.
Description
Technical field
The invention belongs to calculator memory database technical fields, and in particular in widely used graphics processor
(GPU) realize that quick db transaction executes method on.
Background technique
In August, 1999, NVIDIA company have issued the graphic chips Geforce 256 that a code name is NV10.It is
A product for proposing GPU (Graphics Processing Unit) concept, is mainly used to carry out image procossing.In recent years,
As the computing capability of GPU is stronger and stronger, its application field gradually expands to general-purpose computations field.It has already appeared at present
Accelerate the technology of Transaction Processing (OLTP) memory database with GPU, but current scheme is executed using a thread
The sequence executive mode of one affairs.Since an affairs are made of multiple basic operations, and part basic operation can be parallel
It executes, single thread sequence execution pattern cannot sufficiently excavate the concurrency inside affairs, so that the calculating of GPU cannot be made full use of
Resource and memory bandwidth.
Summary of the invention
The purpose of the present invention is to propose to a kind of, and the db transaction based on graphics processor executes method, to overcome existing skill
The drawbacks described above of art, in the case where guaranteeing low overhead, the concurrency that enhancing affairs execute shortens affairs and executes the time, improves
The handling capacity of system execution affairs.
The present invention is based on the db transactions of graphics processor to execute method, it is characterised in that the following steps are included:
Step 1: affairs are split
According to the input data of a collection of affairs, affairs are split into basic operation set, a basic operation is by [data
, transaction number, action type (reading or writing)] it indicates, if two basic operations access the same data item, and wherein at least
One basic operation is write operation, then it is assumed that the two data item conflicts；At least with a number in two affairs t1 and t2
It is conflict according to item, then it is assumed that the two affairs have dependence；
Step 2: building transaction dependency figure
Initial transaction dependency graph is sky, by the sequence of transaction number from small to large successively add all affairs and accordingly according to
Rely transaction dependency figure；Vertex represents an affairs；When meeting following three conditions, one is added between affairs t1 and t2
Side: 1. there are dependences between two affairs t1 and t2, and 2. the transaction number of affairs t1 is less than the transaction number of affairs t2,3. do not deposit
In another affairs t, so that the transaction number of affairs t is between affairs t1 and t2 and affairs t and affairs t1, t2 have dependence
Relationship, affairs t1 are known as the prefix vertex of affairs t2；
Step 3: Lothrus apterus affairs set divides
In the transaction dependency figure constructed, source summit is the vertex on no prefix vertex, corresponding affairs independent of
Any other affairs；The depth of vertex v is the longest path length from source summit to vertex v, and the depth of source summit is zero；Have
The affairs of same depth are put into the same set, and the collection that the vertex that depth is k is constituted is collectively referred to as k-set, the depth of transaction dependency figure
Degree is the maximum value of all vertex depth；
Step 4: operation grouping
In each k-set, all read operations are added same group and call 0- group, all write operations is added same
Group is known as 1- group；Without dependence between operation in 0- group, without dependence between the operation in 1- group；1- group depends on 0
Group；
Step 5: affairs execute
GPU successively executes affairs according to 0-set, the sequence of 1-set ..., k-set；Inside each set, according to 0-
Group, the sequence of 1- group execute operation, and an operation is executed by a GPU thread.
The affairs based on graphics processor of aforementioned present invention execute method, take affairs and split, construct transaction dependency
The step of figure, Lothrus apterus affairs set divide, operation is grouped, affairs execute；By the way that affairs are split as basic operation, according to base
Conflict between this operation obtains the dependence between affairs, transaction dependency figure is constructed, to know between global transaction
Dependence；Transaction phase is being executed, each basic operation is executed by a GPU thread, with existing each affairs by one
The method that a GPU thread executes is compared, and under the premise of guaranteeing that affairs execute correctly, improves the concurrency of affairs execution, contracting
Short affairs execute the times, improve throughput of system.
Detailed description of the invention
Fig. 1 is that the present invention is based on the implementation operating process schematic diagrams that the affairs of graphics processor execute method.
Fig. 2 is input affairs composition schematic diagram.
Fig. 3 is the basic operation schematic diagram in data item.
Fig. 4 is that transaction dependency diagram is intended to.
Fig. 5 is Lothrus apterus affairs assembly diagram.
Fig. 6 is that basic operation is grouped schematic diagram.
Fig. 7 is affairs implementation procedure schematic diagram.
Specific embodiment
With reference to the accompanying drawing by specific embodiment to the present invention is based on the affairs of graphics processor execution methods to make into one
The detailed description of step.
