routing in wireless sensor networks - uni-rostock.de · routing in wireless sensor networks ... •...
TRANSCRIPT
Routing in WirelessSensor Networks
Matthias HandyMatthias HandyUniversity of RostockUniversity of Rostock
[email protected]@uni--rostock.derostock.de
22
Overview
•• Problem Problem FormulationFormulation
•• ChallengesChallenges
•• ClassificationClassification
•• FlatFlat RoutingRouting
•• HierarchicalHierarchical RoutingRouting
•• LocationLocation BasedBased RoutingRouting
•• OtherOther
•• Open Research Open Research IssuesIssues
•• SummarySummary
33
Problem Formulation
Deployment Area
Base Station
Sensor node Sensor node in event regionSensor node outside the event region with routing task
44
Challenges for RPs in WSN
•• NodeNode deploymentdeployment
•• EnergyEnergy vs. vs. QualityQuality
•• DataData reportingreporting methodmethod•• timetime--drivendriven
•• eventevent--drivendriven
•• queryquery--drivendriven
•• NodeNode heterogeneityheterogeneity
•• Fault Fault tolerancetolerance
•• ScalabilityScalability
•• NetworkNetwork dynamicsdynamics
55
Challenges ctd.
•• Transmission media (MAC)Transmission media (MAC)
•• ConnectivityConnectivity
•• CoverageCoverage
•• DataData aggregationaggregation
66
Classification of RPs for WSN
RoutingProtocols
in WSN
NetworkStructure
ProtocolOperation
FlatNetworkRouting
HierarchicalNetworkRouting
LocationBased
Routing
NegotiationBased
Routing
MultipathBased
Routing
QueryBased
Routing
QoSBased
Routing
CoherentBased
Routing
Source: Al-Karaki / Kamal, 2004
77
Flat Routing: Characteristics
•• No No distinctdistinct rolesroles•• DataData--centriccentric routingrouting vs. vs. addressaddress--centriccentric routingrouting•• Evolution of Evolution of flatflat routingrouting: :
•• FloodingFlooding, , GossipingGossiping•• SPIN & SPIN & DirectedDirected DiffusionDiffusion
Sink
Source 1
Source 2
A B
C
1
2
2
1+2
Source 1
Source 2
Sink
A B
C1
1
2
2
2
Address-centric Data-centric
88
Flat Routing: Flooding & Gossiping
•• FloodingFlooding::•• NodeNode A A sendssends datadata to all to all neighborsneighbors
•• NeighborsNeighbors of A send of A send datadata to all to all theirtheir neighborsneighbors ……
•• UntilUntil all all nodesnodes receivedreceived thethe datadata
•• GossipingGossiping::•• basesbases on on floodingflooding
•• datadata isis onlyonly forwardedforwarded to to oneonerandomlyrandomly selectedselected neighborneighbor
•• savessaves energyenergy
Gossiping Example
99
Flat Routing: SPIN•• SSensor ensor PProtocolsrotocols forfor IInformation via nformation via NNegotiationegotiation
(Kulik et al. 2002)(Kulik et al. 2002)•• Motivation: Problems of Motivation: Problems of classicclassic floodingflooding
•• ImplosionImplosion•• OverlapOverlap•• ResourceResource BlindnessBlindness
•• Solution: Solution: SPIN SPIN protocolprotocol familyfamily•• SPINSPIN--PP (PP (forfor pointpoint--toto--pointpoint media)media)•• SPINSPIN--EC (SPINEC (SPIN--PP PP withwith a a lowlow energyenergy thresholdthreshold))•• SPINSPIN--BC (BC (forfor broadcastbroadcast media)media)•• SPINSPIN--RL (SPINRL (SPIN--BC BC forfor lossylossy networksnetworks))
•• Key Key featuresfeatures of SPIN of SPIN protocolsprotocols•• NegotiationNegotiation ((metameta datadata) ) Implosion, Implosion, OverlapOverlap•• ResourceResource adaptationadaptation ResourceResource BlindnessBlindness
•• 3 3 messagemessage typestypes::•• ADVADV•• REQREQ•• DATADATA
A
B C
D
(a)
(a)
(a)
(a)
source
sink
Implosion
Overlap
B
C
A
qr
s
(q,r) (r,s)
1010
Flat Routing: SPIN (2)SPIN-PP optimized for networks using point-to point transmission media
B
A
(1)
B
A
(2)
B
A
(3)
B
A
(5)
B
A
(6)
ADVREQ
DATA
REQREQ
REQ
B
A
(4)
ADVADV
ADV
AD
V
ADV
DATADATA
DATA
SPIN-EC: a node participates only in the three-stage SPIN-PP-protocol,if it believes that in complete all three stages
