department of computer science spatio-temporal histograms hicham g. elmongui*mohamed f. mokbel +...
TRANSCRIPT
Department of Computer Science
Spatio-Temporal Histograms
Hicham G. Elmongui* Mohamed F. Mokbel+ Walid G. Aref*
*Purdue University, Department of Computer Science+University of Minnesota, Department of Computer Science
SSTD’05 Hicham G. Elmongui
2
Motivation
Infrastructure for keeping track and answering continuous queries on moving objects– Moving Queries / Moving Objects– Stationary Queries / Moving Objects– Moving Queries / Stationary Objects– Range Queries, KNN, …
Spatio-TemporalDatabase Server
SSTD’05 Hicham G. Elmongui
3
Motivation
Spatio-TemporalDatabase Server
How many cars on this freeway? Where is my nearest McDonald’s?
SSTD’05 Hicham G. Elmongui
4
Motivation
SELECT M.IDFROM MovingObjects MWHERE M.Type = “Truck”INSIDE Area A;We cannot collect statistics statically
(e.g. histograms) and answer queries efficiently over an extended period of
time
SSTD’05 Hicham G. Elmongui
5
Motivation
Go
to w
ork
Ret
urn
hom
e
Lu
nch
hour
0
0.2
0.4
0.6
0.8
1
12:0
0 A
M
1:0
0 A
M
2:0
0 A
M
3:0
0 A
M
4:0
0 A
M
5:0
0 A
M
6:0
0 A
M
7:0
0 A
M
8:0
0 A
M
9:0
0 A
M
10:0
0 A
M
11:0
0 A
M
12:0
0 P
M
1:0
0 P
M
2:0
0 P
M
3:0
0 P
M
4:0
0 P
M
5:0
0 P
M
6:0
0 P
M
7:0
0 P
M
8:0
0 P
M
9:0
0 P
M
10:0
0 P
M
11:0
0 P
M
12:0
0 A
M
Downtown
A residential area
Not just time makes a
difference, but also space makes
a difference
Nor
mal
ized
Fre
qu
ency
SSTD’05 Hicham G. Elmongui
6
ST-Histograms
Histograms aware of the underlying
space and time dimensions
SSTD’05 Hicham G. Elmongui
7
System Architecture
Query Plan
feedbackQuery Executor
Query Optimizer
ST-Histogram Manager
Continuous Query
Dat
a
SSTD’05 Hicham G. Elmongui
8
Queries as Light Spots
6.25%
6.25%
6.25%
6.25%
6.25%
6.25%
6.25%
6.25%
6.25%
6.25%
6.25%
6.25%
6.25%
6.25%
6.25%
6.25%
SSTD’05 Hicham G. Elmongui
9
Queries as Light Spots
6.98%
6.98%
6.98%
6.98%
6.25%
6.25%
6.01%
6.01%
6.25%
6.25%
6.01%
6.01%
6.01%
6.01%
6.01%
6.01%
6.01%
6.01%
6.01%
6.01%
Q1
6.25%
6.25%
6.25%
6.25%
6.25%
6.25%
6.25%
6.25%
6.25%
6.25%
6.25%
6.25%
10%
SSTD’05 Hicham G. Elmongui
10
Queries as Light Spots
6.15%
6.15%
6.15%
6.15%
15.04% 9.84%
5.05%
5.05%
5.05%
5.05%
5.05%
5.05%
5.05%
6.01%
5.05%
5.05%
5.05%
6.01%
Q2
6.01%
6.01%
6.01%
6.01%
6.01%
6.01%
6.01%
6.01%
6.01%
6.01%
6.98%
6.98%
6.98%
6.98%
Q1
20%
SSTD’05 Hicham G. Elmongui
11
15.04% 9.84%15.04% 9.84%
Queries as Light Spots
6.15%
6.15%
6.15%
6.15%
5.05% 5.05%5.05%
5.05%
5.05%
5.05%
5.05%
5.05%
5.05%
5.05%
5.05%
5.05%
Q1
Q2
SSTD’05 Hicham G. Elmongui
12
Queries as Light Spots
6.29%
6.29%
6.29%
6.29%
4.22%
15.51%
3.24%
10.15%
5.21%
5.21%
5.21%
5.21%
5.21%
5.21%
5.21%
5.21% 1%
5.05% 5.05%Q2
15.04% 9.84%
5.05%
5.05%
5.05%
5.05%
5.05%
5.05%
5.05%
5.05%
6.15%
6.15%
6.15%
6.15%
Q1
SSTD’05 Hicham G. Elmongui
13
Features of ST-Histograms
No computing capabilities assumed for the moving objects– Moving objects update their location periodically with the spatio-
temporal database server
No patterns assumed for queries– Queries come and go anytime anywhere
Diskless spatio-temporal stream database serverEnable for adaptive query processing
SSTD’05 Hicham G. Elmongui
14
ST-Histogram Construction/Refining
Initially
Selectivity of a query
Rate of a query to a grid cell
SSTD’05 Hicham G. Elmongui
15
Experiments – Data Sets
Network-based Generator of Moving Objects (SSDBM’00, GeoInformatica’02)
Map of Greater Lafayette AreaEvery MO updates its location every 10 sec
SSTD’05 Hicham G. Elmongui
16
Estimation Relative Error vs. Query Size
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0.25% 0.50% 0.75% 1% 4%
Query Size
Ave
rag
e R
elat
ive
Err
or
SSTD’05 Hicham G. Elmongui
17
ST-Histogram’s Stability
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
50%
0 5 10 15 20 25 30 35 40
Time
Av
era
ge
Re
lati
ve
Err
or
0.75%1%4%
SSTD’05 Hicham G. Elmongui
18
ST-Histogram vs. Random Sampling
0
0.1
0.2
0.3
0.4
0.5
0.6
RS(10%) RS(25%) RS(50%) RS(75%) ST-Histogram
Ave
rag
e R
elat
ive
Err
or
SSTD’05 Hicham G. Elmongui
19
Related Work
Spatio-temporal histograms– Choi and Chung (SIGMOD’02)– Tao et al (ICDE’03)– Marios et al (SSDBM’03)
Sampling– Random Sampling– Venn Sampling (ICDS’05)
SSTD’05 Hicham G. Elmongui
20
Conclusion
Aware of the underlying space and time dimensionsImplemented in PLACE (a spatio-temporal database server)Efficient for spatio-temporal streaming applicationsRefined upon feedback from query executorUsed in an online/offline modeAccommodate periodicity in moving objects’ behaviorEnable adaptive query processingAverage relative error 8% for practical query sizes
SSTD’05 Hicham G. Elmongui
21
The END
Thank You