a dynamic query-tree energy balancing protocol for sensor networks h. yang, f. ye, and b. sikdar...
DESCRIPTION
Introduction -background- Sensor network Use Querying to collect information across the whole sensor network. A set of sensors is specifically asked to report information of interest. Query Sink Sensor networkTRANSCRIPT
A Dynamic Query-tree Energy Balancing Protocol for Sensor Networks
H. Yang, F. Ye, and B. SikdarDepartment of Electrical, Computer and systems Engin
eering
IEEE WCNC 2004Speaker: Hao-Chun Sun
Outline
Introduction Energy Consumption Model Dynamic Query-tree Energy Balancing
Algorithm (DQEB) Simulation Results Conclusion
Introduction -background-
Sensor network Use Querying to collect information across the
whole sensor network. A set of sensors is specifically asked to report
information of interest.
Query
Sink
Sensor network
Introduction -background-
Basic Sensor Network Query Procedure flooding
Introduction -background-
Sensor network characteristics Limited power Lower processing ability Limited bandwidth Smaller memory
Introduction -background-
Optimal Sensor Network Query Procedure Broadcast Tree Protocols Minimum energy broadcast tree problem is
NP-complete. Many approximate algorithms is proposed.
Introduction -motivation-
These protocols are useful only for the development of static trees.
These protocols are un-weighted protocols whose assumptions are valid only when the nodes are first deployed.
Query Broadcast Tree
Non-leaf Nodes
Leaf Nodes
Sink
Introduction -motivation-
Focus on Developing techniques to improve the
querying procedure to minimize energy consumption and maximize the sensor network lifetime.
Distributed the broadcast load evenly on nodes so that the energy distribution is balanced.
DQEB update the query tree structure to avoid uneven energy depletion.
Energy Consumption Model
Sensor network query tree Sink Non-leaf nodes Leaf nodes
Query Broadcast Tree
Non-leaf Nodes(Receive, Forward)
Leaf Nodes(Receive)
Sink
Energy Consumption Model -weight-
Every node is associated with a weight. ω: [0, 1], nodes weight is initialized to 0. , β is the power attenuation factor and
determines the rate at which depleting power affects the weight of a node.
P is its remaining battery lifetime. Desire to increase the weight faster as the
remaining battery becomes lower.
)1( P
Energy Consumption Model -energy cost-
The energy cost of broadcast depends on the number of leaf and non-leaf nodes in query tree as well as the amount of remaining batter power at a node.
, Tx power=λ× Rx power (γ)
iLiLi
iC
1
LLi Li
iiC )(
Minimum
Dynamic Query-tree Energy Balancing Algorithm (DQEB) Assumptions
A cluster structure Only involves the cluster heads
Nodes have uniform hardware, software and battery capacity.
Nodes are energy-aware. A query tree is assumed to exist with the sink
node as the root and all nodes in the network being either leaf or non-leaf nodes of the tree.
Dynamic Query-tree Energy Balancing Algorithm (DQEB) Overview
UC triggers DQEB algorithm. Neighborhood Information Synchronization (NIS) algorithm Designated parents (DP) selection from alternative parents (AP)
9(0.1) 1(0.1)
7(0.3)8(0.3)
2(0.3)
3(0.5)
4(0.2) 6(0.4)
5(0.6)Update
Coordinator(UC)
Dynamic Query-tree Energy Balancing Algorithm (DQEB) Neighborhood Information Synchronization
Algorithm Neighborhood Information Table (NIT)
State information of neighbors ID, route to root, state,….
Periodic “Hello” message to detect node failure.
If a node’s state information changes in the interval between hello messages, the hello message is substituted by the changed information.