1111
Flat Routing: SPIN (3)SPIN-BC optimized for networks using broadcast transmission media
B
C
EA
D
(1)ADV B
C
EA
DREQ
(2)
B
C
EA
D
(3)DATA B
C
EA
D
(4)
G F
nodes with data
nodes without data
nodes waiting to tx REQ
transmission range
• nodes receiving ADV set arandom timer before broad-casting REQ
• a node cancels its timer whenit overhears a REQ from adifferent node on the sameADV
• DATA is broadcasted onlyonce, regardless how manyREQs are broadcasted
SPIN-RL
• SPIN-BC for lossy networks• adjustments for reliability:
• nodes request data ofoverheard, not answeredREQ messages
• multiple REQs for the samedata are answered
1212
Flat Routing: Directed Diffusion•• Elements of DDElements of DD ((IntanagonwiwatIntanagonwiwat 2001)2001)
•• DataData isis namednamed usingusing attributeattribute--valuevalue pairspairs•• interestsinterests areare disseminateddisseminated througoutthrougout thethe
wsnwsn•• disseminationdissemination setssets up up gradientsgradients to "to "drawdraw" "
eventsevents fromfrom sourcessources to to sinkssinks alongalong multiple multiple pathspaths
•• thethe networknetwork reinforcesreinforces subsetsubset of of availableavailablepathespathes fromfrom sourcesource to sinkto sink
type = four-legged animal
interval = 20 ms
duration = 10 seconds
rect = [-100, 100, 200, 400]
A simple task in a DD network
a) Interest propagation b) Initial gradients set-up c) Data delivery
Interest is 'injected' into the network at the sink node!
1313
3. gradient(data rate,
neighbor ID)
1. interest x
Flat Routing: Directed Diffusion (2)
Interests and Gradients
1
sink
2. interest cache lookup*
* if entry and gradient exist, onlytimestamp and duration are updated
4. interest x
neighbor selection for interest forwarding• broadcast (flooding)• geographic routing• use cached data
Data Propagation
source
1
rect rect
type = four-legged animal interval = 1s
rect = [-100, 100, 200, 400] timestamp = 01:20:40
expiresAt = 01:30:40
Inte
rest
x
1. target detection2. interest cache lookup
3. data msg
4. interest cache & data cache lookup
4. data msg
type = four-legged animal
instance = goat
location = [125, 220]
intensity = 0.6
confidence = 0.85
timestamp = 01:22:40
Data msg
1414
Flat Routing: Directed Diffusion (3)
Reinforcement & Negative Reinforcement
- sink may reinforce one particular neighbor(not path!)
- for higher data rate- sink re-sends interest with smaller interval to
selected node(in the picture: thick path is reinforced)
• Negative reinforcement to 'repair' degraded links by
• timeout high data rate gradients(implicit degradation)
• re-sending interest with lower data rate(explicit degradation)
All interactions in Directed Diffusion are based on local rules!
1515
Flat Routing: GBR•• GGradientradient BBasedased RRoutingouting ((SchurgersSchurgers, , SrivastavaSrivastava 2001)2001)•• BasedBased on on DirectedDirected DiffusionDiffusion
•• DD'sDD's interestinterest messagesmessages countcount hops to hops to thethe sinkssinks•• heightheight of of thethe nodenode = = hophop countcount to sink (to sink (minimumminimum))•• gradientgradient of a link = of a link = heightheight differencedifference of of bothboth link link endpointsendpoints
•• reducesreduces DD'sDD's energyenergy consumptionconsumption byby•• DataData CombiningCombining EntitiesEntities (DCE)(DCE)•• NetworkNetwork TrafficTraffic SpreadingSpreading (NTS)(NTS)
2
2
3
3
2
1 01
1
1
10
1
0
10
Packet is forwarded on the link with the largest gradient!
1
1Gradient
1 Node w. height
3 Source
Sink3
1616
Flat Routing: GBR (2)
B = Area length, R = Tx range
Data Combining Entities (DCE)
Idea: Nodes with multiple streams fromthe same event flowing through themcombine the data
sink source
DCE
DCEs reduce energy consumption but increase delay!
