a case for resource heterogeneity in large sensor networks
DESCRIPTION
A Case for Resource Heterogeneity in Large Sensor Networks. Srikanth Kandula MIT. with Jennifer Hou, Lui Sha (UIUC). IEEE MILCOM Nov 1, 2004. Lot of energy wasted in forwarding alerts. Increased wireless collisions. Problem. Blindly suppressing alerts is bad - PowerPoint PPT PresentationTRANSCRIPT
A Case for Resource Heterogeneity in Large Sensor Networks
Srikanth KandulaMIT
IEEE MILCOMNov 1, 2004
with Jennifer Hou, Lui Sha (UIUC)
IEEE MILCOM 04 2
Blindly suppressing alerts is bad
Dissemination Trees process alerts at intermediate sensors?• may not see all alerts• may not have cpu/memory to correlate
Problem
Need dedicated entities that process sensor data close to point of origin
Need dedicated entities that process sensor data close to point of origin
Lot of energy wasted in forwarding alerts
Increased wireless collisions
IEEE MILCOM 04 3
Resource-rich Accumulator
HowTo co-ordinate and control sensors using a random deployment of accumulators?
HowTo co-ordinate and control sensors using a random deployment of accumulators?
Resource-Heterogeneous networks
Tunnelled Digest
Also provides fine-grained control Serve as a spine for the n/w
IEEE MILCOM 04 4
Accumulator-centered clusters
IEEE MILCOM 04 5
Accumulator-centered clusters
Cluster sensors around closest accumulator Many sizes Multiple hops
An overlay of accumulators
Digests are routed along the overlay
Tunneled Path
Improves network lifetime and goodput of updatesImproves network lifetime and goodput of updates
IEEE MILCOM 04 6
Accumulator centered clusters– Level 1
Cluster sensors around closest accumulator Clusters vary in size Multiple hops
An overlay of accumulators
Digests are routed along the overlay
Tunneled Path
How to efficiently form these clusters?How to efficiently form these clusters?
IEEE MILCOM 04 7
Contributions
Detailed design of a sensor n/w architecture hinged on resource-rich nodes
Protocols for efficient clustering and routing
Demonstrate several distinct benefits through ns2 simulations
Identify opportunities for load-sensitive routing
IEEE MILCOM 04 8
Talk Outline
Internal structure of Level-1 Clusters
Hop-by-hop forwarding of join messages
Routing
Results
IEEE MILCOM 04 9
Level-0 Hexagonal Cells
Idle radios consume power
Only the leader of a cell forwards traffic
IEEE MILCOM 04 10
Level-0 Hexagonal Cells
Idle radios consume power
Only the leader of a cell forwards traffic
Leadership exchanged among all nodes in a cell
Cell size chosen to maintain connectivity
Note: Equivalent Level-0 clusters e.g. SPAN can be used instead
Level-0 clusters save power! Level-0 clusters save power!
IEEE MILCOM 04 11
Talk Outline
Internal structure of Level-1 Clusters
Hop-by-hop forwarding of join messages
Routing
Results
IEEE MILCOM 04 12
Accumulator Broadcasts join messages
One hop to accumulator A1
IEEE MILCOM 04 13
Grids store back-pointer to parent
1
1
1
1
1
1
Back-pointer to parent cell
IEEE MILCOM 04 14
Join messages are re-broadcast!
1
1
1
1
1
1
Two hops to accumulator A1
IEEE MILCOM 04 15
Join messages are re-broadcast!
1
1
1
1
1
1
Two hops to accumulator A1
Join messages to parent’s neighbors will propagate errors
IEEE MILCOM 04 16
Controlled Broadcast - 1
1
1
1
1
1
1
Two hops to accumulator A1
Suppress Join messages to
neighbors of parent
IEEE MILCOM 04 17
Controlled Broadcast - 2
1
1
1
1
1
1
2
2
2
Suppress Join messages to a neighbor if there is a common neighbor of smaller hop count
IEEE MILCOM 04 18
Controlled Broadcast - 2
1
1
1
1
1
1
2
2
2
Three hops to accumulator A1
IEEE MILCOM 04 19
BGP Inefficiencies
1
1
1
1
1
1
2
3
2
2
3
IEEE MILCOM 04 20
Controlled Broadcast
1
1
1
1
1
1
2
3
2
2
3
Three hops to accumulator A1
IEEE MILCOM 04 21
Multiple paths
1
1
1
1
1
1
2
3
2
2
3
3
3
A cell may have multiple parents
IEEE MILCOM 04 22
Multiple paths
1
1
2
1
1
2
1
1
2
3
2
2
3
3
3
Multiple dis-joint paths exist!
