Corso di Reti mobili Reti ad hoc

1
Corso di Reti mobili Reti ad hoc Reti di
Sensori

• S. Chessa

2
Sensor Networks the network

• Ad Hoc Networks specialized for environmental
  monitoring
• Traditional approaches
• Manual deployment (small number of sensors)
• Fixed topology
• Sensed data gathered to a sink node
• Sinks issue requests/commands and receive
  (preprocessed) sensed data
• Sinks provide the user with an interface to the
  network
• Current trend
• Large number of sensors
• Connected as in a multihop ad hoc network
• High network density
• Autonomous when the sinks are disconnected

3
Sensor Networks the network
User
Internet, Satellite Network, etc..
Sink
4
Sensor Networks the sensors

• Each sensor
• Low power, low cost system
• Small
• Autonomous
• Sensors equipped with
• Processor
• Memory
• Radio Transceiver
• Sensing devices
• Acceleration, pressure, humidity, light,
  acoustic, temperature, GPS, magnetic,
• Battery, solar cells,

5
Sensor Networks

• Differences with Ad Hoc Networks
• Number of sensor nodes can be several orders of
  magnitude higher
• Sensor nodes are strongly constrained in power,
  computational capacities, and memory
• Sensor network are denser and sensors are prone
  to failures
• The topology of a sensor network changes very
  frequently due node failures (and mobility?)
• Sensors may not have individual IDs
• Need tight integration with sensing tasks

6
Sensor Networks

• Sensors are deployed in the Sensing Field
• Each sensors samples environmental parameters
• Produces a streams of data
• data stream can be pre-processed locally and then
  forwarded to a sink
• The sinks might be temporarily unavailable
• The network operates autonomously
• Pre-process and store sensed data
• Sensors may implement a database

7
Sensor Networks Applications

• Environmental
• Tracking animals,
• Pollution control,
• Disaster recovery
• Monitor disaster areas,
• Fire/flooding detection,
• Meteorological research
• Security
• Nuclear, Biological and Chemical (NBC) attack
  detection
• Monitoring battlefield,
• Surveillance,

8
Sensor Networks Applications

• Health
• Diagnostics
• Monitoring of physiological data,
• Support for disabled,
• Commercial
• Inventory management,
• Vehicle tracking,
• Toys,
• Domotics,
• Space exploration

9
Sensor Networks

• Sensor networks issues
• Communication is expensive
• Routing is programmed by the sink node
• Programming is focused to specific applications
• Current research
• Point to point geographic routing
• Distributed Data Centric Storage
• Development of general purpose DBMS
• Desiderata
• Guaranteed storage/communication
• Security
• Mobility?

10
Sensor Networks Motes

• Distributed by Crossbow INC. (http//www.xbow.com)
  
• Standard packages
• 8 processing/radio units
• 8 sensing units
• 1 serial programmer
• Operating System (tinyOS)
• Programming environment (nesC)

11
Motes processing / radio

• CPU ATMEL 128L (8 bit, 8Mhz)
• Program memory 128 Kb
• Data memory 4 Kb 512 Kb (?)
• Radio 433 Mhz

12
Motes sensing units

• MTS 300 CA
• Light
• Temperature
• Acoustic
• Sounder

• MTS 510 CA
• Light
• Accelerometer
• Acoustic

13
Motes base station

• MIB 510 CA
• Serial interface
• Able to host MICA boards
• Externally powered

14
Motes Software

• tinyOS
• Developed at UC Berkeley
• Event scheduler minimal C runtime
• Size lt 1 Kb
• Open Source (BSD Licence)
• Many modules (e.g. MAC layers, etc) freely
  available
• nesC
• Event driven language
• Components
• Commands (exported interface)
• Events (imported interface)
• Component-level concurrency
• Compiler for MOTES or emulator
• TinyOS NesC Tutorial
• http//www.tinyos.net/tinyos-1.x/doc/tutorial/

15
Motes Software

• A simple nesC application
• BLINK
• Toggles the red led every second
• module BlinkM
•   provides     interface StdControl 
•   uses     interface Timer    interface
  Leds 

implementation   command result_t
StdControl.init()     call Leds.init()   
return SUCCESS    command result_t
StdControl.start()     return call
Timer.start(TIMER_REPEAT, 1000)     command
result_t StdControl.stop()     return call
Timer.stop()    event result_t
Timer.fired()      call Leds.redToggle()   
return SUCCESS 
16
Motes Software

