Docstoc

Computer Networks Wireless Sensor Networks

Document Sample
Computer Networks Wireless Sensor Networks Powered By Docstoc
					     CS 78 Computer Networks

     Wireless Sensor Networks

             Andrew T. Campbell
          campbell@cs.dartmouth.edu




                                      Application Layer


  our focus                           Transport Layer

• What we will lean
                                          Network
  – Applications
  – Unidirectional Transports
     • Data dissemination (many-to-      Link Layer
       one
     • Reliable transport (one-to-
       many)
                                       Physical Layer
  – Congestion Control
  – Type of traffic




                                                          1
  What are sensor networks?
• Put simply,
  – Sensing, Computing, and Communications
• They connect us to the physical world
  around us
• They represent a new frontier in
  computer networks




        Application domains
• Smart buildings
• Industrial, structural, environmental
  monitoring systems
• People-centric applications
  – Assisted living, health care
  – Urban sensing
• Robotics, cars




                                             2
                Characteristics

  •   Long-lived, mostly-static
  •   Application-specific
  •   Multi-hop wireless
  •   Very energy-constrained
  •   Mobility not an issues of driving factor
  •   People out of the loop




                   Some Sensors




Types of sensors
Seismic, temperature, light, accelerometers, magnetometers,
humidity, chemical, acoustic, image, etc.




                                                              3
   Relationship between sensing
    and communications range
                  sensing
                   range           Zigbee




                            WiFi




                            celluar




    Great Duck Island Experiment, 2002
         (David Culler, UCB/Intel)




http://www.greatduckisland.net/




                                            4
    Duck Island Data




At the Extreme: Smart Dust




                             5
        Communication and Sensing Stack
                             Sensing (temperature/light, etc.) Application Layer



our focus          Data Dissemination/ Reliable Transport/ Congestion Control



                   Multi-hop Routing Layer


       packet      Messaging Layer



                   Radio Packet


            byte   Radio Byte                                    Logger     Temp   Photo
                                                                                           SW


            bit    RFM                    CLOCK         POT               I2C      ADC     HW




       Why is radio different from a
                  wire?




                                                                                                6
        Wireless radio channel
• Shares spectrum with other (interfering) transmitters
• Low power, low range, low bandwidth
• High error rates
• Medium is susceptible to noise, interference,
  blockage and multipath and these channel
  impediments are time-varying
• These random variations distort and diminish the
  received signal due to physical and environmental
  factors non-existent in wireline communications
       • multipath fading, noise and attenuation issues




         Multipath propagation
• The radio signal can get severely distorted at
  the receiver
    – Reflection (objects larger than wave length),
      diffraction (shadow fading) and scattering (objects
      smaller than wave length)
                Wall




                                                       Scattering



                                    Filing
                                   Cabinet

           Transmitter   Diffraction (Shadow Fading)                Receiver



                           Reflection




                                                                               7
                                Signal fading
• The following received signal power results in higher
  BER
      • path loss
                 – determines how the average received signal power
                   decreases with distance between the transmitter and
                   receiver
      • shadow fading
                 – characterizes the signal attenuation due to obstructions
                   from building and other objects
      • Raleigh fading (fast fading):
                 – the rapid fluctuation caused by local multipath
                                              Short-term Fading r(t)
         Signal Strength (dB)




                                                      Long-term Fading m(t)




                                  T T
                                                                       Time (t)




 You end up with delay spread,
   i.e., symbol interference




                                                                                  8
                 Signal Fading Contd.

     Received Signal
       Power (dB)

                                                                 With path loss,
                                                                 shadow fading,
                                                               and Rayleigh faiding


                                                        Path loss
                                                                                      Shadow fading


                                                                                      Rayleigh fading




                                                                                        log (distance)




  Sharing the air using media
            access
• CSMA (carrier sensing media access) is
  widely used
• Problems
  – Hidden terminal problems, near-far



             A                     B                                        D

                                                    C




   Cell: Coverage Range, range over which a node can transmit and receive data reliably.




