CN103049554B - A kind of vector QR tree parallel index method - Google Patents
A kind of vector QR tree parallel index method Download PDFInfo
- Publication number
- CN103049554B CN103049554B CN201210586420.3A CN201210586420A CN103049554B CN 103049554 B CN103049554 B CN 103049554B CN 201210586420 A CN201210586420 A CN 201210586420A CN 103049554 B CN103049554 B CN 103049554B
- Authority
- CN
- China
- Prior art keywords
- tree
- leaf node
- spatial
- leaf
- node
- 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.)
- Expired - Fee Related
Links
Abstract
This patent discloses a kind of vector QR tree parallel index method.Under novel parallel environment, the efficient index for massive spatial data is particularly important, and the height of spatial data index efficiency is the key weighing spatial database overall performance.Current existing parallel index method can not break through the bottleneck that method host node accesses, it is difficult to the problem of load balancing between solution process.For problem above, the present invention devises a kind of parallel index method that Q tree for vector and R tree are collaborative, content includes: 1) central point based on spatial object minimum area-encasing rectangle builds QR tree, utilize manifold Dow process that adjacent space data sets is divided to different process, realize the load balancing of task between process, and optimum indexing path;2) QR tree carrying out master-slave mode storage, optimum indexing reads;3) QR tree is carried out coordinate retrieval, break through root node access bottleneck.The QR tree divided based on spatial object collection builds, QR tree master-slave mode stores, QR tree coordinate retrieval has collectively constituted QR tree parallel index method.QR tree parallel index involved in the present invention can be unit multinuclear, many-core efficient Vector Space Retrieval method a kind of with service offer with the exploitation of the mass data spatial index software of High-Performance Computing Cluster environment.
Description
Technical field
The present invention relates to spatial data management and searching field, particularly relate to magnanimity vector space under a kind of parallel environment
The parallel index method of data.
Background technology
Spatial data has complexity, abstractness, multi-space, polymorphism and non-structured feature, and spatial index is
Spatial data retrieval, the key shared and service, in spatial data services, the efficient index of massive spatial data is particularly important.Mesh
Before, spatial index mainly has grid index[1,2], quaternary tree[3,4](Q tree) and R tree[5,6].Wherein, grid index is mainly for grid
Lattice data, Q tree and R tree are mainly for vector data.Q tree is the class Indexing Mechanism dividing tissue index structure based on space,
Line, face collection are set up Q tree, and data redundancy is big, recall precision is low, but sets up Q tree for point set and retrieve, recall precision
High.R tree is B-tree extension in hyperspace, mainly has R tree, R+[7]、R*[8,9]Tree etc..Owing to wanting Dynamic Maintenance spatial index
Structure, the insertion cost of the method is the highest, and for massive spatial data, the establishment process of index will be the hugest
Greatly.
Along with the development of computer technology, occur in that unit multinuclear, multimachine multinuclear (many-core) and High-Performance Computing Cluster etc. are novel
Parallel environment.Under this environment, the height of spatial data index efficiency is the key weighing spatial database overall performance.Relevant
Person have studied the parallel algorithm of R tree[10-13], such as GPR tree[10], Master-client R tree[11]、Upgraded Parallel R
Tree[12]Deng, but these parallel indexes still can not break through the bottleneck that host node accesses, it is difficult to and the load balancing between solution process is asked
Topic.For this problem, the present invention utilizes multichannel round thinking and collaborative thought, designs a kind of Q tree for vector data
The parallel index method collaborative with R tree.
List of references
[1] Xiao Wei's device, Feng Yucai, Miao Yongwu. spatial object Database Grid Indexing Mechanism. Chinese journal of computers, 1994,17
(10): 45~51.
[2] Liu Yi, Gong Jianya, Guo tie up, the Resource orientation of image data streaming and Selecting research [J] under P2P environment, surveys
Paint journal, 2010,39 (4): 383~389.