Dynamic Query-tree Energy Balancing Algorithm (DQEB) A Greedy algorithm for parent selection
Set-Covering problem
i
5
4
3
14
7
6
8
151
9
10
11
122
13C={3,4,5,7}A=A1={2,4,9}A2={3,4}A3={6}A4={6,8}
m
iAi
1
(w9/d9)
(w2/d2)(w3/d3)
(w4/d4)
(w6/d6)
(w7/d7)
(w8/d8)
K={ }C={3,4,5,7}A={2,3,4,6,8,9}
(w5/d5)
Dynamic Query-tree Energy Balancing Algorithm (DQEB) A Greedy algorithm for parent selection
Set-Covering problem
i
5
4
3
14
7
6
8
151
9
10
11
122
13C={3,4,5,7}A=A1={2,4,9}A2={3,4}A3={6}A4={6,8}
m
iAi
1
(w9/d9)
(w2/d2)(w3/d3)
(w4/d4)
(w6/d6)
(w7/d7)
(w8/d8)
K={9}C={3,4,5,7}A={2,3,4,6,8,99}
(w5/d5)
Dynamic Query-tree Energy Balancing Algorithm (DQEB) A Greedy algorithm for parent selection
Set-Covering problem
i
5
4
3
14
7
6
8
151
9
10
11
122
13C={3,4,5,7}A=A1={2,4,9}A2={3,4}A3={6}A4={6,8}
m
iAi
1
(w9/d9)
(w2/d2)(w3/d3)
(w4/d4)
(w6/d6)
(w7/d7)
(w8/d8)
K={9,8}C={3,4,5,7}A={2,3,4,6,88,99}
(w5/d5)
Dynamic Query-tree Energy Balancing Algorithm (DQEB) A Greedy algorithm for parent selection
Set-Covering problem
i
5
4
3
14
7
6
8
151
9
10
11
122
13C={3,4,5,7}A=A1={2,4,9}A2={3,4}A3={6}A4={6,8}
m
iAi
1
(w9/d9)
(w2/d2)(w3/d3)
(w4/d4)
(w6/d6)
(w7/d7)
(w8/d8)
K={9,8,6}C={3,4,5,7}A={2,3,4,66,88,99}
(w5/d5)
Dynamic Query-tree Energy Balancing Algorithm (DQEB) A Greedy algorithm for parent selection
Set-Covering problem
i
5
4
3
14
7
6
8
151
9
10
11
122
13C={3,4,5,7}A=A1={2,4,9}A2={3,4}A3={6}A4={6,8}
m
iAi
1
(w9/d9)
(w2/d2)(w3/d3)
(w4/d4)
(w6/d6)
(w7/d7)
(w8/d8)
K={9,8,6,3}C={3,4,5,7}A={2,33,4,66,88,99}
(w5/d5)DP={9,8,6,3}
Dynamic Query-tree Energy Balancing Algorithm (DQEB) A Greedy algorithm for parent selection
Set-Covering problem
i
5
4
3
14
7
6
8
151
9
10
11
122
13C={3,4,5,7}A=A1={2,4,9}A2={3,4}A3={6}A4={6,8}
m
iAi
1
(w9/d9)
(w2/d2)(w3/d3)
(w4/d4)
(w6/d6)
(w7/d7)
(w8/d8)
K={9,8,6,3}C={3,4,5,7}A={2,33,4,66,88,99}
(w5/d5)DP={9,8,6,3}
Dynamic Query-tree Energy Balancing Algorithm (DQEB) A Greedy algorithm for parent selection
Disconnected Problem
i
5
4
3
14
7
6
8
151
9
10
11
122
13C={3,4,5,7}A=A1={2,4,9}A2={3,4}A3={6}A4={6,8}
m
iAi
1
(w9/d9)
(w2/d2)(w3/d3)
(w4/d4)
(w6/d6)
(w7/d7)
(w8/d8)
K={9,6}C={3,4,5,7}A={2,3,4,66,8,99}
(w5/d5)
Dynamic Query-tree Energy Balancing Algorithm (DQEB) A Greedy algorithm for parent selection
Disconnected Problem
i
5
4
3
14
7
6
8
151
9
10
11
122
13C={3,4,5,7}A=A1={2,4,9}A2={3,4}A3={6}A4={6,8}
m
iAi
1
(w9/d9)
(w2/d2)(w3/d3)
(w4/d4)
(w6/d6)
(w7/d7)
(w8/d8)
K={9,6}C={3,4,5,7}A={2,3,4,66,8,99}
(w5/d5)DP={9,6,3}
Dynamic Query-tree Energy Balancing Algorithm (DQEB) Differentiating APs
Sibling AP Offspring AP Independent AP
i
5
4
3
14
7
6
8
151
9
10
11
122
13(w9/d9)
(w2/d2)(w3/d3)
(w4/d4)