                                                                                                         9
 Resolving Hidden Terminals
• Using RTS-CTS with CSMA
               RTS
      A                    B       C   D




                     CTS
           A                   B   C   D




               DATA
                           B       C   D
       A




     Data Dissemination,
specifically, directed diffusion,
   which is a many-to-one
 unreliable transport protocol




                                           10
Self-Organizing Sensor Networks




 sensors


       sink        Internet




Self-Organizing Sensor Networks
                   active sensors




Forwarding nodes


  QOS-Point        Internet




                                    11
         Directed Diffusion (interest)




interest message
    is flooded




      Directed Diffusion (data events)




  data events
    flooded




                                         12
       Directed Diffusion (reinforcement)




  paths periodically
reinforced (gradient)




      Directed Diffusion (reinforced data
                    events)




                                            13
       Common Traffic Patterns

•   Periodic Sensing (CBR-like)
•   Event Sensing (discrete events)
•   Impulse Sensing (bang-bang)




        Traffic Patterns (periodic)




                                      14
Traffic Patterns (periodic)




Traffic Patterns (periodic)




                              15
         Traffic Patterns (event)



                                  boom




       Traffic Patterns (impulse)



  earthquake events           fire events




disaster recovery events   contamination, bio-
                             hazard events




                                                 16
        Data Aggregation




        Data Aggregation




    4
         6


2

             4




                           17
        Data Aggregation




        Data Aggregation




    4
         6


2

             4




                           18
    Congestion problem in sensor
             networks




          Impact of the Funneling Effect




       Funneling Impulses

•traffic funnels (more energy consumed)
•traffic intensifies (more congestion)
•greater packet loss (more energy waste)
•more aggregated packets transit (higher energy packets, value increases)




                                                                            19
                    Impact of Funneling Effect


                                                                  data impulses
•Results in a significant
waste in energy                                  periodic and
•Shortens operational                            event
lifetime of network




   •    ns-2 simulation using directed diffusion
         –    30-node sensor network, 2 Mbps IEEE 802.11
         –    6 active sources randomly selected
         –    3 sinks uniformly distributed across sensor field
   •    Caveat: results are based on a CSMA MAC




        Congestion Challenges for Sensors
       • Varying degrees of congestion likely even for low to
         moderate traffic rates
       • Impulse data particularly harmful
             – Sudden, large, correlated impulses of events
             – Difficult to model and control
             – During Impulses data of most important
       • Congestion impacts application fidelity
       • Lost events contribute to energy waste and
         shortened lifetime of the network
       • Could lead to the equivalent of congestion collapse of
         the sensor network




                                                                                  20
                         Congestion Scenarios




(a) dense sources, high rate                        (b) sparse sources, low rate




  active source
  persistently congested node

  transiently congested node

  non-congested node
  sink node                     (c) sparse sources, high rate




     Heterogeneous Sensing Networks
• Reality check
   – In reality, sensing networks will not be
     homogeneous with a single device, common
     application
• Rather,
   – Heterogeneous devices
   – multiple sensing applications will simultaneously
     run in the network
   – Mixed traffic: all traffic types (continuous, event
     and data impulse traffic) in a single network
• Comment: complex problem space




                                                                                   21
          Complex Sensor Networks
                                                          • Multi-sink scenario
                                                               – Aggregators cause micro-funnel
                                                                 effect
                                                          • Mixed congested funnels
                                                               – dense sources, high rate
                                                               – sparse sources, low rate
                                                               – sparse sources, high rate
                                                          •   Funnel “cross-talk”




        Different Objective Functions
               Effective Throughput




                                                  Ideal
               or power (T/D)




                                      Congested
                                       region

                                         Offered load
• Existing objective function
    – Maximize throughput, fairness concerns, etc.
    – May not relevant in sensor networks?
• New objective function should:
    – Balance between offered load and fidelity during congestion
        • Converge on reporting rate that is just sufficient to meet performance or
          fidelity of the sensing apps
        • Meet fidelity criteria while minimizing energy (communications and
          computation) and max network lifetime




                                                                                                  22
                              Metrics

• Energy Tax
   – The overall cost of the network to deliver a single event to the sink
   – # of packet dropped normalized to the # of packet received at the
     sink
• Fidelity Penalty
   – Fidelity: data quality as seen from the application
   – Amount of fidelity degradation tradeoff to avoid congestion
   – # of event received normalized to the # event received in an
     idealized scheme




     Congestion Detection and
       Avoidance (CODA)




                                                                             23
                           CODA protocol
  • Receiver-based congestion detection
      – Accurate and energy-efficient (measurement based) channel monitoring
  • Hop-by-hop backpressure
      – Open-loop, fast timescale (immediate feedback) mechanism that is
        responsive to transient congested conditions
      – Node congestion policy, e.g., regulate rate such as AIMD, drop, priority,
        aggregate
  • Sink-to-multisource regulation
      – Closed-loop, slow timescale (coarse feedback) control mechanism that is
        responsive to persistent congested conditions
      – Congested sources opt in and out of sink control in a distributed manner
      – Computationally powerful sink node monitors application fidelity and
        respond accordingly