[3] Hanan Samet, Robert E.Webber.Storing a collection of polygons using
Quadtrees.ACM Transactions on Graphics (TOG) .1985,4 (3): 182~222.
[4] Li Jianxun, Shen Bing, Jiang Rengui, Chen Tianqing, towards the Quadtree Spatial Index algorithm [J] of image pyramid, meter
Calculation machine engineering, 2011,37 (10): 11-13.
[5] Guttman.R-trees:A Dynamic Index Structure for Spatial
Searching.In:Proceedings of ACM SIGMOD Conference.Boston, MA, 1984.47~57.
[6] Gong Jun, Zhu Qing, Zhang Yeting, Li Xiaoming, Zhou Dongbo, the three-dimensional R tree index extended method of considering levels of detail
[J], surveys and draws journal, 2011,40 (2): 249~255.
[7] Timos K.Sellis, Nick Roussopoulos, Christos Faloutsos.The R+-Tree:A
Dynamic Index for Multi-Dimensional Objects.VLDB.1987.507~518.
[8] N.Beckmann, H.P.Kriegel, R.Schneider, B.Seeger.The R*-Tree:An
Efficient and Robust Access Method for Points and Rectangles.SIGMOD
Conference.1990.322~331.
[9] Sun Dianzhu, Li Yanrui, Zhu Changzhi, Sun Yongwei, the R*-tree node split algorithm of geometric object unified representation
[J], Central China University of Science and Technology's journal, 2010,38 (2): 55~58.
[10] Fu X., Wang D., Zheng W.GPR-tree:A global parallel index structure
For multiattribute declustering on cluster of workstations.In:Proceedings of
APDC ' 97, Shanghai, China, 1997,300~306.
[11] Sehnitzer B., Leutenegger S.T..Master-client R-trees:A new
Parallel R-tree architecture.In:Proceedings of SSDBM, Cleveland, Ohio, USA, 1999,
68~77.
[12] Lai Shu-Hua, Zhu-Feng Hua, Sun Yong-Qiang.A design of parallel R-
Tree on cluster of workstations.In:Proceedings of DNIS, Aizu, Japan, 2000,119~
133.
[13] in ripple, Hao Zhongxiao, the research [J] of distributed parallel spatial index based on DPR tree mechanism, computer technology
With development, 2010,20 (6): 39~42.
Summary of the invention
Under parallel environment, the method improving spatial index efficiency has the following aspects: reduce the height of index tree, minimizing
The load balancing of task and the load balancing of I/O between the degree of overlapping of scope, optimum indexing path, process between index tree node.For
This, the present invention sets about in terms of the division of spatial object collection, QR index construct, index storage and coordinate retrieval etc., and content includes: adopt
By multichannel partitioning, adjacent spatial object (spatial data) collection is divided to different process as far as possible;Based on spatial object
The true scope of little encirclement rectangular centre point and spatial object builds QR tree to the spatial object collection being divided to each process;To
The QR tree built carries out master-slave mode storage;The spatial object collection building QR tree is carried out coordinate retrieval, swollen including Q tree grid
Swollen, grid is rejected, data retrieval operation.
1. the multichannel Spatial-data Integration method of oriented mission equilibrium
Each spatial object calculates the passage at the central point place of its minimum area-encasing rectangle, and is divided to this passage.Each passage
The space adjacent object collection being divided to this passage is divided to different process, equal to realize the load of space data sets between process
Weighing apparatus and the load balancing of retrieval tasks.Comprise the following steps that (in as shown in Figure 1, Figure 2 shown in A flow process):
1) read spatial object collection file (such as Figure 1A), obtain the minimum area-encasing rectangle lower-left angle point of spatial object collection scope
Coordinate v (xmin, ymin), upper right corner point coordinates v (xmax, ymax), set number of channels as numchannel=m × n, calculates each
Scope a of passageij(minxij, minyj, maxxij, maxyij) (such as process B in Fig. 1), computing formula is as follows:
2) the record number { n of all passage Spatial Objects is initialized11, n12..., nmnValue be 0;
3) spatial object o is calculatediCentral point c (the x of minimum area-encasing rectanglec, ycPassage a residing for)ij(formula 5, formula 6), and will
oiIt is divided to passage aij(making the most adjacent space data sets all enter same passage).After spatial object enters passage, press
The spatial object number g of this passage of entrance is cumulative every time obtains this channel space object sum nij(formula 7);Wherein g is every sub-distribution
Enter the quantity of passage, then the spatial object entering this passage is allocated to different process (such as Fig. 1 C).