1717
Flat Routing: GBR (3) Network Traffic Spreading (NTS)
Stochastic scheme:
2
3 2
1 01
1
1
1
1
0
source
sink
source chooses next hop randomly
Energy-Based scheme:
2
3 3
1 01
1
1
1
0
0source
sink
remaining energy
• low-energy nodes increase height• neighbor's heights are updated
Stream-based scheme:
2
3 2
1 01
1
1
10
source
sink
1
** height = 3
• Idea: divert new streams from nodesthat are currently part of the path ofother streams
• stream node tells all neighbors(except stream origin) thatits height has increased
1818
Classification
RoutingProtocolsin WSN
NetworkStructure
ProtocolOperation
FlatNetworkRouting
HierarchicalNetworkRouting
LocationBased
Routing
NegotiationBased
Routing
MultipathBased
Routing
QueryBased
Routing
QoSBased
Routing
CoherentBased
Routing
Source: Al-Karaki / Kamal, 2004
1919
Hierarchical Routing: Characteristics
•• IntendIntend to to increaseincrease flatflatnetworksnetworks''•• efficiencyefficiency
•• scalabilityscalability
•• rolerole basedbased routingrouting schemesschemes•• Cluster Cluster headsheads
•• Cluster Cluster membersmembers
•• OftenOften stepstep--wisewise organizedorganized::•• 1. Cluster 1. Cluster formationformation
•• 2. 2. RoutingRouting
•• ChallengesChallenges•• SelectionSelection of of CHsCHs
•• Cluster Cluster formationformation
•• Medium Medium accessaccess
CM
CH
Sink
2-level hierarchical routing scheme
2020
Hierarchical Routing: LEACH
•• LLowow--EEnergynergy AAdaptive daptive CClusteringlustering HHierarchyierarchy ((HeinzelmanHeinzelman '00)'00)
•• RotatingRotating--CHCH--SchemeScheme / / CHsCHs changechange fromfrom roundround to to roundround
CM
CH
Network Model
Base Station
Threshold Computation
( ) GnnT ∉∀= 0
( ) Gn
PrP
PnT ∈∀
−
=1
mod1
P = cluster head probability
r = number of current round (notframe!)
G = nodes that have not beenclusterhead in the last 1/P rounds
2121
Hierarchical Routing: LEACH (2)Evaluation: NS2-Simulations
Tot
al e
nerg
ydi
ssip
ated
in s
yste
m[J
]
Network diameter
LEACH energy dissipation
100 nodes, message-length 2kb
Energy savings mainly depend ondata aggregation ratio in clusterheads!
2222
Hierarchical Routing: LEACH (3)
•• XLEACH XLEACH –– eXtendedeXtended LEACH (Handy et al. 2002)LEACH (Handy et al. 2002)
•• clustercluster headhead selectionselection basedbased on on energyenergy levellevel::
( )
−
+
−
=max_
_
_
_ 11
1mod1 n
currentns
xman
currentn
E
E
Pdivr
E
E
PrP
PnT
energy level correction
574
746
1104
1337
0
500
1000
1500
LEACH Improved CHS
Life
tim
e (R
ound
s)
FNDHNA
XLEACH
Nodes: 200Area: 200m*200mBase Station Pos.: (100,300)mInitial Energy / Node: 1 JMessage Length: 200 bitCH-Probability: 0.05Path-Loss (intra-cluster): 2Path-Loss (to BS): 2.5
FND=First Node Dies, HNA=Half Nodes Alive
2323
Hierarchical Routing: TEEN / APTEEN
•• TThresholdhreshold sensitive sensitive EEnergynergy EEfficientfficient sensorsensor NNetworketworkprotocolprotocol ((ManjeshwarManjeshwar//AgrawalAgrawal '01)'01)
•• LEACHLEACH--basedbased withwith MultiMulti--LevelLevel--CHsCHs
Network Model
Base Station
simple node 1st level CH 2nd level CH
Principles:
• LEACH: One sensor sample / frame and node (Periodic)
• TEEN: sample sending depends on thresholds (Event-driven)
• Hard Threshold (H) / Soft Threshold (S)
Cluster change
time
Para-meters
Attribute > H
CH receives
msg
(Attr. > H)&
(|Attr.-SV|>S)
CH receives
msg
2424
Hierarchical Routing: TEEN / APTEEN (2)
•• TEEN'sTEEN's drawbackdrawback: no : no transmissionstransmissions belowbelow thresholdsthresholds
•• Solution: Solution: AAdaptive daptive PPeriodiceriodic TThresholdhreshold--sensitivesensitive EEnergynergyEEfficientfficient sensorsensor NNetworketwork protocolprotocol (APTEEN)(APTEEN)
•• CH CH broadcastsbroadcasts::•• AttributesAttributes
•• ThresholdsThresholds
•• TDMATDMA--ScheduleSchedule
•• CountCount Time Time max. max. periodperiod betweenbetween twotwo successivesuccessive reportsreports
Cluster change
time
Parameters& TDMA-Sched.
