Zinaida Benenson Felix FreilingMarkus Bestehorn Marek Jawurek
Query Dissemination with Predictable
Erik Buchmann
Query Dissemination with PredictableReachability and Energy Usage in SensorNetworksAdHoc-Now 2008, Sophia Antipolis
www.kit.edu
Introduction – Sensor Networks
A sensor network consists multiple of sensor nodes, e.g.IntroductionProblem Desc.IdIdea
ReachabilityDirectI di t
MicaZSun SPOT
Indirect
Topology Information
Sensor NodesBattery-poweredEquipped with sensor hardware
EvaluationSetupSimulationBreak Even Equipped with sensor hardware
Limited computing resourcesWireless communication
Break-EvenDeployment
Conclusion
Q&A
Slide 2Markus Bestehorn
Query Processing in WSN
Generic query processing in sensor networks approach:1. Disseminate query through base station
IntroductionProblem Desc.Id q y g
SELECT MAX(temp) FROM sensors …
2. Measure data using sensing hardware3 Process & route query results back to base station
Idea
ReachabilityDirectI di t 3. Process & route query results back to base station
Optimization Goal: Reduce energy consumption!
Indirect
Topology Information
Sending/Receiving data most expensive!
215°C
17°C
EvaluationSetupSimulationBreak Even 2
3
4 6
17°C21°C
QQQ
QQQ
Break-EvenDeployment
Conclusion3
1 5 Basestation20°C
19°C
22°C
QQQ
Q22°CQ&A
Slide 3Markus Bestehorn
Challenges for Query Dissemination
Unnecessary rebroadcasts must be avoidedNodes should receive query only once
IntroductionProblem Desc.Id q y yIdea
ReachabilityDirectI di t
2 4 6Q Q
?
Existing approaches
Indirect
Topology Information
315Q
Q
Existing approachesTopology-Based: Determine rebroadcasting nodes using accurate local topology information
EvaluationSetupSimulationBreak Even 2-Hop topology information is very costly
Optimal Broadcast Dominating Set Problem NP-completeProbabilistic: Nodes rebroadcast with probability p
Break-EvenDeployment
ConclusionHigh p high energy consumptionLow p not all nodes reachedHow to set p?
Q&A
Slide 4Markus Bestehorn
Idea & Agenda
General idea:Acquire basic topology information
IntroductionProblem Desc.Id q p gy
does not consume as much energyUse probabilistic approach to disseminate querySet rebroadcast probability based on basic topology information
Idea
ReachabilityDirectI di t Set rebroadcast probability based on basic topology information
Agenda:
Indirect
Topology Information
Prediction frameworkHow to predict reachability for a given rebroadcast probability p?How to set p based on prediction to reach all nodes?
EvaluationSetupSimulationBreak Even How to set p based on prediction to reach all nodes?
Topology DiscoveryPossibilities to aquire required topology information?
Extensive evaluation
Break-EvenDeployment
ConclusionExtensive evaluation
Simulation and real deployment resultsExplore tradeoff reachability vs. energy consumptionA f th P di ti F k?
Q&A
Slide 5Markus Bestehorn
Accuracy of the Prediction Framework?
Hop Set Modell (1)
Task: Predict the number of reached nodes givenTopology information
IntroductionProblem Desc.Id p gy
Rebroadcast probability pHop Set: Hop Set H[i] contains all nodes that can be reached by the base station via i hops
Idea
ReachabilityDirectI di t reached by the base station via i hopsIndirect
Topology Information H[1]H[2]H[3] H[0]
EvaluationSetupSimulationBreak Even
2 4 6
Break-EvenDeployment
Conclusion31 5
Q&A
Slide 6Markus Bestehorn
Hop Set Modell (2)
Possibilities to reach a node via broadcastDirect: Message is sent from node in H[i-1] to node in H[i]
IntroductionProblem Desc.Id g [ ] [ ]
Indirect: Message is sent from node in H[i] to node in H[i]Backwards: Node in H[j] with j > i forwards message to node in H[i] Simplification: not considered
Idea
ReachabilityDirectI di t H[i] Simplification: not consideredIndirect
Topology Information
EvaluationSetupSimulationBreak Even 4
H[1]H[2]H[3] H[0]
QQQBreak-EvenDeployment
Conclusion
24
6Q
Q
Q
Q
Q
Q&A 31 5
Slide 7Markus Bestehorn
Reachability Prediction
R(h,p) := number of reached nodes in Hop Set h with rebroadcast probability p
IntroductionProblem Desc.Id
p y pR(0,p) = 1 base station is always „reached“R(1,p) = |H[1]|