4) 3 are repeated) step, until all spatial objects all enter respective channel distribution to different processes.
2. central point based on spatial object minimum area-encasing rectangle and the QR tree constructing method of spatial object true scope
QR tree is the spatial index that Q tree is collaborative with R tree: R tree is mainly indexed by Q tree, and R tree is then to spatial data body originally
Body is retrieved.Building QR tree process is: the central point first passing through spatial object minimum area-encasing rectangle enters the leaf joint of Q tree
Point, enters back into the R tree of this leaf node to set up index.The process that realizes of QR parallel index is: each process is to allocated sky
Between data set up that the scope of Q tree in QR tree, and each process is identical, deep equality respectively.
In QR parallel index, each process builds leaf node and the structure two-dimensional array of Q tree (degree of depth is qdepth)
(2qdepth×2qdepth) each unit one_to_one corresponding, i.e. magnanimity spatial object collection is divided into some pieces of subspace object sets
(2depth×2depthIndividual).Every sub spaces object set sets up R tree respectively, optimizes index path by QR index structure.Work as institute
There is spatial object to be all inserted into QR tree, from Q leaf node to root node, the actual range of each node carried out recurrence renewal, and meter
Calculating the swell value of range of search, detailed step is as follows:
1) each process creates index to the space data sets being divided to this process: first, minimum according to spatial object
The central point of area-encasing rectangle calculates the two-dimensional array unit at its place;Secondly, according to the minimum area-encasing rectangle of spatial object it
It is inserted in the R tree in Q leaf node corresponding to two-dimensional array unit;Finally, from Q leaf node to root node recurrence more
The actual range of new Q leaf node, concrete grammar is following (as shown in Figure 2 A):
A) calculate the Q tree degree of depth (qdepth) according to different hardware environment, process number and spatial object number, and create the degree of depth and be
The linear Q tree of qdepth.Meanwhile, Dynamic Two-dimensional array object (4 is createdqdepthIndividual), and each two-dimensional array object comprises sensing
The pointer of corresponding leaf node in Q tree;
B) spatial object information comprises: ID, minimum area-encasing rectangle (Fig. 3 A) and the central point (Fig. 3 B) of minimum area-encasing rectangle.
The insertion of spatial object: first, calculates the Q leaf node that the central point place two-dimensional array unit of minimum area-encasing rectangle is corresponding
(Fig. 3 C, Fig. 3 process D), then, inserts, according to the minimum area-encasing rectangle of spatial object, the R tree (figure that Q leaf node is comprised
3D);
C) repeat step b, all spatial objects are all inserted into the R tree (Fig. 3 process E) that each leaf node of Q tree is comprised;
D) obtain the scope (such as Fig. 4 B) of R tree corresponding to each leaf node of Q tree, the scope of R tree is saved with corresponding Q leaf
The logic scope (such as Fig. 4 A) of point merges, and the new range of calculating is the actual range of Q leaf node (such as the process in Fig. 4
C), the actual range of the non-leaf nodes of Q tree and is updated the actual range of Q leaf node from leaf node recurrence;
E) traversal Q leaf node, is calculated actual range (the long rl of each leaf node by (formula 8)ij, wide rwij, such as figure
Process A in 5) and logic scope (long llij, wide lwij, such as process B in Fig. 5) and the maximum ev of ratioij, obtain all by (formula 9)
The maximum mev (such as Fig. 5 process C) of this ratio in leafy node.
