computing aggregates for monitoring wireless sensor networks jerry zhao, ramesh govindan, deborah...

25
COMPUTING AGGREGATES FOR COMPUTING AGGREGATES FOR MONITORING WIRELESS MONITORING WIRELESS SENSOR NETWORKS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Jerry Zhao, Ramesh Govindan, Deborah Estrin Deborah Estrin Presented by Presented by Hiren Shah Hiren Shah

Upload: augusta-gibbs

Post on 02-Jan-2016

215 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

COMPUTING COMPUTING AGGREGATES FOR AGGREGATES FOR

MONITORING WIRELESS MONITORING WIRELESS SENSOR NETWORKSSENSOR NETWORKS

Jerry Zhao, Ramesh Govindan, Jerry Zhao, Ramesh Govindan, Deborah EstrinDeborah EstrinPresented by Presented by Hiren ShahHiren Shah

Page 2: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

Outline Outline • Motivation and GoalsMotivation and Goals• ContributionsContributions

– ArchitectureArchitecture– Computing Network DigestsComputing Network Digests– Impact of Packet LossImpact of Packet Loss

• ExperimentsExperiments• ConclusionConclusion

Page 3: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

Motivation and GoalsMotivation and Goals• Every Network requires monitoring.Every Network requires monitoring.• Sensor Network is no exception !!!Sensor Network is no exception !!!• Monitoring in traditional networks – Monitoring in traditional networks –

SNMP.SNMP.• Need for a new architecture.Need for a new architecture.

Page 4: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

ArchitectureArchitecture

Page 5: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

Architecture (contd.)Architecture (contd.)• Three software components of the Three software components of the

architecture.architecture.– DumpsDumps

• Detailed node states. E.g. logs with raw data Detailed node states. E.g. logs with raw data reading, packet level details, energy details for a reading, packet level details, energy details for a node. node.

• Dumps are costly to collect.Dumps are costly to collect.

– ScansScans• Abstract views of resource consumption in the Abstract views of resource consumption in the

network without referring to individual node.network without referring to individual node.• Derived using in-network aggregation.Derived using in-network aggregation.

Page 6: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

Architecture (contd.)Architecture (contd.)• Network DigestsNetwork Digests

– Energy cost of computing Energy cost of computing scansscans over over entire network is very high.entire network is very high.

– ScansScans should be invoked only when should be invoked only when necessary.necessary.

– A A digest digest is an aggregate of some is an aggregate of some network property.network property.• E.g. size of the network, average energy left E.g. size of the network, average energy left

at a node etc.at a node etc.

Page 7: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

Architecture (contd.)Architecture (contd.)• DigestsDigests indicate WHEN should indicate WHEN should scansscans

be invoked.be invoked.

• ScansScans indicate WHERE should indicate WHERE should dumpsdumps be collected from. be collected from.

Page 8: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

Computing DigestsComputing Digests• Naive centralized approach does not Naive centralized approach does not

scale well and has a single point of scale well and has a single point of failure.failure.

• Some Requirements of Digest Some Requirements of Digest Computation Computation – Should be energy efficient.Should be energy efficient.– Digests should be available all the time and Digests should be available all the time and

everywhere in the network.everywhere in the network.• Typical Digest functions are Node with Typical Digest functions are Node with

max energy, count of no. of nodes, Sum max energy, count of no. of nodes, Sum of node energies, Avg. residual energy.of node energies, Avg. residual energy.

Page 9: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

Computing Digests Computing Digests (contd.)(contd.)

• These functions are decomposable.These functions are decomposable.• Hence partial result can be added to Hence partial result can be added to

get overall result.get overall result.• Hence one can use in-network Hence one can use in-network

aggregation.aggregation.

Page 10: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

Computing Digests Computing Digests (contd.)(contd.)

• One can use hierarchical approach One can use hierarchical approach for computing the digests. E.g. for computing the digests. E.g. clusters like LEACH.clusters like LEACH.

• However this involves overhead in However this involves overhead in computing leaders and maintaining computing leaders and maintaining the hierarchy.the hierarchy.

• Authors propose Digest Diffusion.Authors propose Digest Diffusion.

Page 11: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

Digest DiffusionDigest Diffusion• Nodes periodically send tuple (Mi, Si, Nodes periodically send tuple (Mi, Si,

Hi).Hi).• Initially node sets Mi to its perceived Initially node sets Mi to its perceived

maximum value e.g. its own residual maximum value e.g. its own residual energy. energy.