• configuration Blink implementation  
  components Main, BlinkM, SingleTimer, LedsC 
  Main.StdControl -gt BlinkM.StdControl 
  Main.StdControl -gt SingleTimer.StdControl 
  BlinkM.Timer -gt SingleTimer.Timer  BlinkM.Leds
  -gt LedsC.Leds

17
Sensor Networks Network Layer

• Important considerations
• Sensor networks are mostly data centric
• Attribute-based addressing and location awareness
• Data aggregation can be useful but it might
  prevent collaborative effort
• Energy efficiency is a key factor
• Traditional routing protocols are not practical
• Large routing tables
• Size of packet headers
• Node IDs are less meaningful than their
  capabilities
• From identity-based to data-driven routing

18
Protocols for Sensor Networks

• Drawbacks of flooding-based data dissemination
• The implosion problem
• node A starts by flooding its data to all of its
  neighbors.
• Two copies of the data eventually arrive at node
  D.
• The system wastes resources in one unnecessary
  send and receive.

• The overlap problem
• Two sensors cover an overlapping geographic
  region.
• The sensors flood their data
• Node C receives two copies of the data marked r.
• Resource blindness
• Sensors are not resource-aware they do not
  adapt their activities based on their energy

19
Protocols for Sensor Networks

• Data centric routing

4
2
7
6
3
5
8
6
6
9
8
4
7
8
9
2
Sink
Temperature lt 5
20
Protocols for Sensor Networks

• Location Awareness

Area B
Area A
4
Area C
2
7
6
3
5
8
6
6
9
8
4
7
8
9
2
Sink
Temperature of the sensors in area C
21
Sensor Protocol for Information via Negotiation
22
SPIN

• Heinzelman et Al., MobiCom 99
• Negotiation
• Resource-adaptation
• Negotiation
• ensures that only useful information will be
  transferred.
• use of meta-data to describe data
• Resource adaptation
• Each sensor has a resource manager that monitors
  energy consumption
• Applications probe the resource manager before
  transmitting/processing data

23
SPIN

• SPIN Messages
• ADV When a sensor node has a new data, it
  broadcasts (to its neighbors) an advertisement
  (ADV) packet that describes the new data by using
  meta data
• REQ used by interested nodes to subscribe data
  advertised with ADV
• DATA to send the data to its subscribers
• Resource manager
• If the residual energy of a sensor is below a
  given bound, the sensor does not partecipate to
  the ADV-REQ-DATA protocol.
• However, it will spend energy to receive the ADV
  messages

ADV
REQ
Data
24
SPIN

• Example of data propagation with SPIN

25
SPIN

• Simulations SPIN VS
• Flooding
• Gossiping
• Nodes forward data to a random subset of
  neighbors
• Ideal
• With perfect knowledge of the network topology
  and shortest paths
• Simulation parameters
• 25 nodes
• Each node with 3 data items of 500 bytes each
• Each data item has a distinct name

26
SPIN simulations

• Percent of total data acquired in the system over
  time for each protocol (SPIN, Flooding,
  gossiping, and ideal)

27
SPIN simulations

• Total amount of energy dissipated in the system
  for each protocol (SPIN, Flooding, gossiping, and
  ideal)

28
SPIN

• Drawbacks
• Energy management
• Nodes cannot turn off the radio to save energy
• Need for synchronization
• The same data travels along multiple paths
• Wastage of energy
• Assumes that all the sampled data should be
  delivered to the sink
• No mechanisms to let the sink request for
  specific data
• No local processing of sensed data

29
Directed Diffusion
30
Directed Diffusion

• Intanagonwiwat et Al., MobiCom 2OOO
• Coordination protocol to perform distributed
  sensing of environmental phenomena
• The sensor network is programmed to respond to
  queries such as
• "How many pedestrians do you observe in the
  geographical region X
• "Tell me in what direction that vehicle in region
  Y is moving"
• Directed diffusion is datacentric
• all communication is for named data
• Data generated by sensors is named by
  attribute-value pairs.
• A node requests data by sending interests for
  named data.

31
Directed Diffusion

• Basic elements of Directed Diffusion
• Data is named using attribute-value pairs.
• A sensing task is disseminated in the network as
  an interest for named data.
• The dissemination of interests sets up gradients
• gradients "draw" events (i.e., data matching the
  interest).
• Data matching the interest flow towards the sink
  of interest along multiple paths.
• The sink reinforces one, or a small number of
  these paths.