Attribute > H
CH receives
msg
(Attr. > H)&
(|Attr.-SV|>S)
CH receives
msg
CH receives
msg
Slot fornode i
CH receives
msg
Slot fornode i
2525
Hierarchical Routing: TEEN / APTEEN (3)
Evaluation: NS2-Simulations
Time (s)
Num
ber
of N
odes
Aliv
e
Number of Nodes Alive
• APTEEN/TEEN perform betterthan LEACH because of event-driven operation
• TEEN performs better thanAPTEEN because of no periodic transmissions
• TEEN/APTEEN parametersallow energy/quality trade-offs
• TDMA-schedule requires clocksynchronization
Results:
2626
Hierarchical Routing: Sensor Aggregates Routing
•• Fang/Zhao/Fang/Zhao/GuibasGuibas '03'03•• Sensor Sensor aggregateaggregate: : setset of of nodesnodes in a in a networknetwork satisfyingsatisfying a a groupinggrouping predicatepredicate•• ProtocolsProtocols: : DAMDAM, EBAM, EMLAM, EBAM, EMLAM
Target Monitoring Scenario
Task: Determine number and appr.location of targets in a 2D-field
• targets are signal sources• each single target is assigned
to a 'cluster leader'
DAM: Distributed Aggregate Management
Purpose: Elect local cluster leaders
Leader election by comparing node's heightin the signal field 'landscape'(Leader is higher than all its neighbors)
sample signalamplitude field
Only one packet type for DAM: the 'DAM-packet'
2727
Hierarchical Routing: Sensor Aggregated Routing (2)
DAM: Distributed Aggregate Management
DAM-Packet:
MaxPr MaxID TransPr TransID
stored sensor state at each node:• maxPrHeard, leaderID, • myPr, myID, myParent• participating
A node forwards a received DAM-packet p only if:
p.maxPr > maxPrHeard && p.transPr + δ > myPr
δ = threshold to filter out tiny spurious peaks
DAM problem: undercounting (multiple targets belong to a single cluster leader)
2828
Hierarchical Routing: Sensor Aggregated Routing (3)
EBAM: Energy based activity monitoring
• DAM problem: undercounting (multiple targets belong to a single cluster leader)• EBAM solution: estimate volume of each cluster
single sourcesignal profile
superposedsignal
ceiling
EBAM operation• after DAM-phase, each leaf sensor
reports its reading (up to ceiling) to itsparent
• parents aggregate reports of all leafsand report to their parents
• until cluster leader (root) is reached
leaf 1st level parent 2nd level parent
3rd protocol EMLAM:incorporates target's movement
cluster leader
2929
Classification
RoutingProtocolsin WSN
NetworkStructure
ProtocolOperation
FlatNetworkRouting
HierarchicalNetworkRouting
LocationBased
Routing
NegotiationBased
Routing
MultipathBased
Routing
QueryBased
Routing
QoSBased
Routing
CoherentBased
Routing
Source: Al-Karaki / Kamal, 2004
3030
Location Based Routing: Characteristics
•• NodesNodes addressedaddressed byby locationlocation
•• LocationLocation obtainedobtained byby•• Distance Distance estimationestimation
•• NeighborNeighbor discoverydiscovery
•• BeaconsBeacons
•• GPSGPS
•• Frank Frank saidsaid itit beforebefore ;;--))
3131
Location Based Routing: GAF
•• GGeographicaleographical AAdaptive daptive FFidelity (idelity (XuXu et al. 2001)et al. 2001)
•• Motivation:Motivation: IdleIdle energyenergy dominatesdominates energyenergy consumptionconsumption in in AdAd--hochoc--networksnetworks
•• GAFGAF--SolutionSolution:: putput redundant redundant nodesnodes in in sleepsleep mode mode byby usingusing a a virtualvirtual gridgrid
12
3
45
nominal radio range
if node 2 is awake 3 and 4are extraneous for communicationbetween 1 and 5
Discover redundant nodes with a virtual grid:
12
3
45
A B C
r
r r r
5Rr ≤
R=nominal radio range
Virtual grid size:
3232
Location Based Routing: GAF (2)
GAF state transitions
Three states in GAF:• sleeping: power down radio• discovery: find nodes within the
same grid (initial state)• active: node participates in routing
• node starts in discovery state• after Td , node
• broadcasts discovery message• enters active state• sets Timer Ta• periodically re-broadcasts discovery
message while in active state• timer can be suppressed by other discovery
messages• after Ta , node returns to discovery state• active node can change to sleep state,
when a higher-ranked node handles routing
3333
Other: DCP (Bluetooth)•• DDataata CCollectionollection PProtocolrotocol (Handy '04)(Handy '04)•• CooperationCooperation StrategyStrategy: Clustering: Clustering
•• Cluster Cluster HeadHead, Cluster , Cluster MembersMembers
•• PeriodicPeriodic Cluster Cluster ReorganizationReorganization•• EnergyEnergy consumptionconsumption•• TopologyTopology changeschanges
•• DCP DCP doesdoes notnot maintainmaintain connectionsconnections duringduringsteadysteady--statestate ((unlikeunlike BluetoothBluetooth scatternetsscatternets!)!)