2) the QR tree having built up is inserted spatial data: first, according to being inserted into spatial object minimum area-encasing rectangle
Central point obtains the R tree information that in each process of its pre-inserted, Q leaf node is comprised, relatively each process (piThese R trees in)
In object number n that comprisesj, obtain njThe ID (process number) of minimum that process (formula 10), and this object is inserted into process
By in the R tree that ID is comprised.
ID=min0≤i < mnjFormula 10
The master-slave mode storage method of 3.QR tree
Each process all comprises QR tree, in order to improve recall precision, it is to avoid need during retrieval QR tree is traveled through, this
Bright QR tree is divided into Q tree and R tree collection two parts store, during retrieval only need to according to Q tree retrieval result reading part R tree
(comprising the R tree of retrieval object).Meanwhile, the renewal to QR tree also only need to be to the part R tree comprised in QR tree and corresponding Q leaves
Child node is updated.
The QR tree that the storage of each process each creates, predominantly two parts: first, store Q tree;Then, storage depends on Q tree
R tree collection, its step is as follows:
1) storage Q tree: as unit's index of R tree, comprise the spatial object of R tree and correspondence in Q leaf nodal information
Storage position;
2) storage R tree: be each R tree file designation with the leaf node information of process ID and R tree place Q tree.
4.QR tree coordinate retrieval method
QR tree comprises a Q tree and some R trees, when retrieving massive spatial data, by QR coordinate retrieval: first lead to
Cross Q tree assistance and retrieve R tree collection, assisted the R tree collection search space data itself obtained the most again by Q tree.By collaborative, right
Q tree carries out expansive working and obtains maximum Q number leaf node collection, and the R tree obtaining comprising retrieval object by rejecting collects, thus prominent
Break root node access bottleneck.First, calculate the Q leaf set of node that each process intersects with frame retrieval, and each process is examined
The leaf node of rope Q tree expands mev (formula 10) rank, obtains the Q leaf node maximum collection that may intersect with range of search;So
After, concentrate with maximum the true scope of each leaf node to ask friendship (see Fig. 6), reject disjoint leaf node, intersected
Q leaf set of node, and then retrieval R tree corresponding to Q leaf node.Concrete steps (as shown in Fig. 2 B flow process).Retrieval
Process is as follows:
1) the Q tree (such as Fig. 6 A, I) during each process reads QR tree;
2) the selected region needing retrieval, if this regional extent is Search_Rect, as shown in a in Fig. 6 B, C, D, E, F
Region;
3) Q tree is intersected in Search_Rect with QR tree leaf node collection (LeafnodeSet, red as in Fig. 6 B are calculated
Shown in territory, zone), region as shown in a in Fig. 6 C, and search domain is carried out the expansion of mev rank (if in mev=1, Fig. 6 D 1,2,
3,4,6,7,8,9 regions show expansion area), obtain retrieving leaf node maximum collection.As in Fig. 6 I, " L " of labelling is yellow
It it is the leaf node that intersects with Search_Rect of logic scope;
4) take the actual range realrect of leaf node in leaf node collection (newLeafnodeSet), calculate
(Fig. 6 E grey area is the reality of newLeafnodeSet to the node that realrect intersects with search domain Search_Rect
Scope);Rejecting disjoint node, (Fig. 6 F grey area is the Q leaf intersected with range of search to retain crossing node
The actual range of node).As in Fig. 6 I R labeled be the leaf that all intersects with Search_Rect of logic scope, actual range
Node;
5) step 4 is repeated, until traversal leaf node maximum concentrates all of leaf node, the result wherein rejected such as figure
Region shown in (2,3,5,9) in 6G.As in Fig. 6 I " R " labeled be actual range intersect with Search_Rect leaf joint
Point.