32
Directed Diffusion

• Interests are named by a sequence of
  attribute-value pairs that describe the task.
• Example of a simple animal tracking task
• type four-legged animal // detect animal
  location
• interval 20 ms // send back events every 20
  ms
• duration 10 seconds // .. for the next 10
  seconds
• rect -100, 100, 200, 400 // from sensors
  within rectangle
• Coordinate may refer to a GPS-based coordinate
  system
• The data sent in response to the interest is also
  named using a similar naming scheme.
• Example
• type four-legged animal // type of animal
  seen
• instance elephant // instance of this type
• location 125, 220 // node location
• intensity 0.6 // signal amplitude measure
• confidence 0.85 // confidence in the match
• timestamp 012040 // event generation time

33
Directed Diffusion

• Interests are periodically generated by the sink
• The first broadcast is exploratory
• The next broadcasts are refreshes of the interest
• Necessary because dissemination of interests is
  not reliable
• Nodes receiving an interest may forward the
  interest to a subset of neighbors
• nodes must be assigned with a unique ID
• Directed diffusion works also in presence of
  multiple sinks

34
Directed Diffusion

• Nodes cache received interests
• Different interests with same time interval,
  area, and type (but, for example, different
  sampling rate) are aggregated
• Interests in the cache expire when the duration
  time is elapsed
• Each interest in the cache is associated with a
  gradient, i.e. the node from which it was
  received
• Gradients might be associated with different
  sampling rate
• Note that the same interest may be received from
  different nodes

Interest propagation
Gradients set up
4
1
5
sink
2
3
6
35
Directed Diffusion

• A gradient is a direction and a data rate
• Gradients are used to route data matching the
  interest toward the sink whom originated the
  interest
• A data may be routed along multiple paths
• Data is routed along a single path if a preferred
  gradient is used
• Examples of data propagation

4
1
5
sink
2
3
6
36
Directed Diffusion

• A sensor node which detects an event matching
  with an interest in the cache
• Start sampling the event at the largest sampling
  rate of the corresponding gradients
• The node sends sampled data to neighbors
  interested in the event
• This information is stored in the gradients
  associated to the interest in the cache
• If a gradient g has a lower rate then the others,
  data along g is sent with lower rate
• Neighbors forward the data only if a
  corresponding interest (with a gradient) is still
  in the cache
• However if that data has already been sent it is
  dropped

37
Directed Diffusion

• Reinforcements
• Used when the sink start receiving data matching
  an exploratory interest from node u
• The sink reinforces node u to improve the quality
  of received data
• Exploratory interests use a low sampling rate
• Reinforces of interests specify an interest with
  larger sampling rate
• In turn, a node receiving a reinforce of an
  interest reinforces one of its neighbors
• Reinforces are propagated through the path along
  which the data flows

38
Directed Diffusion

• Drawbacks
• Assumes that the sink is permanently connected to
  the network
• the network does not operate autonomously
• Sensors do not process data (apart aggregation)
• The sensors just send the data matching the
  interests to the sink
• Does not exploit processing and storage
  capabilities of the sensors
• Flexible network design needs flexible routing

39
Greedy Perimeter Stateless Routing(GPRS)
40
Routing with GPS GPSR

• Karp Kung, MOBICOM 2OOO
• Assumptions
• The nodes are deployed on a two-dimensional space
• Nodes are aware of their position and of the
  position of their neighbors
• For example the nodes are equipped with GPS
• The source knows the coordinate of the
  destination
• Packet headers contain the destination coordinate
• The protocol is scalable
• No need for route discovery
• Few control packets
• Nodes maintain only local information
• Large route caches or routing table are not
  necessary
• Packet headers do not need to store routes

41
GPSR

• GPRS comprises two modes
• Greedy forwarding
• Perimeter forwarding
• Greedy forwarding
• Consider a packet with destination D
• the forwarding node x select as next hop a
  neighbor y such that
• y is closer to D than x
• Among neighbors y is the closest to the
  destination
• Greedy forwarding fails if the packet encounters
  a void

