Power Aware Routing for Sensor Databases

1
Power Aware Routingfor Sensor Databases
(IEEE INFOCOM 2005)

• Ki Sung Lee
• 06 Oct. 05
• DB Lab.

2
Overview

• Introduction
• Background
• Motivation
• Assumptions
• Preview
• Fully aggregated queries
• Unaggregated queries
• Partially aggregated queries
• Experimental result
• Summary

3
Background

• Sensor database
• To provide a SQL like data interface in sensor
  networks
• In-network aggregation
• Processes queries by using a routing tree
• Power of sensor nodes
• Battery power
• Severely power-limited
• Impossible to replenish
• Communication cost
• The largest factor in power consumption

4
Motivation

• Problem of previous power aware routing
• No consideration of aggregated data packets
• Problem of previous query processing
• No consideration of the power optimality of the
  routing tree
• ?Need to make energy efficient routing in a
  sensor network for data aggregation queries

5
Assumptions

• Sensor network
• N sensor nodes (1,2,,N)
• Energy ei, i 1, , N
• Node 1 acts as the base station with unlimited
  power supply (e18)
• Characteristics
• Every node generates one unit of data every time
  unit
• Transmitting one unit of data costs one unit of
  energy
• Receiving one unit of data costs cr lt1 units of
  energy
• Two sensors can communicate with each other if
  they are within range r
• Sensors transmit with fixed power
• There exists a time synchronization model
• System lifetime
• Metrics for power efficiency
• The time required for first node failure to occur

6
Preview

• Classification of database queries
• Fully-aggregated query
• Processed by transmitting a single value
• SELECT AVG(temp) FROM sensors
• Unaggregated query
• Processed by transmitting all incoming values
• SELECT nodeid FROM sensors WHERE (tempgt70)
• Partially aggregated query
• Processed by transmitting values with an upper
  bound for the amount of data
• SELECT HISTOGRAM(temp) FROM sensors

7
Preview (contd)

• Hardness
• Finding the optimal routing tree is NP-complete
• Approximation algorithms
• For fully aggregated queries
• Constant factor approximation algorithm
• For unaggregated and partially aggregated queries
  
• Heuristic algorithms with excellent performance
  in practice

8
Fully aggregated queries

• Without reception cost
• With reception cost

Proposition In the model where the receive cost
is assumed to be zero, every spanning tree is
optimal for aggregated queries, even with
arbitrary node energy levels.
Theorem Finding maximum lifetime routing tree
for fully aggregated queries with reception
costs is NP-complete.
Proof same as MINIMUM DEGREE SPANNING TREE
(MDST) which is NP-complete
Upper bound Topt emin (Topt optimal
routing lifetime)
9
Fully aggregated queries (contd)

• AGGREGATED-TREE algorithm
• Transformation of routing problem to the MDST
  problem
• By augmenting auxiliary nodes to each node
• Solving the MDST problem
• By OPT1 approximation algorithm for finding the
  MDST
• ? constant factor approximation algorithm

10
Unaggregated queries

• Hardness of the optimal routing tree problem
• Finding an upper bound
• By an integer programming with some relaxation
• To investigate the possibility of approximation
  algorithms

Theorem Finding maximum lifetime routing Tree
for unaggregated queries is NP-complete
Proof By reduction from SET-COVER to this
problem
11
Unaggregated queries (contd)

• Energy Conserving Routing Tree (ECRT)
• Heuristic algorithm
• Very similar to Prims MST algorithm

Algorithm ECRT(G(V,E),e) Initialize tree T to
contain the single node root. while Not all nodes
are in T do Find lifetime t of the tree T Find
the set of nodes N adjacent to T Add node v ? N
which reduces T by the least amount end
while Output T
12
Unaggregated queries (contd)

• Local optimization
• Locally optimal the lifetime of the tree can not
  be improved by switching the parent of any single
  node
• Optimal tree is always locally optimal
• Can be used as a stand alone algorithm

13
Unaggregated queries (contd)
Algorithm LOCAL_OPT(G(V, E), T, e) Done lt-
FALSE while doneFALSE do done lt- TRUE for all
v ? V do if switching the parent of v improves
T then switch parent of v to improve T done
lt- FALSE end if end for end while Output T
14
Partially aggregated queries

• Adapt the approximation algorithms used to solve
  the unaggregated routing tree problem

15
Experimental result

• System lifetime for unaggregated queries

Max Flow
ECRTLocal Local Opt
ECRT
N400 nodes all equal energy of 1000 units
Min Hop
16
Summary

• 3 types of the query
• NP-complete optimal routing tree problems
• Constant factor approximation algorithm for fully
  aggregated queries
• Heuristic algorithms with excellent performance
  in practice for other queries