Embodiment 1:
Fig. 1 is that the present invention is based on the implementation operating process schematic diagrams that the affairs of graphics processor execute method.
The present embodiment executes method based on the affairs of graphics processor, specifically includes the following steps:
Step 1: affairs are split
Fig. 2 gives the input affairs composition of embodiment, and every a line indicates an affairs in figure.This batch of affairs include affairs
It number is T1, T2, T3, four incremental affairs of T4 and transaction number, R indicates read operation, and W indicates that write operation, a, b, c are three need
The data item to be accessed, such as Ra indicate item of read data a.The composition of affairs T1 operates are as follows: Ra, Wa, Rb, Wb.The group of affairs T2
At operation are as follows: Rc, Wc, Ra.The composition of affairs T3 operates are as follows: Rb, Wb.The composition of affairs T4 operates are as follows: Rc, Wc, Wb.This implementation
Affairs input of the example based on attached drawing 1 carries out the affairs based on graphics processor and executes method.
Fig. 3 indicates the operation chart of input affairs.
One basic operation is indicated by [data item, transaction number, action type], and according to the input of affairs, affairs are split
For basic operation (1. referring to the flow operations frame in Fig. 1).The basic behaviour for needing to carry out in each data item is listed in attached drawing 3
Make, every a line indicates all basic operations for needing to carry out in a data item.Two basic operations access the same data item,
And wherein at least one basic operation is write operation, then the two data item conflicts；It is at least closed in two affairs t1 and t2
It is conflict in a data item, then the two affairs have dependence.
Step 2: building transaction dependency figure
Fig. 4 is that transaction dependency diagram is intended to.
Originally transaction dependency figure is empty.Affairs T1 is added in transaction dependency figure (referring to the flow operations in Fig. 1 first
Frame is 2.), due to not adding side without the affairs with affairs T1 conflict.Then, affairs T2 is added to transaction dependency figure
In, the operation [a, T2, R] of affairs T2 and the operation [a, T1, W] of T1 conflict with each other, and the transaction number of affairs T1 is less than affairs T2, no
There are transaction numbers between affairs T1, affairs T2 and and their affairs for all conflicting, add an affairs T1 and be directed toward affairs
The side (3. referring to the flow operations frame in Fig. 1) of T2.Affairs T3 can be similarly added in transaction dependency figure figure.Finally, affairs
T4 and T2 conflicts about data item c and affairs T1, T3 conflict about data item b, due to having affairs T3 between affairs T1 and T4,
So not adding side between affairs T1 and T4, side only is added between affairs T2 and T4, T3 and T4.The transaction dependency built
Figure is as shown in Fig. 4.
Step 3: Lothrus apterus affairs set divides
Fig. 5 is Lothrus apterus affairs assembly diagram.
In the transaction dependency figure that second step is built, the depth of affairs T1 is 0, and 0-set is added in affairs T1.Affairs
The depth of T2, T3 are 1, and 1-set is added in affairs T2, T3.The depth of affairs T4 is 2, and 2-set is added in affairs T4.To complete
At the division (4. referring to the flow operations frame in Fig. 1) of Lothrus apterus affairs set.Attached drawing 5 gives ready-portioned k-set set.
Step 4: basic operation is grouped
Fig. 6 is that basic operation is grouped schematic diagram.
Each k-set is the affairs set of Lothrus apterus, and gathering internal affairs can execute parallel, is needed at this time into one
The dependence of basic operation inside step analysis affairs.Analysis is found, read-write dependency is only existed inside affairs, other shapes are not present
The dependence of formula.Therefore the read operation in each k-set is divided into 0- group, write operation is divided into 1- group.1- group depends on
0- group.After segmentation in 0-set 0- group content are as follows: [a, T1, R], [b, T1, R], the content of 1- group are as follows: [a, T1, W], [b, T1,
W] (5. referring to the flow operations frame in Fig. 1).Whole basic operation grouping situations is given in attached drawing 6.
Step 5: affairs execute
Fig. 7 is affairs implementation procedure schematic diagram.
In order to guarantee that affairs correctly execute, need according to 0-set, 1-set, the sequence of 2-set successively executes affairs, this hair
The bright concurrency that in such a way that a thread executes an operation, raising affairs are executed.Meanwhile inside set, operation is pressed
It is executed according to 0- group, the sequence of 1- group.According to the basic operation division result of the 4th step, the execution that affairs are given in attached drawing 6 is suitable
One read operation is indicated with a reading word in Fig. 7, is write for the ease of expression by sequence (6. referring to the flow operations frame in Fig. 1)
Operation is write with one and is indicated.