42
GPSR
w
u
void
x
D
v
z
43
GPSR

• Perimeter mode forwarding is executed when greedy
  forwarding finds a void
• Routes around the void
• Based on the right hand rule
• When arriving from y to x
• Selects as next edge the one sequentially
  counterclockwise about x from edge (x,y)
• Traverses the interior of a closed polygonal
  region (face) in clockwise edge order
• Intuitively it explores the polygon enclosing the
  void to route around the void
• In the previous example it would produce x w
  u D

z
2
x
3
1
y
44
GPSR

• However, graph G corresponding to the sensor
  network is a non-planar embedding of a graph
• Edges may cross
• the right hand rule may take a degenerate tour of
  edges that does not trace the boundary of a
  closed polygon
• In the example, from x to v the right hand rule
  produces the path
• x v w u x

w
v
u
z
x
45
GPSR graph planarization

• For this reason GPRS applies the perimeter mode
  to a planar graph P obtained from G
• Relative Neighborhood Graph of G
• Gabriel Graph of G
• Properties
• If G is connected then P is connected
• P is obtained from G by removing edges
• P is computed with a distributed algorithm
  executed along with the perimeter mode packet
  forwarding

46
GPSR graph planarization

• Relative Neighborhood Graph (P) of G
• Edge (u,v)?P iff
• (u,v)?G
• d(u,v) ? Max(d(u,w),d(v,w)) for each w?N(u)?N(v)
• Consider the forwarding node u
• u considers each neighbor v?N(u)
• edge (u,v) is kept iff the above property is
  satisfied

w
u
v
This area must be empty in order to include (u,v)
in P
47
GPSR graph planarization

• Gabriel Graph (P) of G
• Edge (u,v)?P iff
• (u,v)?G
• d2(u,v) ? d2(u,w)d2 (v,w) for each w?N(u)?N(v)
  
• GG constructed with a distributed algorithm as
  RNG
• RNG is a subgraph of GG
• RNG has smaller link density
• RNG or GG are both suitable to GPRS

w
u
v
This area must be empty in order to include (u,v)
in P
48
GPSR graph planarization
Full graph, Gabriel Graph and Relative
Neighborhood Graph
49
GPSR perimeter mode

• Packet header in perimeter mode
• Let x be the node where the packet enters in
  perimeter mode
• Consider the line x-D
• GPSR forwards the packet on progressively closer
  faces on the planar graph, each of which
  intersects x-D

Field Function
D Destination Location
x Location where packet entered in perimeter mode
Lf Point on x-D where the packet entered current face
eo First edge traversed on current face
M Packet mode greedy or perimeter
50
GPSR perimeter mode

• A planar graph has two types of faces
• Interior faces
• Closed polygonal regions bounded by the graph
  edges
• One exterior face
• The unbounded face outside the outer boundary of
  the graph
• On each face GPRS uses the right hand rule to
  reach an edge which crosses x-D (and that is
  closer to D than x)
• At that edge GPRS moves to the adjacent face
  crossed by x-D
• Each time it enters a new face the packet
  records
• In Lf the point on the intersection between x-D
  and the current edge
• In eo the current edge
• GPRS returns to greedy mode if the current node
  is closer to D than x
• Perimeter mode is intended to recover from a
  local maximum

51
GPSR perimeter mode
Greedy mode
D
Perimeter mode
x-D
Lf
F2
F1
x
52
GPSR perimeter mode

• If D is reachable from x (G is connected) then
  GPRS always finds a route
• if D is not reachable
• Either D lies inside an interior face Fi
• Or D lies in the exterior face Fe
• The packet will reach the face (either Fi or Fe)
• Then it will tour around the face until it
  travels the edge eo for the second time
• At that point the packet is discharged

53
GPSR

• GPRS and mobility
• GPRS relies on updated information about the
  position of the neighbors
• It need a freshly planarized graph
• Using stale planarized graph may result in
  performance degradation
• Performing planarization at topology changes is
  not sufficient
• Nodes may move within a nodes transmission range
  
• Proactive approach nodes periodically (at each
  beacon interval) communicate their position their
  neighbors
• This information is used to keep updated the list
  of neighbors and to force planarization

54
GPSR - simulation

• Decreasing the beaconing time the delivery rate
  of GPRS
• Routing overhead (beacon packets) is independent
  of mobility
• Beacons are proactive

Delivery rate
Routing overhead
55
GPSR - simulation

• Path length nearly optimal if the network is
  dense
• 95 of packet delivered through the shortest path
  VS 85 of DSR
• Difference due to the caching of DSR, some paths
  in the cache may be no longer optimal
• Intuitively greedy routing approximates shortest
  paths

56
Data-Centric Storage (DCS) and Geographic Hash
Tables (GHT)
57
DCS GHT

• Ratnasamy et Al., MONET 2003
• Focus
• The sensor network can operate in an unattended
  mode
• Samples and Records information about the
  environment
• Need for
• Data-dissemination techniques to extract data
• Data-centric storage
• Based on
• Geographic routing protocols
• Peer-to-peer lookup systems