                   Congestion Detection
• Combination of measured channel loading
  conditions and buffer occupancy
• Where? receiver-based is energy efficient
• Low cost (energy efficient) sampling
  techniques for measuring channel load
   – Must listen anyway, carrier sense for reception of
     packets during duty cycle
   – A receiver/forwarding node’s queue is not empty
     => measure only at this time                         load sensing epoch (start
   – Stop sensing epoch when packet is transmitted        on reception of packet)
                                                          radio off (mostly off, S-
                                                          MAC, SPAN)
                                                          listening mode (low duty
                                                          cycle)




                                                                                      24
        How do we detect congestion?


                                 •   CSMA theoretical throughput upper
                                     bound

                                                     1       !C
                                         Smax #          (" = )
                                                  1+ 2 "      L

                                 • Indication of high probability of
                                   congestion

                                         $ " #S max (# ! 1)
 Once congestion is detected
 nodes signal their upstream             buffer occupancy≈1
 neighbors via backpressure




                    Measuring the β Value
                A       B

                            β
        2β




                   1       !C
       Smax #          (" = )
                1+ 2 "      L
•50 iterations
• β = 0.030 +/- 0.003 (95% confidence level)
•Therefore, predicted Smax (channel throughput upper
bound) ~ 70%




                                                                         25
Hop-by-Hop Backpressure Algorithm
• A node broadcasts backpressure messages as long as it
  detects congestion
   – If congestions persists at a node it keeps sending backpressure
     messages (could get lost) for N times
• Backpressure messages propagate toward sources
   – However, source nodes may not receive backpressure
   – In dense networks likely to propagate directly to sources
• Based on localized congestion policy nodes take action
   – Regulate sending rate based on some policy (e.g., AIMD) priority
     drop (e.g., plain events over aggregates)
   – Nodes receiving backpressure messages only propagate the
     message if they locally also detection congestion




      Depth of Congestion Indicator

• Depth of congestion (M) indicator
    – The number of hops (M) a backpressure
      messages traverses before a non-congested node
      is encountered
    – Represents the instantaneous depth of congestion
      metric
• Routing
    – Indicator for routing protocol to select better path
• Load balancing
    – Silently suppress routing or diffusion signaling




                                                                        26
         Energy Tax and Fidelity Penalty
•source rate impulses 8pps, 4pps, 7pps   •3x reduction in energy tax with
                                         minimum fidelity penalty for CODA




•Backpressure implemented with multiplicative rate reduction
•CODA’s rate control impacts fidelity but minimizes this penalty
•CODA’s priority scheme favors “chosen node” Src-2 (source node 2)




            Sink-to-Multisource Regulation

    • Closed loop control over slow timescale
    • The sink only regulates source nodes that are
      responsible for, or impacted by, persistent
      congested conditions
    • Closed loop regulation operates over slower
      timescale where the sink is capable of
      asserting control over multiple heterogeneous
      sources
    • The sink is also best placed to understand the
      fidelity rate of the received signal and take
      application specific actions, if needed




                                                                             27
              Regulation Algorithm

• When a source rate (r) is less than some fraction (λ)
  of the maximum theoretical throughput (Smax) the
  source regulates itself
• When this value is exceeded a source is more likely
  to contribute to congestion and triggers (sets the sink
  regulation bit in packets) sink regulation
• At this point a source requires constant slow
  timescale feedback (via ACKs from the sink) to
  maintain its rate
   – The reception of ACKs at the sources serves as a self-clocking
     mechanism allowing sources to maintain their current rates (e.g., 1
     ACK per 100 events at the sink)
   – Failure to receive one or more ACKs forces sources to reduce their
     rate based on some policy (e.g., AIMD)




          Regulation: Sink Actions

• Application specific actions such as only sending
  ACKs along certain reinforced paths
• Stop sending ACKs based on its own
  measurements of its local channel loading
  conditions
• Because sink measures fidelity it can take
  application specific actions when the reporting rate
  is less then desired fidelity




                                                                           28
When does a source drop out of sink
           regulation?
• If the sources’ channel load drops to an acceptable level (r< λ
  Smax) or the sources are reinforced with lower rates (e.g., via
  directed diffusion reinforcement) then they start to regulate
  themselves again
• The sink maintain information based on application specific data
  types and can “force” (via signaling) sources to lower their
  threshold value (λ)
    – This forces selected sources to trigger sink regulation before others
      do.
    – This provides an implicit source priority scheme as part of the sink
      regulation mechanism




One-to-many reliable transport
 protocol, specifically, Pump-
Slowly, Fetch-Quickly (PSFQ)
           transport




                                                                              29
        Traditional Viewpoint:
  “Sources-to-Sink Communications”
                                          Query
Data Dissemination
                      Sensor
                                          Data Sink

