Spatial Queries in Wireless Broadcast Systems

1
Spatial Queries in Wireless Broadcast Systems
Authors Baihua Zheng, Wang-chien Lee, and Dik
Lun Lee, 2004
Presented By Qifeng Lu
2
Agenda

• Motivation
• Introduction
• Related work
• Hilbert-curve index structure
• Windows query
• KNN query
• Partition based Hilbert-curve index structure
• Cost model
• Performance evaluation
• Conclusion
• Critique

3
Motivation

• As a mobile user equipped with a GPS receiver
  walking in 5 feet/second in an ad hoc network
  with 1000 mobile nodes, one may want to know the
  nearest restaurant. How to process this query
  efficiently, timely, and accurately?
• How to deal with scalability and mobility in the
  mobile computing environment to support location
  dependent spatial queries (including window
  queries and kNN queries)?

4
Introduction

• Mobile computing
• Scalability
• More and more applications and users involved
• Mobility
• Frequent topology changes in an ad hoc network
• Context awareness
• Capability to recognize and react to the real
  world context
• Location-awareness
• Spatial queries
• Location-Dependent Spatial Queries (LDSQ)
• KNN and Window Queries

5
Deal with Mobility!
Restaurants
Mobile User at the source location
Mobile user at the destination
6
Agenda

• Motivation
• Introduction
• Related work
• Hilbert-curve index structure
• Windows query
• KNN query
• Partition based Hilbert-curve index structure
• Cost model
• Performance evaluation
• Conclusion
• Critique

7
Related work

• Wireless Broadcast
• Packet flooding
• Highly scalable to serve a huge client base
• Smart (adaptive) wireless broadcast
• Server decides on the content of broadcasts
  dynamically, in response to client mobility and
  demand patterns

8
Related work

• Wireless broadcast with interleaving technique
• Index information broadcast along with objects
• Assist mobile users filter out unwanted
  information during query processing and reduce
  power consumption

9
(1,m) Interleaving
10
Data Retrieval from the wireless broadcast channel

• Each index segment contains a full index over all
  data objects
• LOCAL location-dependent data retrieval without
  submitting the location information to the
  server!
• Initial probe
• The client tunes into the broadcast channel and
  determines when the next index will be broadcast.
  
• It then turns into the power saving mode until
  the next index arrives.
• Index search
• The client searches the index
• It follows a sequence of index nodes (by
  selectively tuning into the broadcast channel) to
  locate the desired data objects to determine when
  to tune into the broadcast channel to receive
  them. It waits for the arrival of the data in the
  power saving mode.
• Data retrieval
• The client tunes into the channel when the
  desired data arrives and downloads the data

11
On Air Index

• Index is on air
• Index available to clients only when it is
  currently being broadcast
• Index for traditional databases available at
  anytime
• The performance of data retrieval algorithm at
  each client depends on the broadcast sequence!

• What happened if you visit R2 first, then R1?

12
Agenda

• Motivation
• Introduction
• Related work
• Hilbert-curve index structure
• Windows query
• KNN query
• Partition based Hilbert-curve index structure
• Cost model
• Performance evaluation
• Conclusion
• Critique

13
Hilbert Curve

• Mapping multi-dimensional space to a one
  dimensional space

14
Window Query
Claim 1. For a given window, the point p inside
the query window that has the largest
Hilbert-curve index value must be lying on the
bounding box of the query window.
Proof Step 1 Assume that there is a point q
inside the query window which has a larger index
value than p. Since the Hilbert curve is a
continuous path to visit every point in the
search space, there must be a point r outside of
the query window, having a larger index value
than q Step 2 Considering a line connecting q
and r, it must intersect the bounding box on
point b. Since the index values of the points on
the Hilbert curve are monotonously increasing,
the index value of b which is between q and r
must be larger than that of q. Consequently, the
index value of b is larger than p, which has the
biggest value according to our statement. Hence,
the previous assumption fails and the point p
having the largest value should be on the
bounding box. Step 3 Similarly, the smallest
15
Window query
16
KNN query
17
Partitioned Hilbert Curve
18
Cost Model
19
Agenda

• Motivation
• Introduction
• Related work
• Hilbert-curve index structure
• Windows query
• KNN query
• Partition based Hilbert-curve index structure
• Cost model
• Performance evaluation
• Conclusion
• Critique

20
Performance Evaluation

• Data set
• Uniform distributed data set (a)
• Cities and villages of Greece (b)

21
Performance Evaluation
22
Performance Evaluation
23
Conclusion

• A new research direction of provisioning spatial
  information and supporting spatial queries in the
  wireless data broadcast systems is identified and
  studied.
• A new index structure based on the Hilbert curve
  is proposed.
• A cost model is developed to measure the
  performance of the proposed index and to provide
  technical insights.
• A simulation is conducted to compare our proposal
  with state-of-the-art indexes, using both
  synthetical data and real data

24
Critique

• Access latency is longer for the partitioned
  Hilbert Curve due to the extra indices for the
  sub grids
• It assumes there is a centralized server with
  full real time knowledge of all the objects. In
  the real world, however, the coverage of a server
  is limited and delay exists to get the new
  location of a moving object

25
Questions ????