58
DCS GHT

• Data-centric storage
• Events are named (keys) and corresponding data
  are stored by names in the network
• Queries are directed to the node that stores
  events of that key
• Two operations supported by DCS
• Put(k,v) stores the observed data according to
  its key k
• Get(k,v) retrieves whatever value associated to
  key k

59
DCS GHT

• Design criteria of a DCS
• Node failures battery exhaustion, HW failures,
• Topology changes due to node mobility, failures,
  
• Scalability number of nodes, network density
• Energy constraints
• Persistence a stored pair (k,v) must remain
  available despite failures and topology changes
• Consistency a query for key k must reach the
  node where pairs (k,v) are actually stored
• Scaling in database size storage should not
  overburden a node as the number of pairs (k,v)
  increase

60
DCS GHT

• Geographic hash table
• Hash(k) returns a pair of coordinates (x,y)
• (x,y) should be included in the network boundary,
  it is assumed this information is known to each
  node
• (k,v) is stored by the node u which is the
  closest to coordinates (x,y)
• Queries related to key k are routed to (x,y)
• GHT built on top of the GPRS routing protocol
• Mobility or failure of u may result in
  unavailability of stored value v
• GHT uses a Perimeter Refresh Protocol (PRP) to
  provide persistence and consistency
• PRP selects one node as home node for key k
• PRP replicates v on the nodes in the perimeter
  around (x,y)

61
DCS GHT

• Home node Home perimeter

source
Home Node
v
Produces value v with key k Hash(k) (x,y)
Home Perimeter
u
w
z
(x,y)
q
s
62
DCS GHT

• Perimeter Refresh Protocol (PRP)
• Accomplish replication of key-value pairs
• GHT routes the packet (k,v) around the perimeter
  enclosing (x,y)
• Where (x,y)Hash(k)
• The perimeter is identified by GPSR
• (k,v) is stored in the home node
• Each node in the home perimeter stores a replica
• Nodes on the home perimeter are said replica
  nodes

63
DCS GHT

• Perimeter Refresh Protocol (PRP)
• The home node u of key k generates periodical
  refresh packets which
• Is sent to coordinate (x,y)Hash(k)
• Note that the home node might have moved
• Contains (k,v)
• Tours around the perimeter around (x,y)
• The refresh packets preserve consistency the
  home node should be the closest to (x,y)

64
DCS GHT

• Perimeter Refresh Protocol (PRP)
• If the refresh packet reaches a node v closer to
  (x,y) than the old home node u
• v generates a new refresh packet towards (x,y)
• It is egligible as a new home node for k
• Else v forwards the packet towards (x,y)
• When the refresh packet reaches its source v
• v becomes the home node for k
• Sets up a refresh timer for k
• Replicate (k,v) in the new perimeter

65
DCS GHT

• Perimeter Refresh Protocol (PRP)
• However the home node may fail
• Need to enforce persistence
• Hence each replica node sets up a takeover timer
• The timer is reset when a refresh packet is
  receved
• When the timer expires the replica node generates
  a refresh packet for (k,v) towards (x,y)

66
DCS GHT

• Perimeter Refresh Protocol (PRP)
• Pairs (k,v) are not cached forever
• If a home (replica) node moves it might not be
  associated to key k anymore
• Discharging (k,v) should not affect availability
• Home and replica nodes use death timers
• Each pair expires after a death timeout
• The death timeout should be larger than the
  refresh and takeover timeouts

67
DCS GHT

• Perimeter Refresh Protocol (PRP)

v (replica)
r (replica)
u (home)
w (replica)
(x,y)
s (replica)
After takeover time w generates a refresh
q (home)
z (replica)
q becomes the new home node
68
DCS GHT

• A home node for key k might be overburdened if
  too many values with key k are produced
• Structured replication (SR)
• uses a hierarchy (of depth d) of event names
• Hash(k) is the root of the hierarchy
• To each key k are associated a root and 4d-1
  mirrors
• A node stores the pair (k,v) to its closest
  mirror of Hash(k)
• Retrieval of a value involves queries to the root
  and all mirrors
• Trades storage overhead with communication

69
DCS GHT

• An example of Structured Replication with d2

(0,100)
(100,100)
Root (3,3)
Level 1 Mirrors
Level 2 Mirrors
(0,0)
(100,0)
70
DCS GHT