base station always broadcasts H[1]
Idea
ReachabilityDirectI di t base station always broadcasts
Hop Set H[1] always reached
Nodes in s bseq ent Hop Sets are reached
H[1]Indirect
Topology Information
Nodes in subsequent Hop Sets are reachedDirectly Direct(h,p)Example: Direct(2,p)=4Indirectly Indirect(h p) H[1]
EvaluationSetupSimulationBreak Even Indirectly Indirect(h,p)
Example: Indirect(2,p)=2H[1]
H[2]Break-EvenDeployment
ConclusionR(h,p) := Direct(h,p) + Indirect(h,p) with h > 1Q&A
Slide 8Markus Bestehorn
Direct Reachability Prediction
Basic Idea to compute Direct(h,p)Possible rebroadcasters |H[h-1]| nodes
H[i]H[i-1]IntroductionProblem Desc.Id | [ ]|
Potential Rebroadcasters R(h-1,p) nodesRebroadcasters R(h-1,p)·p nodes
|H[h-1]|R(h 1 )
Idea
ReachabilityDirectI di t |H[h 1]|R(h-1,p)
R(h-1,p) ·p
Indirect
Topology Information
P(„Node in H[h] directly reached“) can be computed
EvaluationSetupSimulationBreak Even Avg. Number of connections from
H[i] to H[i-1] Connectivity[h]Detailed description in the paper
Break-EvenDeployment
Conclusion p p p
Direct(h,p) = P(reached directly)·s[h]
pRebroadcast
Probability
H[i]Nodes reached
in i Hops
Q&A
Slide 9Markus Bestehorn
p
Indirect Rechability Prediction
Idea to compute Indirect(h,p):Potential Rebroadcasters Direct(h,p)
IntroductionProblem Desc.Id ( ,p)
Rebroadcasters Direct(h,p)·pAverage Number of connections within a Hop set
Interconnectivity[h]
Idea
ReachabilityDirectI di t Interconnectivity[h]
Indirect(h,p)=Direct(h,p)·p·Interconnectivity[h]Indirect
Topology Information H[1]H[2] H[0]Evaluation
SetupSimulationBreak Even
4 6
Implicit Assumption:
Break-EvenDeployment
Conclusion3 5
Implicit Assumption:Reached nodes distributed evenly within hop sets
Q&A pRebroadcast
Probability
H[i]Nodes reached
in i Hops
Slide 10Markus Bestehorn
p
Reachability Prediction (3)
R(h,p) computes reached nodes in Hop Set h with rebroadcast probability p
IntroductionProblem Desc.Id
p y pComputing total reachability for given p:
( )( )][min)( hHphRpR ∑=Idea
ReachabilityDirectI di t
Minimum required because Direct(h p) + Indirect(h p) > H[h]
( )( )][,,min)( hHphRpRh∑=Indirect
Topology Information
Minimum required because Direct(h,p) + Indirect(h,p) > H[h]possibleEvaluation
SetupSimulationBreak Even
Also available:Number of sent messages / rebroadcasting nodes
Break-EvenDeployment
ConclusionNumber of sent messages / rebroadcasting nodesNumber of received messagesAllows estimation of energy consumption!
Q&A
Slide 11Markus Bestehorn
Topology Information
Required Topology Information for Reachability PredictionSet Size: Number of Nodes in each Hop Set H[h]
IntroductionProblem Desc.Id p [ ]
Connectivity: Avg. Number of connections a node in H[h] has to nodes in H[h-1]Interconnectivity: Avg Number of connections a node in H[h]
Idea
ReachabilityDirectI di t Interconnectivity: Avg. Number of connections a node in H[h]
has to other nodes in H[h]Example:
Indirect
Topology Information
H[i]H[i-1] Set size Connectivity
EvaluationSetupSimulationBreak Even … i-1 i …
… 2 3 …… i-1 i …… 1.5 2 …
I t ti it
Break-EvenDeployment
ConclusionInterconnectivity
… i-1 i …… 0 4/3 …
Q&A
Slide 12Markus Bestehorn
… 0 4/3 …
Acquiring Topology Information
Several options to get required topology information:Echo Algorithm
IntroductionProblem Desc.Id g
Expansion Wave: Explore network by initiating a flooding at the base stationContraction Wave: Aggregate topology information towards base
Idea
ReachabilityDirectI di t
gg g p gystation
Drawback: Energy consumption, ScalabilityGossiping: Nodes attach routing information to messages
Indirect
Topology Information Gossiping: Nodes attach routing information to messages
Advantage: No extra messagesDrawback: Routing information disperses slowly
Routing Protocol Extraction: Extract topology information
EvaluationSetupSimulationBreak Even Routing Protocol Extraction: Extract topology information
from data structures of routing protocolDrawback: Only possible for some protocols (AODV)
N t
Break-EvenDeployment
ConclusionNote:
Even for Echo Algorithm Prediction pays off after a few query disseminations!