In the present embodiment, method is executed by the affairs based on graphics processor, by being split as grasping substantially by affairs
Make and an operation is executed by a thread, while guaranteeing that affairs execute correct, enhances the concurrency of affairs execution.
In the sequence execution pattern that a thread executes an affairs, the affairs of the present embodiment were needed by 10 read-write stages.?
A thread proposed by the present invention executes in the execution pattern of a basic operation, it is only necessary to pass through 6 read-write stages, shorten
Affairs execute the time, and improve resource utilization ratio, improve throughput of system.
Claims (1)
1. a kind of db transaction based on graphics processor executes method, it is characterised in that the following steps are included:
Step 1: affairs are split
According to the data access requirements of a collection of affairs, affairs are split into multiple basic operations, a basic operation is by [data
, transaction number, action type] it indicates, if two basic operations access the same data item, and wherein at least one base
This operation is write operation, then it is assumed that the two data item conflicts；In two affairs t1 and t2 at least with a data item be punching
Prominent, then it is assumed that the two affairs have dependence；
Step 2: building transaction dependency figure
Initial transaction dependency graph is sky, successively adds all affairs by the sequence of transaction number from small to large and corresponding dependence is arrived
Transaction dependency figure；Vertex represents an affairs；When meeting following three conditions, a line is added between affairs t1 and t2:
1. there are dependences between two affairs t1 and t2,2. the transaction number of affairs t1 is less than the transaction number of affairs t2, is 3. not present
Another affairs t, so that the transaction number of affairs t is between affairs t1 and t2 and affairs t and affairs t1, t2 have dependence to close
System, affairs t1 are known as the prefix vertex of affairs t2；
Step 3: Lothrus apterus affairs set divides
In the transaction dependency figure constructed, source summit is the vertex on no prefix vertex, and corresponding affairs are independent of any
Other affairs；The depth of vertex v is the longest path length from source summit to vertex v, and the depth of source summit is zero；With identical
The affairs of depth are put into the same set, and the collection that the vertex that depth is k is constituted is collectively referred to as k-set, and the depth of transaction dependency figure is
The maximum value of all vertex depth；
Step 4: basic operation is grouped
In each k-set, all read operations are added same group and call 0- group, all write operations are added same group and are claimed
For 1- group；Without dependence between operation in 0- group, without dependence between the operation in 1- group；1- group depends on 0 group；
Step 5: affairs execute
Affairs are successively executed according to the sequence of 0-set, 1-set ..., k-set；Inside each affairs set, according to 0- group, 1-
The sequence of group executes operation, and an operation is executed by a GPU thread.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810778840.9A CN109033301B (en) | 2018-07-16 | 2018-07-16 | Database transaction execution method based on graphic processor |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810778840.9A CN109033301B (en) | 2018-07-16 | 2018-07-16 | Database transaction execution method based on graphic processor |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109033301A true CN109033301A (en) | 2018-12-18 |
| CN109033301B CN109033301B (en) | 2021-07-06 |
Family
ID=64643124
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810778840.9A Active CN109033301B (en) | 2018-07-16 | 2018-07-16 | Database transaction execution method based on graphic processor |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109033301B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110287212A (en) * | 2019-06-27 | 2019-09-27 | 浪潮商用机器有限公司 | A kind of data service handling method, system and associated component |
| CN110287006A (en) * | 2019-06-10 | 2019-09-27 | 广州思创科技发展有限公司 | A kind of data dispatching method and system |
| CN116775315A (en) * | 2023-08-22 | 2023-09-19 | 北京遥感设备研究所 | Multi-core CPU concurrent transaction allocation method based on dependency graph |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060080285A1 (en) * | 2004-10-13 | 2006-04-13 | Sybase, Inc. | Database System with Methodology for Parallel Schedule Generation in a Query Optimizer |
| CN103077006A (en) * | 2012-12-27 | 2013-05-01 | 浙江工业大学 | Multithreading-based parallel executing method for long transaction |
