Document Sample
Routing Powered By Docstoc
					                          Lecture 4:

                            Anish Arora


Introduction to Wireless Sensor Networks

  Material uses slides from Wattenhofer,
                           Gouda, Estrin
                           Routing Overview

•   Patterns:
       Convergecast
          one shot subscription or persistent subscription
          subscriber in-network or from base station
          if in-network and one-shot subscriber, then subscriber could be moving
       Broadcast
          potentially directed/local
          potentially with feedback (PIF)
          potentially scoped (e.g. data centric routing)

                          Routing Overview

•   Model assumptions:
       Availability of locations
       Density/planarity
       Node/link heterogeneity

•   Requirements:
       Latency
       Reliability
       Energy
       Scalability
       Convergence

        Convergecast Protocol Classification

•   Distance vector protocols
       Key issues:
          Link selection
          Route metric:
             o   Expected number of transmissions on path
             o   Expected transmission time
             o   Distance advanced towards destination

•   Greedy protocols: issue of dealing with holes
•   Geometric protocols
•   Randomized protocols
•   Gradient-descent protocols
•   Multi-path protocols, even flooding
•   Hierarchical protocols (potentially exploiting clusters)
Location-based/Geometric/Geographic Convergecast

   •   Sensor nodes addressed according to their locations

   •   No routing tables stored in nodes!

Kleinrock et al.        MFR et al. Geometric Routing proposed

Kranakis, Singh,        Face      First correct algorithm
Urrutia                 Routing

Bose, Morin,            GFG       First average-case efficient algorithm
Stojmenovic, Urrutia              (simulation but no proof)
Karp, Kung              GPSR      A new name for GFG

Kuhn, Wattenhofer,      GOAFR     Worst-case optimal and average-case
Zollinger                         efficient, percolation theory
    Correct Geometric Routing: Face Routing

•   [Kranakis, Singh, Urrutia CCCG 1999]

                     Face Routing

•   Remark: Planar graph can easily (and locally!) be
    computed with the Gabriel Graph, for example

    Face Routing

s                  t

    Face Routing

s                  t

    Face Routing

s                  t

    Face Routing

s                  t

    Face Routing

s                  t

    Face Routing

s                  t

    Face Routing

s                  t

                     Face Routing Properties

•   All necessary information is stored in the message
     Source and destination positions
     Point of transition to next face

•   Completely local:
     Knowledge about direct neighbors’ positions sufficient
     Faces are implicit

                                   “Right Hand Rule”

•   Planarity of graph is computed locally (not an assumption)
     Computation for instance with Gabriel Graph
                  Efficiency in Face Routing

•   Theorem: Face Routing reaches destination in O(n) steps
•   But: Can perform poorly compared to the optimal route

•   Need to bound search area adaptively                      17
                          Grid Routing

Key ideas:
    Embeds tree in logical grid

    Well suited for bursty convergecast traffic
       Avoids fast link reliability estimation
           o   Preselects innerband links
        Focuses only on up/down link detection
        Attempts to spread load uniformly
           o   Parent chosen randomly and rotated periodically

    Deals with holes randomly
       Cycles avoided by limiting number of diversions

    Base station snoops

                            The Logical Grid

•   The motes are named as if they form an M*N logical grid
•   Each mote is named by a pair (i, j) where
                  i = 0 .. M-1 and j = 0 .. N-1
•   The base station is mote (0,0)
•   Physical connectivity between motes is a superset of their
    connectivity in the logical grid:
          (0,1)                           (0,1)   (1,1)   (2,1)


                                          (0,0)   (1,0)   (2,0)
          (0,0)           (1,0)

                         Potential Parents

•   A mote (i, j) dominates another mote (x, y) iff i≥x and j≥y

•   If (i, j) dominates (x, y), then distance from (i, j) to (x, y) is

•   Let H be a “small” positive integer, called the hop size

    A potential parent of a mote (i, j) is a mote (x, y) such that
      (i, j) dominates (x, y) and
      distance from (i, j) to (x, y) = H
       (except in special cases where (i,j) is close to some
       edge of the grid)

                     Communication Pattern

•   Each mote (i, j) can send msgs whose ultimate destination is mote
    (0, 0)
•   The motes need to maintain an incoming spanning tree whose root
    is (0, 0): each mote maintains a pointer to its parent

                                                    (H = 2)

•   When a mote (i, j) has a msg, it forwards the msg to its parent.
    This continues until the msg reaches mote (0, 0)
                     Protocol Message

•   When a mote (i, j) has a parent, then every random period,
    whose average is 20 seconds, mote (i, j) sends the msg:

                connected(i, j)

    Otherwise, mote (i, j) does nothing

•   Every random period, whose average is 20 seconds, mote (0, 0)
    sends the msg:

                connected(0, 0)

                   Maintaining a Parent

•   Initially, no mote has a parent

•   When a mote (i, j) receives a connected(x, y) msg, where (x, y)
    is a potential parent of (i, j), (i, j) makes (x, y) its (new) parent

•   Thus, the parent of a mote is changed, in a round robin fashion,
    among the active potential parents of that mote – load
    balancing and fast fault recovery