6), after carrying out expanding, rejecting operation, read the R tree corresponding to Q leaf set of node and carry out retrieving (such as Fig. 6 H), obtain
Take retrieval result;
If expanding and obtaining all leaf nodes that newLeafnodeSet comprises Q tree, then need to travel through all of leaf node
Seek the common factor with range of search, so obtaining the computationally intensive of result by expansion.Therefore, directly start to use from Q root vertex
The actual range of node and range of search ask friendship, obtain, by tree hierarchy retrieval, the leaf node collection intersected with range of search,
Thus R tree retrieval corresponding to acquired Q leaf set of node again.
(2) beneficial effect
1, utilize the present invention, use multi-channel data division methods spatial object collection to be divided, by adjacent space pair
As being assigned to different process efficiently, it is achieved load balancing between process, improve recall precision, to millions vector target collection
Retrieval time is stable at Microsecond grade;
2, utilize the present invention, breach the bottleneck that spatial index root node accesses, by calculating the ranks of Q leaf node
Number position, uses array pointer directly to access the R tree corresponding to leafy node, improves recall precision, and recall precision is R tree
More than 2 times;
3, utilize the present invention, be some sub-block spatial object collection by Q tree by Spatial-data Integration, and child objects respectively
R tree set up by collection so that the insertion that carries out index, deletion etc. update and operate that become only need to be to the R tree corresponding to Q leaf node
It is updated, greatly improves index IO and update efficiency, comparing R tree under big data quantity, improve two orders of magnitude;
4, utilize the present invention, it is adaptable to the management of the especially vector data of the massive spatial data under New Hardware environment with
Retrieval service, can develop the management of high-performance massive spatial data accordingly, retrieve and inquire about service software.
Accompanying drawing explanation
Fig. 1 is the partition process of multi-channel data
Fig. 2 is the parallel flow process of the collaborative index of Q tree
Fig. 3 is the process that spatial data inserts Q tree
Fig. 4 is the renewal process of Q tree node scope
Fig. 5 is the calculating process of range of search expanded threshold value
Fig. 6 is the search method of spatial data
Fig. 7 is that 64 passages divide the process of spatial data in { (0,0)~(8,000,8,000) } regions
Fig. 8 is the process of the R tree collection storing Q tree and correspondence thereof in 8 processes
Fig. 9 be range of search be the retrieving of (4,500,4,500)-(4,800,4,950)
Detailed description of the invention
Use the example of QR tree parallel index.The size of spatial data scope is Rect{ (0,0), (8,000,8,000) },
The number SpatialDataSets=512,000 of spatial object collection.
1., as a example by the multichannel of 64 (8 × 8), data above collection is divided by matrix MultiChannelArray
Process as shown in Figure 7:
1) be Rect{ (0,0) scope, (8,000,8,000) } space data sets be initialized as 64 (8 × 8) individual passage
(such as Fig. 7 A), the long ChannelLength=1 of each subchannel, 000, wide ChannelWidth=1,000, initially set each
Passage has distributed data record value NumofObjects=0;
2) traversal spatial object collection, obtains the minimum area-encasing rectangle (MBR) of each object.Calculate the central point place of MBR
Passage MultiChannelArray [i] [j] (0≤i < 8,0≤j < 8), and enter this passage, and passage
NumofObjects adds up 1, such as Fig. 7 B;
3) each passage MultiChannelArray [i] [j] spatial object round in order to coming into divides
(rotation therapy) to each process, until the space data sets of passage divide complete (each process about can obtain 512,000/8=64,
000 spatial object), such as Fig. 7 C;
4) the spatial object collection that all passages obtained divide is stored in file by each process.