• Data Centric Storage based on Geographic Hash
  Table
• nodes should be aware of their coordinates
• Nodes should know the network boundary
• Built on top of GPRS
• Perimeter Refresh Protocol to enforce persistence
  and consistency
• Structured Replication to enforce scaling in
  database size

71
Physical and virtual Coordinates
72
Geographic Routing and Localization

• Traditional routing protocols for ad hoc networks
  are not practical
• Large routing tables or path caches
• Size of packet headers
• Geographic routing appears to be the best option
• Coordinates are mandatory
• To support geographic routing
• Support the implementation of a data centric
  storage (DCS with GHT)
• Provide a relation between sensed data and
  locations

73
Sensor Coordinates

• Coordinates can be obtained by equipping nodes
  with GPS
• Additional cost
• Not always feasible (for example indoor)
• When no GPS system is available
• Either a few anchor nodes know their position
• other nodes compute coordinates with a variety of
  methods
• Or virtual coordinates are used
• typically they are hop-distances dependent

74
Sensor Coordinates without GPS

• Approximate physical coordinates of the sensors
• A few nodes (anchors) know their exact position
• by means of special hardware (GPS)
• or because they are manually positioned in well
  known locations
• The anchors broadcast beacons
• The other nodes estimate their positions with
  distributed protocols
• Nodes may estimate their distance from the
  anchors
• Time of arrival techniques
• Signal strength
• Or they may estimate their position based on
  triangulation
• angle of arrival

75
Range and Range-Free techniques

• Two basic techniques
• Using special Ranging hardware (e.g. signal
  strength, T.O.A., etc).
• Range Free Technique.
• Cost-Effective No special hardware for
  ranging.
• Topology based (Hop counting) techniques.

76
Range techniques

• Drawbacks
• Need special antennas to estimate angle of
  arrival
• Signal strength or time of arrival techniques may
  be affected by external perturbations
• Interferences
• Humuidity
• Evaluation of the coordinate system accuracy

77
Range-Free Techniques

• Virtual coordinates
• Unrelated to the physical coordinates of the
  sensors
• Typically based on hop distances
• Can support efficiently geographical routing
• With sparse networks might be better than
  physical coordinates
• Correspondence between virtual and physical
  coordinates left to the sink node

78
Sensor Coordinates without GPS

• Geographic routing without location information
  (Rao et al.) MOBICOM 2003
• Investigates three cases
• Perimeter nodes are know they know their
  location
• Perimeter nodes know they are on the perimeter
  but they dont know their location
• Nodes know neither their location, nor whether
  they are on the perimeter
• For each case they give a protocol which assign
  virtual coordinates

79
Sensor Coordinates without GPS

• Perimeter nodes are known they know their
  location
• Given node i let
• Ni be the set of its neighbors
• xi its x-coordinate
• yi its y-coordinate
• Initially each node (except perimeter node) is
  assigned coordinate (100,100)
• Node i approximates its virtual coordinates
  iteratively
• xiSUM(xk k? Ni) / Ni
• yiSUM(yk k? Ni) / Ni
• The iteration is repeated d times

80
Sensor Coordinates without GPS

• Perimeter nodes are known they know their
  location
• After d iterations evaluate
• success rate of greedy routing over the virtual
  coordinates
• Average path length

• Initial position of nodes
• After 10 iterations
• After 100 iterations
• After 1000 iterations

81
Sensor Coordinates without GPS

• Perimeter nodes are known they know their
  location
• Simulation with d1000 (and 16 neighbors per
  node)
• Virtual coordinates
• Greedy routing success rate 99,3
• Average path length 17.1
• Physical coordinates
• Greedy routing success rate 98,9
• Average path length 16,8
• It is not necessary that all perimeter nodes
  participate to the protocol
• If only 8 perimeter nodes participate
• Greedy routing success rate 98,1
• Average path length 17.3

82
Sensor Coordinates without GPS

• Only perimeter nodes are known
• Each perimeter node broadcasts an HELLO message
  to the entire network
• Each perimeter nodes knows its hop distance with
  the other perimeter nodes
• This vector distance is the perimeter vector
• Each perimeter node broadcasts its perimeter
  vector to the entire network
• Each perimeter node knows the hop distance
  between any pair of perimeter nodes

83
Sensor Coordinates without GPS

• Only perimeter nodes are known
• Each perimeter node computes a triangulation to
  compute the virtual coordinates of the other
  perimeter nodes
• Such as to minimize
• SUM i,j in the perimeter (hop-dist(i,j)
  dist(i,j))2
• Where dist(i,j) is the euclidean distance over
  the assigned coordinates
• Then the previous protocol is applied
• However the nodes can be assigned initial
  coordinates taking into consideration the
  information available from the previous steps