(w6/d6)
(w8/d8)
(w5/d5)Sibling AP
Offspring AP
Independent AP
Dynamic Query-tree Energy Balancing Algorithm (DQEB) A Greedy algorithm for parent selection
Disconnected Problem
i
5
4
3
14
7
6
8
151
9
10
11
122
13C={3,4,5,7}A=A1={2,4,9}A2={3,4}A3={6}A4={6,8}
m
iAi
1
(w9/d9)
(w2/d2)(w3/d3)
(w4/d4)
(w6/d6)
(w7/d7)
(w8/d8)
K={ }C={3,4,5,7}A={2,3,4,6,8,9}
(w5/d5)C={3,4,5,6,7}A={2,8,9}
Dynamic Query-tree Energy Balancing Algorithm (DQEB) A Greedy algorithm for parent selection
Disconnected Problem
i
5
4
3
14
7
6
8
151
9
10
11
122
13C={3,4,5,7}A=A1={2,4,9}A2={3,4}A3={6}A4={6,8}
m
iAi
1
(w9/d9)
(w2/d2)(w3/d3)
(w4/d4)
(w6/d6)
(w7/d7)
(w8/d8)
K={ }C={3,4,5,6,7}A={2,8,9}
(w5/d5)
Dynamic Query-tree Energy Balancing Algorithm (DQEB) A Greedy algorithm for parent selection
Disconnected Problem
i
5
4
3
14
7
6
8
151
9
10
11
122
13C={3,4,5,7}A=A1={2,4,9}A2={3,4}A3={6}A4={6,8}
m
iAi
1
(w9/d9)
(w2/d2)(w3/d3)
(w4/d4)
(w6/d6)
(w7/d7)
(w8/d8)
K={9}C={3,4,5,6,7}A={2,8,99,3}
(w5/d5)
Dynamic Query-tree Energy Balancing Algorithm (DQEB) A Greedy algorithm for parent selection
Disconnected Problem
i
5
4
3
14
7
6
8
151
9
10
11
122
13C={3,4,5,7}A=A1={2,4,9}A2={3,4}A3={6}A4={6,8}
m
iAi
1
(w9/d9)
(w2/d2)(w3/d3)
(w4/d4)
(w6/d6)
(w7/d7)
(w8/d8)
K={9,8}C={3,4,5,6,7}A={2,88,99,3,7}
(w5/d5)
Dynamic Query-tree Energy Balancing Algorithm (DQEB) A Greedy algorithm for parent selection
Disconnected Problem
i
5
4
3
14
7
6
8
151
9
10
11
122
13C={3,4,5,7}A=A1={2,4,9}A2={3,4}A3={6}A4={6,8}
m
iAi
1
(w9/d9)
(w2/d2)(w3/d3)
(w4/d4)
(w6/d6)
(w7/d7)
(w8/d8)
K={9,8,7}C={3,4,5,6,7}A={2,88,99,3,77,6}
(w5/d5)
Dynamic Query-tree Energy Balancing Algorithm (DQEB) Resolving Update Conflicts
Two nodes trigger DQEB concurrently. Lock mechanism
Once the UC triggers all its children for their APs’ information, all children will freeze themselves.
Any UC that does not get a response from some of its children waits for a given time.
Simulation Results
Network Size: 600 m x 600 m Nodes number: 1000 Nodes positions: Uniformly distributed Network is connected and Query tree has
been constructed initially. λ=3 β=3 , w=(1-P) β
Simulation Results
Power Distribution Balance v.s Time
Simulation Results
Life Time v.s Death Rate threshold
Simulation Results
Power STD v.s Connectivity
Simulation Results
Life Time v.s Connectivity
Conclusion
We proposed an energy aware, distributed protocol to dynamically update query tree structures in sensor network.
Decisions are taken at each non-leaf node to locally minimize the cost of the broadcast tree by switching a non-leaf node with low remaining power to a leaf node so that its energy depletion rate is decreased.