Spatial Query Processing for High Resolutions

1
Spatial Query Processing for High Resolutions
Hans-Peter Kriegel, Martin Pfeifle, Marco Pötke,
Thomas Seidl
Database Group
8th International DASFAA-Conference 26 - 28
March, 2003, Kyoto, Japan
Institute for Computer Science University of
Munich, Germany
2
Outline of the Talk
1.) Introduction 2.) RI-Tree 3.)
HRI-Approach 4.) Experimental Evaluation 5.)
Conclusions
3
Outline of the Talk
1.) Introduction 2.) RI-Tree 3.)
HRI-Approach 4.) Experimental Evaluation 5.)
Conclusions

• ? Spatial Databases
• Voxelized Objects
• High Resolutions

4
Spatial Database Management Systems

• System Requirements
• ? Effectivity
• ? Efficiency
• ? Scalability
• Concurrency Control
• Recovery

complex
spatial objects
(based on extensible ORDBMS)
5
Voxelized Spatial Objects
6
Query Processing for High Resolutions
O R D B M S
High Resolution
Spatial - DB
Filter - Step
e.g. RI-Tree
HRI-Approach
Candidate Set for Application Specific
Refinement Step
C A D
Refinement Step
Result
7
Outline of the Talk

• Relational Interval Tree
• VLDB 00
• Extensible Indexing

1.) Introduction 2.) RI-Tree 3.)
HRI-Approach 4.) Experimental Evaluation 5.)
Conclusions
8
Relational Interval Tree (RI-Tree)
? Foundation Interval Tree Edelsbrunner 1980
9
Relational Interval Tree (RI-Tree)
root 2h1
? Foundation Interval Tree Edelsbrunner 1980
? 1. Idea Virtualization of the Primary Structure
10
Relational Interval Tree (RI-Tree)
? Foundation Interval Tree Edelsbrunner 1980
? 1. Idea Virtualization of the Primary Structure
11
RI-Tree Intersection Query
1. Procedural Step
2. Declarative Step
12
RI-Tree Intersection Query
1. Procedural Step
? arithmetic traversal of the primary
structure ? collecting the visited nodes in
transient tables ? number of I/0-accesses 0
2. Declarative Step
posting one single
SQL-statement number of
I/O-accesses O(hlogbn r/b)
13
RI-Tree Integration into an ORDBMS
Declarative Embedding Object-relational DML and
DDL
Extensible Optimization Framework Object-relationa
l interface for selectivity estimation and cost
prediction functions.
Extensible Indexing Framework Object-relational
interface for index maintenance and querying
functions.
User-defined Index Structure VLDB 00 SSTD 01
User-defined Cost Model SSDBM 02
Relational Implementation Mapping to built-in
indexes (B-trees) SQL-based query processing
Relational Implementation Mapping to built-in
statistics facilities SQL-based evaluation of
cost model
14
Outline of the Talk
1.) Introduction 2.) RI-Tree 3.)
HRI-Approach 4.) Experimental Evaluation 5.)
Conclusions

• Grey Intervals
• Storage of Grey
• Intervals in an
• ORDBMS
• Intersection Queries
• based on
• Grey Intervals

15
Grey Intervals
A voxelized real-world object
16
Storage of a Grey Object Interval Sequence
17
Multi-Step Query Processing for Intersection
Queries
result set
DB
A
C
B

A1 Q1
A3 Q2
A2 Q1

A1 Q1
B3 Q1
A3 Q2
B2 Q1

?

B3 Q1
B3 Q1
B2 Q1
C1 Q1
C2 Q1
C1 Q1
D1 Q1
Query Q
18
2. filter step FAST-GREY-TEST

• intersection test based on the
• aggregated information of the grey
  intervals.

intersection
no intersection
19
3. filter step BLOB-TEST

• intersection test based on the
• examination of the black interval sequence

...
...
I1
I2
runtime analysis
bit-oriented approach

• finding the starting point A
• of the interlacing area

O (1)

• testing the
• L voxels

O (L)
20
3. filter step BLOB-TEST

• intersection test based on the
• examination of the black interval sequence

A
L
...
...
w7
w1 w2 w3 w4 w5 w6 w7 w8
w6
w4
I1
n1 8 nL1 1 nL2 3
I2
runtime analysis
offset-oriented approach

• finding the starting point
• of the interlacing area

O (log n1)

• testing the
• nL1 resp. nL2 intervals

O (nL1nL2)
21
SQL-Statement
SELECT candidates.id FROM ( SELECT db.id AS
id, table (AggInfos(db.intervalsequence,
q.intervalsequence)) AS ctable FROM
GreyIntervals db, GreyQueryIntervals q WHERE
intersects (hull(db.intervalsequence),
hull(q.intervalsequence)) GROUP BY
db.id ) candidates WHERE EXISTS ( SELECT 1
FROM GreyIntervals db, GreyQueryIntervals q,
candidates.ctable ctable WHERE db.rowid
ctable.dbrowid AND q.rowid ctable.qrowid
AND blobintersection (db.intervalsequence,
q.intervalsequence)                  )
table (AggInfos(db.intervalsequence,
q.intervalsequence))
intersects (hull(db.intervalsequence),
hull(q.intervalsequence))
blobintersection (db.intervalsequence,
q.intervalsequence)
22
Outline of the Talk
1.) Introduction 2.) RI-Tree 3.)
HRI-Approach 4.) Experimental Evaluation 5.)
Conclusions
23
Experimental Evaluation
CAR
PLANE
approx. 200 parts approx. 10.000
parts approx. 7 million intervals approx.
9 million intervals resolution 33 bit (0 ..
8.589.934.591) resolution 42 bit (0 ..
4.398.046.511.103)

• Examination of the HRI approach based on
  different
• MAXGAP Parameters 10 100 1,000 10,000
  100,000 1,000,000
• Comparison between the HRI approach
• and the spatial variant of the RI-tree
  SSTD 01

24
Experiments
25
Experiments Secondary Storage
(RI-Tree)

• With the HRI method we can improve the storage
  requirement
• by an order of magnitude.

26
Experiments Runtime for collision queries
CAR
PLANE
27
Experiments Runtime for collision queries
CAR
RI-tree
PLANE
RI-tree
28
Outline of the Talk
1.) Introduction 2.) RI-Tree 3.)
HRI-Approach 4.) Experimental Evaluation 5.)
Conclusions
29
Conclusions

• the HRI approach is a multi-step index structure
• suitable for spatial query processing for
  high resolutions

? good secondary storage utilization
? small main memory footprint
? improved query response time behaviour
30

?
?
?
Any questions?
?
?
?
?