84
Sensor Coordinates without GPS

• Only perimeter nodes are known
• With d10 (and 16 neighbors per node) achieve
  same performance than previous protocol with
  d1000
• Greedy routing success rate 99,2
• Average path length 17.2
• Due to a better initialization of non-perimeter
  nodes

85
Sensor Coordinates without GPS

• No Location Information
• Add a preliminary phase
• Two bootstrap nodes broadcast a beacon
• The nodes that, within two hops, are the farthest
  from the bootstrap nodes are classified perimeter
  node
• Applies the same protocol as before
• With d10
• Greedy routing success rate 99,6
• Average path length 17.3

Example of virtual coordinates
86
Sensor Coordinates without GPS
Success rate of greedy routing with virtual and
physical coordinates
87
Sensor Coordinates without GPS

• The protocols are resilient to message losses
• Due to redundancy of information in the perimeter
  vectors
• The virtual coordinates can be mapped onto a
  circle
• Gives well defined area for implementing a
  distributed hash table

88
The Virtual Coordinate Assignment protocol
89
Virtual Coordinate System

• A virtual coordinate system is defined by
  electing three anchor nodes (X,Y,Z).
• Each coordinate is a triplet (x,y,z), x is the
  minimum hop distance between the node and the X
  anchor.
• The coordinates are virtual because they are not
  related to the physical position of a node
  (Euclidean coordinates).
• Different nodes may have the same coordinates.

90
Virtual Coordinate System
Y
(x,y,z)
Z
X
91
Why three anchor?
Virtual Coordinate System

• A zone is a set of nodes sharing the same triplet
  of coordinates
• With less than three anchors a zone might be
  split into two distinct regions.

92
The zones
Virtual Coordinate System

• With the virtual coordinate system, the routing
  protocol exploits geographic routing to reach the
  destinations zone and proactive routing within
  the zone.
• Hence, in order to efficiently support routing,
  the size of a zone must be small.
• The size of the zones depends on the position of
  the anchors and on the network density
• We show that the zones are minimized if the
  anchors are as far as possible from each other
• The anchors must be close to the network border

93
Virtual Coordinate Assignment Protocol
Virtual Coordinate System

• The sink node begins the VCap protocol
• It broadcasts a W_SET message containing an hop
  counter to identify nodes on the network border
• Nodes in the network use the hop counter in the
  W_SET message to determine their hop distance
  from the sink
• Anchor X is elected by a broadcast-based
  distributed protocol
• Nodes with maximum hop distance from the sink (in
  their local neighborhood) are eligible as X.

94
Virtual Coordinate Assignment Protocol
Virtual Coordinate System

• The election of anchor X
• Among the nodes with maximum hop distance from
  the sink, it is elected as X the node with
  maximum ID.
• The election of anchors Y and Z exploit a similar
  protocol
• Heuristics enforcing the property that X, Y and Z
  are as far as possible from each other are used.
• Example Z should be on the border and
  equidistant from X and Y

95
Virtual Coordinate Assignment Protocol
Virtual Coordinate System

• Feasible positions for anchor Z

96
Hop Distance V.S. Euclidean distance
Virtual Coordinate System

• A node coordinate express the minimum hop
  distance between the node and an anchor
• Let ? be the network density, and r the
  transmission range.
• Let h and D be the minimum hop distance and
  Euclidean distance between two nodes,
  respectively.
• Let l(h), u(h) such that l(h) D u(h).
• Assumption

97
Bound to the size of a Zone
Virtual Coordinate System

• Feasible locations for an anchor, given two nodes
  (a and b) in the same zone at distance d

r
a
b
a
d
b
d
a
a
d
b
b
98
Bound to the size of a Zone
Virtual Coordinate System

• In a circular sensor network of diameter R, where
  the anchors are placed on the vertexes of an
  equilateral triangle inscribed in R.
• The maximum distance d between nodes in the same
  zone satisfy

99
Simulative analysis of the zone size

• Maximum distance d between nodes in the same
  zone in circular networks with R1 and r 0.05.
• The figure plots the minimum, average and maximum
  of dN in units of r with densities in the range
  of 10 to 70. (min(dN)/r, E(dN)/r, max(dN)/r).