                      Event
           Interest
                               Interest




                       Occasional loss is OK




                       Reliable Transports

  • Signaling to change
    specific parameters on
    sampling, compression,
    reporting rates
  • Demand loading new
    algorithms
  • Low frequency updates
  • Must work with existing
    app and minimize
    contention




                                                      30
           Transport Design Issues
• Reliable data applications for sensor networks
   –   Reliable signaling messaging
   –   Reconfiguation of sensors in the field
   –   Management of sensors
   –   Retasking and reprogramming of sensors
• Challenges
   – One to many reliable delivery paradigm
   – Wide range of error conditions experienced
   – Application specific challenges, others..




  “Sink-to-Sources” Communications

                                Pump
                Sensor
                                                     Loss Event
                                 Data Sink
             New Execution
                image
                                                  Fetch

                                                   Reliable
                                                   Multicast?

          Occasional loss is disastrous!




                                                                  31
         PSFQ Transport
    (Pump Slowly, Fetch Quickly)
• Ensure reliable delivery with min. support
  from infrastructure ~ non-IP.
• Minimal overhead/signaling
  – lost detection, recovery
• High error tolerance
  – Work efficiently under low and high packet loss
• Scalable and energy efficient
  – Localization, aggregation, multimodal operations




       PSFQ Protocol Features

• Power efficient Negative ACK (NACK)
  system – don’t propagate
• Power efficient local error recovery
  – Hop by hop error recovery – not e2e
• 1->N reliable delivery
• Built on a fundamental relationship
  between pump and fetch
• Multi-modal communications operation




                                                       32
End-to-End Error Recovery


                              (1-p)n



                            Prob. to detect
                            loss in NACK
                            system.




Fetch/Pump Relationship




                                              33
             Multi-Modal Operations

• The impact of loss propagation
                                                 A           B            C




                                                 A           B            C




                                                 A           B            C



• In-sequence data forwarding.
• “Multihop forwarding” vs. “Store-and-forward”.




                  PSFQ Operations
• Pump Operation (at sender)
   – Application specific part (timely dissemination of code fragments)
   – Basic rate based flow control
   – Avoid redundant messaging and minimize contention/collisions
     over shared wireless channel (critical in dense networks)
   – Localize loss: avoid propagation of loss events to downstream
     nodes
• Fetch Operation (at receiver)
   – Local loss detection and recovery based on pump/fetch ratio
   – Data and control could be lost
   – Multiple loss windows managed as “loss aggregation” – helps to
     minimize signaling/energy cost
   – Loss gap windows can be serviced by multiple nodes
   – Cancel NAK if receiver “overhears” NACK for same segment from
     neighbors
   – Avoid NAK implosion problem NACKs are not propagated
   – Proactive fetch for end of file conditions
• Aggregated Reporting (send/ receivers) - optional
   – Health of the receivers/ network




                                                                              34
                   Pump Operation

• Timers – Tmin, Tmax for rate base flow control
• Packets delayed for a random period between Tmin,
  Tmax before broadcast
• TTL/data centric addressing – no end node
  addresses – all/group/clusters sensors addressable
• Tmin
    – Provides Time-buffer for local recovery operations
    – Suppress rebroadcast [mobicom99] operation
        • One rebroadcast 60% additional coverage; four = < 1%
• Tmax – loose delay bounds
  D(n) = Tmax × n × (Number of hops)
• Cached data packets
   – may be application specific (all or buffer cache)




                   Fetch Operation

• Timers
   – Tr (<< Tmax)
   – Tmax/Tr ~ Pump/Fetch ratio
• Loss Aggregation – windows of loss.
• Proactive Fetch
   – Loss of last segment
   – Loss of all segment
   – How long should wait before proactive
     fetch?




                                                                 35
       Retasking Example




2Mbps, CSMA/CA channel access.
Tmax is 100ms, Tmin is 50ms and Tr is 20ms




          Error Tolerance




                                             36
Average Latency




                  37
38

				
DOCUMENT INFO
Shared By:
Categories:
Stats:
views:57
posted:7/30/2010
language:English
pages:38
Description: What are sensor networks? • Put simply, – Sensing, Computing, and Communications • They connect us to the physical world around us • They represent a new frontier in computer networks. Wireless radio channel • Shares spectrum with other (interfering) transmitters• Low power, low range, low bandwidth• High error rates• Medium is susceptible to noise, interference, blockage and multipath and these channel impediments are time-varying • These random variations distort and diminish the received signal due to physical and environmental factors non-existent in wireline communications • multipath fading, noise and attenuation issues