Topk Monitoring in Wireless Sensor Networks

1
Top-k Monitoring in Wireless Sensor Networks

• Minji Wu, Jianliang Xu, Xueyan Tang, and
  Wang-Chien Lee
• IEEE TRANSACTIONS ON KNOWLEDGE AND DATA
  ENGINEERING, VOL. 19, NO. 7, JULY 2007

2
Outline

• Introduction
• Filter-based Monitoring Approach
• FILA Overview
• Query Reevaluation
• Filter Setting (Uniform versus Skewed)
• Filter Update (Eager versus Lazy)
• Performance Study
• Simulation Setup
• Eager versus Lazy Filter Update
• Performance Comparison against TAG and Range
  Caching
• Conclusions

3
Introduction

• Top-k Query
• Environmental Monitoring
• A top-k query is issued to find out the nodes and
  their corresponding areas with the highest
  pollution indexes for the purpose of pollution
  control or research study.
• Network Management
• A top-k query may be issued to continuously
  monitor the sensor nodes with the least residual
  energy.

4
Introduction

• In traditional database systems
• Focused on snapshot top-k queries
• This paper focuses on continuously monitoring
  top-k queries in sensor networks.
• Utilize previous top-k result to obtain a new
  top-k result.

5
Top-1 query

• TAG (S. Madden et al. , OSDI 02)

BS
51
56
52
t1
35
A total of nine messages are sent
t2
38
C
t3
37
43
51
45
56
48
52
A
B
t1
t1
43
51
t2
t2
45
56
t3
t3
52
48
6
Top-1 query

• Range Caching (C. Olston et al., SIGMOD01)

BS
t1
35
48
A total of four messages are sent
t2
38
C
t3
37
20, 39
52
48
39, 47
47, 80
A
B
t1
t1
43
51
t2
t2
45
56
t3
t3
52
48
7
Problem Definition

• Consider a top-k monitoring query that
  continuously requests the (ordered) list of
  sensor nodes R with the highest readings, that is

8
FILA Overview

• (1) Filter Setting
• the base station computes a filter li, ui for
  each sensor node i and sends it to the node for
  installation.
• (2) Query Reevaluation
• (3) Filter update

9
Query Reevaluation

• Sensor-initiated updates
• (1) Internal update
• (2) Join update
• (3) Leave update

Leave update
Internal update
Join update
Critical bound
10
A Simple Case

• Consider a simple case where only one
  sensor-initiated update is received by the base
  station

Only n1 needs to be probed
11
A Simple Case
Only the sensor nodes whose current readings are
higher than v2 respond to the probe
12
General Cases

• Tinternal the set of internal updates
• Tjoin the set of join updates
• Tleave the set of leave updates
• T the old top-k set
• If T' T - Tleave Tjoin ? k
• the new top-k set must be a subset of T'
• Otherwise, if T' lt k
• the nodes that are not in T' have to be probed.

13
An Example of Top-3 Monitoring
14
Another Example of Top-3 Monitoring
15
(No Transcript)
16
Filter Setting

• Uniform filter setting
• It is simple and favorable when the readings of
  all sensor nodes follow a similar changing
  pattern.

17
Filter Setting

• Skewed filter setting
• taking into account the changing patterns of
  sensor readings.
• Suppose the average time for the reading of node
  i to change beyond is fi(?)
• 1/fi(?) the rate of sensor-initiated updates by
  node i

18
Filter Setting

• We let every node measure the average delta
  change di of their sensor readings at a fixed
  rate.
• Skewed filter setting

19
Filter Update

• Eager filter update
• If a new filtering window li', ui' is different
  from the old one li, ui then the new filter
  li', ui' is immediately sent to node i
• Lazy filter update
• If a new filtering window li', ui' fully
  contains the old one li, ui, that is, li',
  ui' ? li, ui then the base station delays the
  filter update until node is reading violates the
  old filter li, ui.

20
Performance Study

• Simulation Setup
• Energy cost in transmitting a message
• s message size
• ? distance-independent term (50 nj/b)
• ? coefficient (100 pj/b/m2)
• q distance-dependent term ( 2)
• d distance
• Energy cost in receiving a message
• ? is set at 50 nJ/b

21
Performance Study

• A Sensor initiated update message
• Sensor ID 4 bytes
• Sensor Reading 4 bytes
• A filtering window is characterized by 8 bytes.

22
Network Layouts
23
Real Data Traces

• Simulated using the real traces provided by the
  Live from Earth and Mars (LEM) project at the
  University of Washington.
• Two kinds of sensor readings are used
• temperature (TEMP)
• Dew point (DEW)
• logged by the station at the University of
  Washington from August 2004 to August 2005
• Total 500000 sensor readings
• Extract many subtraces starting at different
  dates
• Each subtrace contains 20000 readings
• The subtraces were used to simulate the physical
  phenomena in the immediate surroundings of
  different sensor nodes.

24
Real Data Traces
25
Evaluation Metrics

• Network Lifetime
• the network lifetime is defined as the time
  duration before the first sensor node runs out of
  power.
• Average Energy Consumption
• It is defined as the average amount of energy
  consumed by a sensor node per time unit.
• Monitoring Accuracy
• This is defined as the mean accuracy of monitored
  results against the real results.

26
Eager versus Lazy Filter Update(multihop, k 10)
Average energy consumption.
Network lifetime.
27
Eager versus Lazy Filter Update
Energy consumption by layer
28
Performance Comparison against TAG and Range
Caching(single hop, k 3)
Average energy consumption.
Network lifetime.
29
Performance Comparison against TAG and Range
Caching (single hop, k 3)
Monitoring accuracy
30
Performance Comparison against TAG and Range
Caching(Multihop, k 10)
Average energy consumption.
Network lifetime.
31
Performance Comparison against TAG and Range
Caching(Multihop, k 10)
Monitoring accuracy
32
Conclusion

• This paper exploited the semantics of top-k query
  and proposed a novel energy-efficient monitoring
  approach called FILA.
• Two filter setting algorithms (that is, uniform
  and skewed) and two filter update strategies
  (that is, eager and lazy) have been proposed.

33
Filter Setting

• Under random walk model

0.5
0.5
l
The average time for the reading to change beyond
? can be expressed as