Q&A
Slide 13Markus Bestehorn
q y
Evaluation - Setup
Network: 125 to 425 nodesNode Degree: 4 – 16
IntroductionProblem Desc.Id g
Different Topology Types used, e.g.Uniform: Nodes are placed uniformly around basestationG i G i di ib i f d d b i
Idea
ReachabilityDirectI di t Gaussian: Gaussian distribution of nodes around basestation
100 topologies per topology type, 40 queries per topologyEnergy prediction based values measured on MicaZ
Indirect
Topology Information Energy prediction based values measured on MicaZ
Criteria for success:
EvaluationSetupSimulationBreak Even Accurate Prediction for Reachability and Energy
Optimization of probabilistic rebroadcast parameter pto reach ALL nodes with query
Break-EvenDeployment
Conclusionto reach ALL nodes with querywithout rebroadcasting at each node
Exploration of rebroadcast probability – reachability tradeoff
Q&A
Slide 14Markus Bestehorn
Evaluation – Simulation Results
Result for node degree 16, 425 nodesIntroductionProblem Desc.Id Uniform GaussianIdea
ReachabilityDirectI di t
p0
Indirect
Topology Information
EvaluationSetupSimulationBreak Even
Findings:
Break-EvenDeployment
Conclusion
Reachability & energy prediction accurateFor most experiments, there exists a p0<1: Increasing p beyond p0 does not pay off regarding reachability!
Q&A
Slide 15Markus Bestehorn
p0 p y g g yenergy savings without reducing reachability
Break Even Point
Exemplary computation:Uniform topology
IntroductionProblem Desc.Id p gy
425 nodes, node degree 16Assuming
T l di i h E h Al i h
Idea
ReachabilityDirectI di t Topology discovery using the Echo Algorithm
Energy consumption values measured on MicaZIndirect
Topology Information
Topology Discovery consumes 722 mAsQuery dissemination with simple flooding (p=1) consumes 370 A
EvaluationSetupSimulationBreak Even 370 mAs
Using prediction framework for 99% reachabilityp=0.6 220 mAs
Break-EvenDeployment
Conclusion pResult:Topology Discovery pays off after 5 queries!
Q&A
Slide 16Markus Bestehorn
Evaluation – SPOT Deployment
17 SPOTs + Basestation deployed10 Queries were disseminated into the
IntroductionProblem Desc.Id 10 Queries were disseminated into the
network using Simple flooding (p=1)P b bili ti fl di
Idea
ReachabilityDirectI di t Probabilistic flooding
Prediction algorithm was used to reachAll nodes
Indirect
Topology Information
At lowest possible rebroadcast prob. pResult:
Broadcast Reached Sent Msg Received
EvaluationSetupSimulationBreak Even Broadcast
AlgorithmReached
NodesSent Msg. Received
Msg.Simple 16.3 16.3 63.8Probabilistic 15 4 10 2 34
Break-EvenDeployment
Conclusion
Probabilistic Rebroadcast Optimization~30% less sent messages
Probabilistic 15.4 10.2 34Q&A
Slide 17Markus Bestehorn
almost 50% less received messages
Summary
Explored relations betweenReachability
IntroductionProblem Desc.Id y
Energy consumption for query disseminationEnergy spent to acquire topology information
I t d d l ti l f k
Idea
ReachabilityDirectI di t Introduced analytical framework
Determines p0<1 for probabilistic broadcasting to reach all nodes
Indirect
Topology Information
Allows predictions regarding sent / received messages Energy consumption
EvaluationSetupSimulationBreak Even gy p
Energy spent for topology information pays off after a few (5) query disseminations
Even if echo algorithm is used!
Break-EvenDeployment
ConclusionEven if echo algorithm is used!
Evaluation using Simulation & real Sensor networkQ&A
Slide 18Markus Bestehorn
Outlook
Integrate „backwards“ reachability intoframework
IntroductionProblem Desc.Id
More topology information required?Payoff?
Relation between query dissemination and query result
Idea
ReachabilityDirectI di t Relation between query dissemination and query result
accuracyIndirect
Topology Information
p0
~100% reachability100% accuracy
~100% reachability100% accuracy
<100% reachability? accuracy
<100% reachability? accuracy
EvaluationSetupSimulationBreak Even
Dynamic usage of different broadcast algorithmsProbabilistic approach good for dense networks
Break-EvenDeployment
ConclusionProbabilistic approach good for dense networksSwitch to other broadcast algorithms in less populated areas of the network?
Q&A
Slide 19Markus Bestehorn
Thank you for your attention!
IntroductionProblem Desc.Id
Questions?Idea
ReachabilityDirectI di tIndirect
Topology Information
EvaluationSetupSimulationBreak EvenBreak-EvenDeployment
Conclusion
Q&A
Slide 20Markus Bestehorn