1
32
cluster head
clustermembers
Cluster
- Cluster head selection- Cluster formation- PFA delivery
Collection of Sensor Data
Setup Phase Steady-State Phase
3434
Other: DCP (Bluetooth)
B
12 3
4
57
6
810
9
11
1. Cluster Head Selection
randomly determined
2. Base station Inquiry
detect CH and 1-hop-CM
3. Base station transmits PFAfirst CMs (1,3,4) then CHs (2)
4. Discovered nodes turn off I-Scan"invisible" mode
5. 1-hop-distant CH inquiryCM and CH discovered by BS are not
detected
6. 1-hop-distant CH transmit PFAfirst CMs (6) then CHs (5,7)
BS-Cluster
12
Abbrevations:BS = Base StationCM = Cluster MemberCH = Cluster HeadPFA = Packet Forward Address
Setup Phase in Detail
3535
Other: DCP (Bluetooth)
Steady State Phase in Detail
•• CM CM transfertransfer sensorsensor datadata to CHto CH
•• CH CH preprocesspreprocess sensorsensor datadata ((datadatacompressioncompression//fusionfusion))
•• CH CH forwardforward aggregatedaggregated datadata to PFA/BSto PFA/BS
B
12 3
4
57
6
810
9
11
BS-Cluster
12
Nodes disconnect immediately• clusters are not limited to piconet size• energy savings for low data rates• reduced interference
(How many Bluetooth piconets fit into a room?)
Periodical or event-triggered transmission schemes are applicable!
3636
Open Research Issues*
•• ExploitExploit RedundancyRedundancy•• TieredTiered ArchitecturesArchitectures•• ExploitExploit spatialspatial diversitydiversity and and densitydensity of of sensorsensor//actuatoractuator
networksnetworks•• AchieveAchieve desireddesired global global behaviorbehavior withwith adaptive adaptive localizedlocalized
algorithmsalgorithms•• LeverageLeverage datadata processingprocessing insideinside thethe networknetwork and and exploitexploit
computationcomputation nearnear datadata sourcessources•• Time and Time and locationlocation synchronizationsynchronization•• SelfSelf--configurationconfiguration•• SecureSecure routingrouting
* taken from: Al-Karaki & Kamal 2004
3737
Summary
•• RoutingRouting protocolsprotocols forfor wiredwired networksnetworks and and adad--hochoc networksnetworks areare notnotapplicableapplicable forfor sensorsensor networksnetworks
•• RoutingRouting protocolsprotocols forfor sensorsensor networksnetworks havehave to to bebe•• energyenergy conservingconserving
•• scalablescalable
•• robustrobust
•• fault tolerantfault tolerant
•• selfself--organizingorganizing
•• Most Most routingrouting protocolsprotocols forfor sensorsensor networksnetworks cancan bebe categorizedcategorized intointo•• flatflat ((datadata--centriccentric))
•• hierarchicalhierarchical
•• locationlocation--basedbased
•• DirectedDirected Diffusion, SPIN and LEACH Diffusion, SPIN and LEACH areare ""ancestorsancestors" of " of manymanymodern modern routingrouting protocolsprotocols forfor sensorsensor networksnetworks