The data divided above by manifold Dow process are set up spatial index by Q tree degree of depth qdepth=3 in 2.QR tree, its
Process is as follows:
1) each process initialization degree of depth is 3, and scope is Rect{ (0,0), (8,000,8,000) } Q tree, and create one
Array of pointers pQArray (8 × 8) of individual its leaf node of sensing;
2) 64 the most divided blocks of data collection are set up spatial index and (64,000 spatial objects are set up rope by each process
Draw), obtain minimum area-encasing rectangle and the central point of minimum area-encasing rectangle of each spatial object, calculate minimum area-encasing rectangle center
Point place Q leaf node, inserts the R tree corresponding to Q leaf node (each process all comprises 64 R tree indexes).
3) the R tree scope that in 64 Q leaf nodes, the logic scope of each leaf node is corresponding merges, and obtains new
Scope is the actual range of Q leaf node, from 64 nodes of Q tree, actual range recurrence is updated other nodes of Q tree.
3. pair Q tree degree of depth be 3 Q index store, process is as follows:
1) each process stores its Q tree each created, and adds process ID name index file (such as Fig. 8 A), storage with engineering name
Q tree information includes: index range Rect{ (0,0), (8,000,8,000) }, Q tree degree of depth qdepth=3 and 64 leaves of Q tree joint
The R tree that some actual range information (as shown in B process in Fig. 8), each leaf node are corresponding stores position and the spatial object of R tree
Storage position (such as Fig. 8 D);
2) according to leaf node correspondence two-dimensional array ranks information name R tree storage file in the Q tree of R tree place (such as figure
8C), the storage of R tree collection is as shown in process E in Fig. 8.
4. pair index range is Rect{ (4,500,4,500), (5,000,5,000) }, the QR of maximum swelling value mev=1
The retrieving of tree is as follows:
1) upper some Point (4,500,4,500) the place Q tree node calculating frame retrieval is pArray [5] [5], separately
Outer 1 Point (4,800,4,950) is also pArray [5] [5], then location initial retrieval Q tree node is pArray [5]
[5], as shown in Figure 9 A;
2) according to maximum swelling value mev, outward expansion single order obtains retrieving maximum collection, and retrieval leaf node scope is
pArray[4][4]、pArray[4][5]、pArray[4][6]、pArray[5][4]、pArray[5][5]、pArray[5]
[6], pArray [6] [4], pArray [6] [5], pArray [6] [6], as shown in Figure 9 B;
3) true scope of each node and the common factor of range of search in calculating retrieval maximum magnitude, read crossing leaf and save
The R tree that point is corresponding is retrieved, finally retrieved set of node be pArray [5] [4], pArray [5] [5], pArray [5] [6],
PArray [6] [6] (such as Fig. 9 C), the R tree reading node corresponding carries out data retrieval.
Particular embodiments described above, has been carried out the purpose of the present invention, technical scheme and beneficial effect the most in detail
Describe in detail bright, be it should be understood that the specific embodiment that the foregoing is only the present invention, be not limited to the present invention, all
Within the spirit and principles in the present invention, any modification, equivalent substitution and improvement etc. done, should be included in the guarantor of the present invention
Within the scope of protecting.
Claims (4)
1. a vector QR tree parallel index method, it is characterised in that:
1) in QR tree construction step based on spatial object minimum area-encasing rectangle central point and spatial object true scope, towards appointing
Business equilibrium multichannel Spatial-data Integration by each passage foundation spatial object minimum area-encasing rectangle central point by adjacent
Spatial object collection is divided to different process, it is achieved the load balancing of space data sets and the task balance of retrieval between process;Based on
The QR tree of spatial object minimum area-encasing rectangle central point and spatial object true scope builds, and each Q leaf node comprises
Two scopes: logic scope, actual range, the logic scope of Q leaf node merges with corresponding R tree scope, the new model obtained
Enclose the actual range for leaf node;Traversal Q leaf node, calculates actual range and the logic scope ratio of each leaf node
The maximum of value, obtains this ratio maximum mev in all leaf nodes;
2), in QR tree master-slave mode storing step, the QR tree that the storage of each process each creates, the storage of R tree collection depends on depositing of Q tree
Storage;
3), in QR tree coordinate retrieval step, each process retrieves R tree by Q tree assistance and collects, then collects search space data by R tree
Itself.