IEEE MILCOM 04 23
Level-1 cluster boundary
1
1
2
1
1
2
1
1
2
3
2
2
3
3
3
One hop to accumulator A2
IEEE MILCOM 04 24
Level-1 cluster boundary
1
1
2
1
1
2
1
1
2
3
2
2
3
3
3 1
1
1
1
1
1
Two hops to accumulator A2
IEEE MILCOM 04 25
Level-1 cluster boundary
1
1
2
1
1
2
1
1
2
3
2
2
3
3
2
2
1
1
1
1
2
1
1
Cell changes allegiance!
IEEE MILCOM 04 26
Level-1 cluster boundary
1
1
2
1
1
2
1
1
2
3
2
2
3
3
2
2
1
1
1
1
2
1
1
Three hops to accumulator A2
IEEE MILCOM 04 27
Boundary notification
1
1
2
1
1
2
1
1
2
3
2
2
3
3
2
2
1
1
1
1
2
1
1
Negative Ackx hops from A1
IEEE MILCOM 04 28
Boundary notification
1
1
2
1
1
2
1
1
2
3
2
2
3
3
2
2
1
1
1
1
2
1
1
Negative Ack:x hops from A1
Border Cell:3 hops to A2
Border Cell:3 hops to A1
IEEE MILCOM 04 29
Level-1 Cluster formation: Summary
Controlled broadcast of join messages Naïve O(N^2) O(N) messages, N = # of cells
Routing information piggybacks over join messages Multiple paths from a cell to closest accumulator Multiple paths between adjacent accumulators
Join broadcasts are periodically repeated Join messages route around holes
IEEE MILCOM 04 30
Talk Outline
Internal structure of Level-1 Clusters
Hop-by-hop forwarding of join messages
Routing
Results
IEEE MILCOM 04 31
Sensor Alerts Accumulator
1
1
2
1
1
2
1
1
2
3
2
2
3
3
2
2
1
1
1
1
2
1
1Sensor forwards alert to cell
leader
Alert follows parent pointers back to the
accumulator
IEEE MILCOM 04 32
Accumulator Accumulator Sink
1
1
2
1
1
2
1
1
2
3
2
2
3
3
2
2
1
1
1
1
2
1
1
Accumulator forwards digests to some border
cellDigest crosses the border
and follows parent pointers
IEEE MILCOM 04 33
Accumulator Accumulator Sink
Accumulators run OSPF on the overlay
Pick one of many paths
Improved resilience
IEEE MILCOM 04 34
Talk Outline
Internal structure of Level-1 Clusters
Hop-by-hop forwarding of join messages
Routing
Results
IEEE MILCOM 04 35
Power Savings – CDF of remaining power
Pro
babi
lity
Remaining Power (J)
Homogenous Network
Heterogeneous Clusters
Sensors that generate updates gain a little due to decreased forwarding
load
IEEE MILCOM 04 36
Power Savings – CDF of remaining power
Pro
babi
lity
Remaining Power (J)
Homogenous Network
Heterogeneous Clusters
Long-haul sensors gain more due to
decrease in forwarded traffic
IEEE MILCOM 04 37
Power Savings – CDF of remaining power
Pro
babi
lity
Remaining Power (J)
Homogenous Network
Heterogeneous Clusters
Idle sensors gain a lot by turning off their radios
IEEE MILCOM 04 38
Improvement in GoodputF
ract
ion
of U
pdat
es R
ecei
ved
Duration between generation of updates (s)
14 accum
Zero accum
500 sensors spread over 4sq-km
357 Cells, 15 CBR sources
A small number of accumulators is enough to improve goodput
significantly
IEEE MILCOM 04 39
Improvement in GoodputF
ract
ion
of U
pdat
es R
ecei
ved
Duration between generation of updates (s)
14 accum
Zero accum
500 sensors spread over 4sq-km
357 Cells, 15 CBR sources
More accumulators are needed when updates are generated
faster
IEEE MILCOM 04 40
Reduction in join message overheadT
otal
num
ber
of jo
in m
essa
ges
Number of Accumulators
Naïve Algo.
Controlled Broadcast
Significantly reduces # of join messages when clusters are
large
500 sensors spread over 4sq-km
357 Cells
IEEE MILCOM 04 41
Reduction in join message overheadT
otal
num
ber
of jo
in m
essa
ges
Number of Accumulators
Naïve Algo.
Controlled Broadcast
Slightly beneficial even when
Level-1 clusters are small
500 sensors spread over 4sq-km
357 Cells
IEEE MILCOM 04 42
Related Work
We explore the use of heterogeneous sensors in designing better dissemination trees.
Clustering and Routing in Sensor Networks Geographic Area Forwarding (GAF), SPAN, k-
cluster based routing, ZRP, Spine, Min-ID/Max degree
Dissemination Trees in sensor networks Diffusion, TTDD
IEEE MILCOM 04 43
Conclusion
A small number of resource-rich sensors Improve network lifetime Improve goodput of updates Provide fine-grained control
Hierarchical clusters to use resource-rich sensors can be built efficiently
Scope to design robust routing protocols – more work to be done!
IEEE MILCOM 04 44
Questions?