• Si is source of maximum. Initialized to Si is source of maximum. Initialized to i.i.

• Hi is hop distance of the source of Hi is hop distance of the source of maximum.maximum.

Page 12: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

Digest DiffusionDigest Diffusion• Upon receiving a tuple from neighbor Upon receiving a tuple from neighbor

a node performs following processing.a node performs following processing.– If Mj >Mi then set Mi=Mj ; Si=Sj ; If Mj >Mi then set Mi=Mj ; Si=Sj ;

Hi=Hj+1;Hi=Hj+1;– Also set parent Pi=j;Also set parent Pi=j;

• Algorithm converges in steps Algorithm converges in steps proportional to network diameter.proportional to network diameter.

• Approach is scalable. Periodic Approach is scalable. Periodic messages can be PIGGYBACKED. messages can be PIGGYBACKED.

Page 13: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

Computing other digestsComputing other digests

Page 14: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

Computing other digestsComputing other digests• Hence a tree is needed without any Hence a tree is needed without any

overlaps.overlaps.• Digest diffusion establishes a tree rooted Digest diffusion establishes a tree rooted

at the node with the maximum value.at the node with the maximum value.• Every node i aggregates values from its Every node i aggregates values from its

children and passes the partial result to children and passes the partial result to its parent along the tree.its parent along the tree.

• The root will have the final aggregate The root will have the final aggregate value.value.

Page 15: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

Computing other digestsComputing other digests• Metric for establishing the tree Metric for establishing the tree

should be carefully chosen such that should be carefully chosen such that the tree is relatively stable. E.g. link the tree is relatively stable. E.g. link degree of a node is a bad choice degree of a node is a bad choice since it fluctuates a lot with node since it fluctuates a lot with node failures.failures.

Page 16: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

Digest Tree MaintenanceDigest Tree Maintenance• With node failures the tree should be With node failures the tree should be

suitably changed.suitably changed.• Packet loss needs to be taken into account Packet loss needs to be taken into account

since absence of heartbeat does not mean since absence of heartbeat does not mean the neighbor has failed.the neighbor has failed.

• Let T0 be the interval between Let T0 be the interval between broadcasts.broadcasts.

• If heartbeat is not received from a parent If heartbeat is not received from a parent or child in Tp= 4To seconds change the or child in Tp= 4To seconds change the tree. tree.

Page 17: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

Impact of Packet LossImpact of Packet Loss• Packet losses are frequent in Packet losses are frequent in

wireless environments. wireless environments. • Packet losses make aggregation tree Packet losses make aggregation tree

unstable and hence affect the quality unstable and hence affect the quality of aggregate digest.of aggregate digest.

Page 18: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

Impact of packet lossImpact of packet loss

Page 19: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

Impact of packet lossImpact of packet loss

Page 20: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

Impact of packet lossImpact of packet loss

Page 21: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

Solution Solution • Link Quality Profiling and RejectionLink Quality Profiling and Rejection

– ““blacklist” links with poor quality and blacklist” links with poor quality and asymmetry.asymmetry.

– A node chooses as parent a node with which A node chooses as parent a node with which it has good and symmetric communication.it has good and symmetric communication.

• Use packet sequence number to Use packet sequence number to estimate how many of the packets sent estimate how many of the packets sent by the neighbor are getting lost.by the neighbor are getting lost.

• Exchange “ I CAN HEAR YOU “ lists to Exchange “ I CAN HEAR YOU “ lists to identify asymmetric links.identify asymmetric links.

Page 22: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

ExperimentsExperiments• Used REAL Berkley Used REAL Berkley

Motes !!!Motes !!!• Studied Studied

performance of performance of various schemes various schemes against packet loss.against packet loss.

• Used root mean Used root mean square error to square error to quantify quantify performance.performance.

Page 23: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

ExperimentsExperiments

Page 24: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

ConclusionConclusion• Proposed an architecture for Proposed an architecture for

monitoring sensor network.monitoring sensor network.• Suggested design for computing Suggested design for computing

network digests.network digests.• Carefully handling lossy and Carefully handling lossy and

asymmetric links reduces error in asymmetric links reduces error in digest computation.digest computation.

• Future work includes experiments on Future work includes experiments on larger testbed and designing a full-larger testbed and designing a full-fledged monitoring suite. fledged monitoring suite.

Page 25: COMPUTING AGGREGATES FOR MONITORING WIRELESS SENSOR NETWORKS Jerry Zhao, Ramesh Govindan, Deborah Estrin Presented by Hiren Shah

Questions ?Questions ?