Vector QR tree parallel index method the most according to claim 1, it is characterised in that surround square based on spatial object minimum
The QR tree construction step of shape central point and spatial object true scope comprises 2 features:
1) spatial object is abstract is the central point of minimum area-encasing rectangle, and by the Q tree in this abstract insertion QR tree, and insert QR
R tree in tree is then the minimum area-encasing rectangle of spatial object;
2) in QR tree, Q tree is linear quadtree, and has and Q leaf node two-dimensional array one to one, by this two-dimemsional number
Group goes to access the R tree that Q leaf node is comprised.
Vector QR tree parallel index method the most according to claim 1, it is characterised in that QR tree master-slave mode storing step comprises
2 features:
1) storage of Q tree comprises following information: space data sets scope, the Q tree degree of depth, Q leaf node actual range, leaf save
R tree index position, R tree that point is corresponding index corresponding space data sets position;
2) first storing Q tree, store position and the position of R tree comprised spatial object collection of R tree in Q leaf node, R tree relies on
In Q tree, it is made a look up, R tree collection and Q leaf node one_to_one corresponding.
Vector QR tree parallel index method the most according to claim 1, it is characterised in that QR tree coordinate retrieval step comprises 3
Individual feature:
1) retrieval of Q tree logic scope obtains Q leaf set of node, expands the scope of rank mev, Q leaf set of node according to Q tree
Outward expansion mev rank, obtain maximum retrieval Q leaf set of node;
2) assist R tree is retrieved by Q tree: the actual range of each leaf node in maximum retrieval Q leaf set of node
Asking friendship with range of search, if intersecting, then spatial object itself is retrieved by the R tree index reading this leaf node corresponding;
If non-intersect, then reject this Q leaf node;
3) if the range of search after Peng Zhanging is more than Q leaf range of nodes, then Q tree is begun stepping through from Q root vertex, real with node
Border scope calculates with range of search and obtains the Q leaf node intersected, and reads the R tree intersecting Q leaf node corresponding and comes sky
Between object itself retrieve.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210586420.3A CN103049554B (en) | 2012-12-31 | 2012-12-31 | A kind of vector QR tree parallel index method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210586420.3A CN103049554B (en) | 2012-12-31 | 2012-12-31 | A kind of vector QR tree parallel index method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103049554A CN103049554A (en) | 2013-04-17 |
CN103049554B true CN103049554B (en) | 2016-09-14 |
Family
ID=48062195
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201210586420.3A Expired - Fee Related CN103049554B (en) | 2012-12-31 | 2012-12-31 | A kind of vector QR tree parallel index method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103049554B (en) |
Families Citing this family (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105723365B (en) | 2013-11-19 | 2019-09-03 | 华为技术有限公司 | Method for optimum indexing, main database node and subscriber database node |
CN103679803B (en) * | 2013-12-03 | 2016-07-06 | 大连海事大学 | Sea ice dynamic modelling method in a kind of navigation simulator scene |
CN105138560A (en) * | 2015-07-23 | 2015-12-09 | 北京天耀宏图科技有限公司 | Multilevel spatial index technology based distributed space vector data management method |
CN106095920B (en) * | 2016-06-07 | 2019-06-11 | 四川大学 | Distributed index method towards extensive High dimensional space data |
CN106933511B (en) * | 2017-02-27 | 2020-02-14 | 武汉大学 | Space data storage organization method and system considering load balance and disk efficiency |
CN110110186A (en) * | 2018-01-23 | 2019-08-09 | 微软技术许可有限责任公司 | Vector index technology of the space partition tree in conjunction with neighborhood graph |