For low density the nodes of a zone are
dispersed. As density increases, d approaches
8/3r.
100
VCAP Overhead

• Memory Overhead
• Each node stores only the values of the x,y, and
  z coordinates.
• Communication Overhead
• Step 1 (W_SET) is a simple network broadcast.
• Steps 2,3 and 4 (election of X,Y,Z) are also
  based on broadcast.
• The overhead of each step can be reduced trading
  communication with latency.
• A pragmatic approach exploits random delays of
  the messages depending on the ID of the
  forwarding node.
• Large delays reduce the overhead but increase the
  protocol latency.

101
Coordinate Smoothing

• Coordinate smoothing
• Each coordinate is re-computed as the average of
  the neighbors coordinate
• Increases the resolution of the coordinate system
• By reducing the size of the zones
• Improves the performance of greedy routing
• Reduce the effect of errors (due to packet
  losses) in the coordinate system

102
Coordinate Smoothing

• Fraction of nodes which share the same coordinate
  with at least another node

VCap
VCap one smoothing step
103
Simulation of geographic routing over virtual
coordinates

• Simulation parameters
• Circular networks of radius R100 Transmission
  Range r 8. Network Density (average neighbors
  for node) ranges from 8 to 50.
• Simulation of a greedy routing algorithm
• Statistics
• Degree of reachability (the possibility of
  setting a source-dest path)
• Average path lengths.
• Samples size of 1000.

104
Effect of smoothing
Reachability with greedy geographic routing using
virtual coordinates.
105
Reachability
Reachability with greedy geographic routing using
virtual (VCap), Eucledian (PHY), virtual one
smoothing step (SVC) coordinates
106
Reachability
Reachability at low densities with greedy
geographic routing using virtual (VCap),
Eucledian (PHY), virtual 1 smoothing step
(SVC).
107
Average path length
Average path length with greedy routing.
108
Detours and greedy routing
Reachability with greedy routing and detours with
virtual coordinates and one smoothing step.
109
2-hop lookup greedy routing
Reachability with 2-hop lookup greedy routing
with virtual coordinates and one smoothing step.
Circular domain with R 100, r 8
110
2-hop lookup greedy routing
Average path length with 2-hop lookup greedy
routing with virtual coordinates and one
smoothing step.
111
Packet overhead
Packet overhead per node (network with 300 nodes)
112
Effect of packet loss
For each node, number of neighbors with at least
one coordinate error
113
Simulation Outcomes

• The reachability approaches 100 as the network
  density increases.
• Reachability with virtual coordinates is
  comparable to reachability with Euclidean
  coordinates.
• With very low density routing on virtual
  coordinates performs better!
• The average path length with virtual coordinate
  is slightly larger.

114
Conclusions and Future Works

• Goals achieved
• A distributed Virtual Coordinates Assignment
  Protocol (VCap) for large scale sensor networks.
• The resolution of VCap (the maximum diameter of a
  zone) is independent of the network size. We
  studied its relationship with network density.
• Geographic routing over virtual coordinates has a
  good degree of reachability compared to Euclidean
  coordinates. Especially in dense networks.
• Virtual Coordinates do not require GPS equipped
  nodes.
• Future Works
• Design of a geographic routing protocol with
  guaranteed delivery over virtual coordinates.
• Virtual coordinates with (low) mobility

115
Data Management in Sensor NetworksThe MaD-WiSe
Aproach
116
MaD-WiSe

• Aims
• Design a database for sensor networks
• The sensor network is programmed by high level
  SQL queries
• The queries are optimized and translated in a
  query plan
• The query plan assigns subqueries to individual
  sensors
• Eash sensor comprises a query processor which
  executes its assigned query

117
MaD-WiSe

• MaD-WiSe is implemented in NesC on TinyOS
• designed as a three layer software
• Network
• Transport (Stream System)
• Query Processor

118
MaD-WiSe Network layer

• Based on the Virtual Coordinate assignment
  protocol
• Exploits a geographic greedy routing protocol
• Provides connection-less and connection-oriented
  services
• (planned) implementation of application-driven,
  energy efficiency strategies

119
MaD-WiSe Stream System

• Provides support to management of data streams in
  the network
• Streams can be
• Sensor streams (corresponding to a local
  transducer)
• Local streams (used to write/store processed
  data)
• Remote streams (used to access data
  sensed/processed by remote sensors)
• Implemeted over the connection-oriented services
  of the network layer
• Interface to the upper layers
• Open and Close streams
• Read and write from/to streams

120
MaD-WiSe Query Processor

• Implements the operators of the algebra
• Such as join, select, average,
• Implemented over the Stream System