| CN103942342A (en) * | 2014-05-12 | 2014-07-23 | 中国人民大学 | Memory database OLTP and OLAP concurrency query optimization method |
| CN104021145A (en) * | 2014-05-16 | 2014-09-03 | 华为技术有限公司 | Mixed service concurrent access method and device |
| CN104317850A (en) * | 2014-10-14 | 2015-01-28 | 北京国双科技有限公司 | Data processing method and device |
| CN106445687A (en) * | 2016-09-27 | 2017-02-22 | 金蝶软件（中国）有限公司 | Large transaction execution method and system |
| CN108139927A (en) * | 2015-10-01 | 2018-06-08 | 华为技术有限公司 | The routing based on action of affairs in online transaction processing system |
-
2018
- 2018-07-16 CN CN201810778840.9A patent/CN109033301B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060080285A1 (en) * | 2004-10-13 | 2006-04-13 | Sybase, Inc. | Database System with Methodology for Parallel Schedule Generation in a Query Optimizer |
| CN103077006A (en) * | 2012-12-27 | 2013-05-01 | 浙江工业大学 | Multithreading-based parallel executing method for long transaction |
| CN103942342A (en) * | 2014-05-12 | 2014-07-23 | 中国人民大学 | Memory database OLTP and OLAP concurrency query optimization method |
| CN104021145A (en) * | 2014-05-16 | 2014-09-03 | 华为技术有限公司 | Mixed service concurrent access method and device |
| CN104317850A (en) * | 2014-10-14 | 2015-01-28 | 北京国双科技有限公司 | Data processing method and device |
| CN108139927A (en) * | 2015-10-01 | 2018-06-08 | 华为技术有限公司 | The routing based on action of affairs in online transaction processing system |
| CN106445687A (en) * | 2016-09-27 | 2017-02-22 | 金蝶软件（中国）有限公司 | Large transaction execution method and system |
Non-Patent Citations (1)
| Title |
|---|
| CHENG ZHONG等: "Parallel Identifying(l, d)-Motifs in Biosequences Using CPU and GPU Computing", 《SPRINGER》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110287006A (en) * | 2019-06-10 | 2019-09-27 | 广州思创科技发展有限公司 | A kind of data dispatching method and system |
| CN110287212A (en) * | 2019-06-27 | 2019-09-27 | 浪潮商用机器有限公司 | A kind of data service handling method, system and associated component |
| CN116775315A (en) * | 2023-08-22 | 2023-09-19 | 北京遥感设备研究所 | Multi-core CPU concurrent transaction allocation method based on dependency graph |
| CN116775315B (en) * | 2023-08-22 | 2024-01-02 | 北京遥感设备研究所 | Multi-core CPU concurrent transaction allocation method based on dependency graph |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109033301B (en) | 2021-07-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109033301A (en) | A kind of db transaction execution method based on graphics processor | |
| US8442935B2 (en) | Extract, transform and load using metadata | |
| US8364751B2 (en) | Automated client/server operation partitioning | |
| US20110185359A1 (en) | Determining A Conflict in Accessing Shared Resources Using a Reduced Number of Cycles | |
| CN106021484A (en) | Customizable multi-mode big data processing system based on memory calculation | |
| CN110209650A (en) | The regular moving method of data, device, computer equipment and storage medium | |
| CN103761236A (en) | Incremental frequent pattern increase data mining method | |
| CN109508355A (en) | A kind of data pick-up method, system and terminal device | |
| US20180357298A1 (en) | Performance of Distributed Databases and Database-Dependent Software Applications | |
| CN111897891B (en) | Data processing method and device | |
| CN113535258B (en) | Pattern file loading method and system | |
| CN112148713B (en) | Method and device for data migration between heterogeneous databases | |
| CN108133005A (en) | A kind of environmental model analogy method, terminal device and storage medium based on memory database | |
| CN108334565A (en) | A kind of data mixing storage organization, data store query method, terminal and medium | |
| CN116560629A (en) | Serialization method, device, medium and equipment for data transmission object | |
| CN103246498A (en) | Memory storage structures supporting relational data parallel processing and achieving method thereof | |
| CN113434522B (en) | Parallel updating method and updating system on state tree oriented to alliance chain | |
| CN109857832A (en) | A kind of preprocess method and device of payment data | |
| CN111753027B (en) | Method and device for implicit knowledge graph representation facing multiple entities | |
| US20220091831A1 (en) | A streaming compiler for automatic adjoint differentiation | |
| CN114138735A (en) | Method for quickly loading Janus graph data in batches | |
| CN102063308A (en) | Method for controlling processing flow of seismic prospecting data | |
| EP3113038A1 (en) | A data handling method | |
| CN110989995A (en) | Processing method and system for artificial intelligence application based on open source deep learning framework | |
| GB2539898A (en) | A data handling method |
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 |