CN109284295B (en) * | 2018-10-17 | 2021-09-17 | 郑州云海信息技术有限公司 | Data optimization method and device |
CN110083598B (en) * | 2019-03-22 | 2021-05-25 | 深圳先进技术研究院 | Spark-oriented remote sensing data indexing method, system and electronic equipment |
CN111221937B (en) * | 2020-01-10 | 2023-05-09 | 江苏大学 | Method for constructing topic R tree by dynamic K value clustering |
CN111414445B (en) * | 2020-03-17 | 2024-02-06 | 周凯 | Address inverse analysis method applying geographic information |
CN112612787A (en) * | 2020-12-08 | 2021-04-06 | 深圳大学 | Big data oriented index optimization method |
-
2012
- 2012-12-31 CN CN201210586420.3A patent/CN103049554B/en not_active Expired - Fee Related
Non-Patent Citations (5)
Title |
---|
A Design of Parallel R-tree on Cluster of Workstations;Shuhua Lai et al;《DNIS 2000》;20001231;第LNCS 1966卷;全文 * |
Design and Implementation of QR+Tree Index Algorithms;Mao Huaqing et al;《International Conference on Wireless Communications, Networking and Mobile Computing》;20070925;全文 * |
QR-TREE: A HYBRID SPATIAL INDEX STRUCTURE;YU-CHEN FU et al;《Proceedings of the Second International Conference on Machine Learning and Cybernetics》;20031105;全文 * |
Research on Distributed Hilbert R tree Spatial Index Based on BIRCH Clustering;Yizhou Yang et al;《2012 20th International Conference on Geoinformatics》;20120617;全文 * |
孙晓明 等.基于QR树索引的面包含关系求解算法.《计算机工程》.2010,第36卷(第21期),全文. * |
Also Published As
Publication number | Publication date |
---|---|
CN103049554A (en) | 2013-04-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103049554B (en) | A kind of vector QR tree parallel index method | |
Zhu et al. | An efficient 3D R-tree spatial index method for virtual geographic environments | |
Hsu et al. | Key formulation schemes for spatial index in cloud data managements | |
CN111339236B (en) | Construction method and device for multi-source space database of power transmission line corridor | |
CN109492060A (en) | A kind of map tile storage method based on MBTiles | |
Wan et al. | Column store for GWAC: a high-cadence, high-density, large-scale astronomical light curve pipeline and distributed shared-nothing database | |
Wang et al. | Parallel trajectory search based on distributed index | |
Wang et al. | Efficient visibility analysis for massive observers | |
CN103714124B (en) | Ultra-large-scale low-voltage data processing method | |
Alam et al. | A performance study of big spatial data systems | |
Zhang et al. | High-performance online spatial and temporal aggregations on multi-core cpus and many-core gpus | |
Lin et al. | A spark-based high performance computational approach for simulating typhoon wind fields | |
Xie et al. | On massive spatial data retrieval based on spark | |
CN103106254B (en) | The parallel joining method of polygon vector data file | |
Liu et al. | Analysis of spatial indexing mechanism and its application in data management: A case study on spatialite database | |
Wang et al. | Distributed retrieval for massive remote sensing image metadata on spark | |
CN105654137A (en) | Mass remote sensing data classification and organization method | |
Xia et al. | Remote Sensing Image Data storage and search method based on pyramid model in cloud | |
Tian et al. | A prototype spatio-temporal database built on top of relational database | |
Wang et al. | HBase storage schemas for massive spatial vector data | |
CN115408382A (en) | Vector space information gridding storage management method for large-scale satellite data | |
Ding et al. | RDB-KV: A cloud database framework for managing massive heterogeneous sensor stream data | |
Duan et al. | Assessment of MongoDB's spatial retrieval performance | |
Stojanovic et al. | A hybrid MPI+ OpenMP application for processing big trajectory data | |
Teng et al. | GPU-based Real-time Contact Tracing at Scale |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20160914 Termination date: 20161231 |