Balanced- -energy Sleep energy Sleep Balanced Scheduling Scheme for High Scheduling Scheme for High Density Cluster- -based Sensor based Sensor Density Cluster Networks Networks † and P. J. Deng * , Y. Han ** , W. Heinzelman † and P. Varshney Varshney * J. Deng * , Y. Han ** , W. Heinzelman * * Syracuse University Syracuse University * ** National Taipei University, Taiwan National Taipei University, Taiwan ** † University of Rochester University of Rochester †
Motivation Motivation � Consider: Consider: � � Sensor network with randomly distributed Sensor network with randomly distributed � sensors sensors � Application: provide coverage of area for Application: provide coverage of area for � surveillance (QoS) surveillance (QoS) � Assumption: Assumption: � � Sensor density is higher than necessary for Sensor density is higher than necessary for � meeting QoS meeting QoS
Motivation (cont.) Motivation (cont.) � Characteristics of sensor networks Characteristics of sensor networks � � Low energy Low energy � � Low bandwidth Low bandwidth � � Networks expected to last for months unattended Networks expected to last for months unattended � � Energy Energy- -efficiency is crucial efficiency is crucial � � Exploit redundancy by powering down Exploit redundancy by powering down � unnecessary sensors unnecessary sensors � Save energy for later when nodes are more important Save energy for later when nodes are more important � � Sleep Scheduling Problem: Which sensors to Sleep Scheduling Problem: Which sensors to � power down? power down?
Cluster- -based Networks based Networks Cluster � Base station cannot manage sensors Base station cannot manage sensors � directly directly � Clustering provides framework for Clustering provides framework for � � Local control Local control � � Resource management Resource management � � Channel access Channel access � � Data fusion Data fusion � � Within a cluster, how to set nodes to sleep? Within a cluster, how to set nodes to sleep? �
Assumptions Assumptions � Dense, static, circular clusters Dense, static, circular clusters � � Variable transmission power to reach cluster head Variable transmission power to reach cluster head � � x x = distance from sensor to cluster head = distance from sensor to cluster head � � Nodes distributed as 2D Poisson point process Nodes distributed as 2D Poisson point process � � Energy savings is expected energy consumption Energy savings is expected energy consumption � were the node awake were the node awake γ = λ ⋅ ⋅ + E active ( x ) k [max( x , x )] k 1 min 2
Initial Scheduling Techniques Initial Scheduling Techniques � Randomized scheduling (RS) Randomized scheduling (RS) � � Randomly put sensors to sleep Randomly put sensors to sleep � = β < p 1 � Each sensor sleeps with probability Each sensor sleeps with probability � s � Distance Distance- -based scheduling (DS) based scheduling (DS) � � Probability Probability p p linearly related to linearly related to x x � � Sensors further from cluster head put to sleep with Sensors further from cluster head put to sleep with � higher probability higher probability β 3 x = ≤ ≤ s p ( x ) 0 x R 2 R
Coefficient of Variation Coefficient of Variation # nodes 500 # nodes 500 k 1 10 - -6 6 J/pkt/m J/pkt/m 2 2 k 10 1 k 2 k 0.1 J/sec 0.1 J/sec 2 x min 10 m x 10 m min λ 25, 50, 100 λ 25, 50, 100 pkt/sec pkt /sec R 100 m R 100 m γ γ 2 2 � Analytically determine coefficient of variation of Analytically determine coefficient of variation of � energy consumption for RS and DS energy consumption for RS and DS
Balanced- -Energy Scheduling (BS) Energy Scheduling (BS) Balanced � p(x p(x) ) chosen so nodes consume same amount of chosen so nodes consume same amount of � energy, on average energy, on average � Let Let E E BS (x) ) be expected energy consumption of a be expected energy consumption of a BS (x � node at distance node at distance x x from cluster head from cluster head � Find Find p(x p(x) ) such that such that E E BS BS (x (x) ) does not depend on does not depend on x x � Can only energy balance certain portion β β b � Can only energy balance certain portion of nodes b of nodes � � Nodes close to cluster head not energy balanced Nodes close to cluster head not energy balanced � = − = ≤ ≤ ( b ) E ( x ) [ 1 p ( x )] E ( x ) E x x R BS active BS b
Results Results
Performance Evaluation Performance Evaluation � Analytically Analytically � determine determine expected energy expected energy consumption consumption � λ λ = 100 = 100 pkts/s pkts/s �
Performance Evaluation (cont.) Performance Evaluation (cont.) � BS achieves goal of lower coefficient of variation BS achieves goal of lower coefficient of variation �
Network Lifetime Network Lifetime � T( T( β β d ) = time when = time when β β d fraction of sensors run out d ) d fraction of sensors run out � of energy of energy � Initial sensor energy = Initial sensor energy = Ψ Ψ � � For BS, For BS, β β b b fraction of nodes consume same energy fraction of nodes consume same energy � Ψ β = � When When β β d = β β b T ( ) � d = b BS d ( b ) E BS � In RS, nodes farther away consume more energy In RS, nodes farther away consume more energy � � Run out of energy faster than closer nodes Run out of energy faster than closer nodes � � In DS, network lifetime calculated numerically In DS, network lifetime calculated numerically �
Lifetime Results Lifetime Results
Lifetime Results (cont.) Lifetime Results (cont.) � BS has 70% (150%) longer lifetime than BS has 70% (150%) longer lifetime than � DS (RS) for β β d = 0.1 DS (RS) for d = 0.1 � BS has better lifetime than DS and RS for BS has better lifetime than DS and RS for � all points except β β d = 0.5 and β β s < 0.27 all points except d = 0.5 and s < 0.27 � Small Small β β s � fewer sensors energy balanced fewer sensors energy balanced s � � � 50% sensors run out of energy quickly 50% sensors run out of energy quickly �
Sensing Coverage Sensing Coverage RS BS DS
Sensing Coverage Distribution Sensing Coverage Distribution Initial sensing coverage Sensing coverage distribution distribution after 40% nodes run out of energy
50% Sensors Remaining 50% Sensors Remaining DS BS RS
Conclusions Conclusions � Sleep scheduling important to extend network Sleep scheduling important to extend network � lifetime lifetime � Balanced Balanced- -energy scheduling effective in energy scheduling effective in � extending lifetime while maintaining coverage extending lifetime while maintaining coverage � Future work Future work � � Explore different initial energies Explore different initial energies � � Dynamically changing clusters and cluster heads to Dynamically changing clusters and cluster heads to � balance energy among all nodes balance energy among all nodes
Recommend
More recommend