                           Losing the Parent

•   If a mote (i, j) does not receive any connected(x, y) msg from any
    of its potential parents for 120 seconds, then (i, j) loses its parent

•   If a mote (i, j) has no parent and receives a connected(x, y) msg,
    where (x, y) is not a potential parent of (i, j), then (i, j) makes
    (x, y) its “foster parent” but (i, j) will not send connected(i, j) msgs
    as long as (i, j) has no parent

                   Using the Routing Protocol

•   When a mote (i, j) has a data msg to forward, it checks whether
    (i, j) has a parent or a foster parent

     if (i, j) has a parent or a foster parent (x, y), (i, j) sends a
       data(x, y) msg, intended for (x, y)

     otherwise, (i, j) discards the data msg

•   A mote (i, j) has a data msg to forward iff either the mote itself has
    generated the msg or it has received the data(i, j) msg

        Using the Routing Protocol by the Root

•   When mote (0, 0), the base station, receives any data(x, y), it
    forwards the msg text to its resident application (the base
    station snooping)

                   Grid Routing in Exscal

•   Each mote is assigned three potential parents for a base station,
    based on a location of a mote in a logical grid
       A mote reads potential parent information from internal flash.
       “Potential Parents” session will cover how to compute potential
        parents for each mote in the demo topology

•   Provide primary and secondary base stations for each mote -
    overcome a base station failure
       A sensor can be connected to the secondary base station, only when
        its primary base station fails

•   Connected message format
         connected(myID, currentBaseStationID)

                     Data-centric routing

•   Sensor networks can be considered as a virtual database

•   Implement query-processing operators in the sensor network

•   Queries are flooded through the network (or sent to a
    representative set of nodes)

•   In response, nodes generate tuples and send matching tuples
    towards the origin of the query

•   Intermediate nodes may merge responses or aggregate

                             Directed Diffusion

•   Protocol initiated by destination (through query)

•   Data has attributes ; sink broadcasts interests

•   Nodes diffuse the interest towards producers via a sequence of local

•   Nodes receiving the broadcast set up a gradient (leading towards
    the sink)

•   Intermediate nodes opportunistically fuse interests, aggregate,
    correlate or cache data

•   Reinforcement and negative reinforcement used to converge to
    efficient distribution
                   Illustrating Directed Diffusion

Setting up gradients
                                                      Sending data

       Source                          Source

                             Sink                            Sink

                                                     stable path

                                Sink                        Sink

     from node failure
                       Data Naming

•   Expressing an Interest
     Using attribute-value pairs
     E.g.,   Type = Wheeled vehicle     // detect vehicle location
              Interval = 20 ms           // send events every 20ms
              Duration = 10 s            // Send for next 10 s
              Field = [x1, y1, x2, y2]   // from sensors in this area

                   Gradient Set Up

•   Inquirer (sink) broadcasts exploratory interest, i1
     Intended to discover routes between source and sink

•   Neighbors update interest-cache and forwards i1

•   Gradient for i1 set up to upstream neighbor
     No source routes
     Gradient – a weighted reverse link
     Low gradient  Few packets per unit time needed

                  Exploratory Gradient

                        Exploratory Request


                  Low                    Low

Bidirectional gradients established on all links through flooding

               Event-data propagation

•   Event e1 occurs, matches i1 in sensor cache
     e1 identified based on waveform pattern matching

•   Interest reply diffused down gradient (unicast)
     Diffusion initially exploratory (low packet-rate)

•   Cache filters suppress previously seen data
     Problem of bidirectional gradient avoided


                                 Reinforced gradient
                                         Reinforced gradient
         A sensor field                      Sink

•   From exploratory gradients, reinforce optimal path for
    high-rate data download  Unicast

     By requesting higher-rate-i1 on the optimal path

     Exploratory gradients still exist – useful for faults

               Path Failure / Recovery

•   Link failure detected by reduced rate, data loss
     Choose next best link (i.e., compare links based on
       infrequent exploratory downloads)
•   Negatively reinforce lossy link
     Either send i1 with base (exploratory) data rate
     Or, allow neighbor’s cache to expire over time

                                                 Link A-M lossy
      Event          D                           A reinforces B
                             M                   B reinforces C …
                                      A          D need not
                     C                           A (–) reinforces M
                              B           Sink
                                                 M (–) reinforces D

                    Loop Elimination

                P                     Q

         D                M                 A

•   M gets same data from both D and P, but P always
    delivers late due to looping
     M negatively-reinforces (nr) P, P nr Q, Q nr M
     Loop {M  Q  P} eliminated
•   Conservative nr useful for fault resilience

                 Local Behavior Choices

1. For propagating interests      3. For data transmission
   In our example, flood             Different local rules can result in
   More sophisticated behaviors         single path delivery, striped
     possible: e.g. based on            multi-path delivery, single
     cached information, GPS            source to multiple sinks …

2. For setting up gradients       4. For reinforcement
   Highest gradient towards
                                     reinforce one path, or part
     neighbor from whom we
                                        thereof, based on observed
     first heard interest
                                        losses, delay variances etc.
   Others possible: towards
     neighbor with highest           other variants: inhibit certain
     energy                             paths because resource levels
                                        are low

                        Simulation studies


•   Compare diffusion to a)
    flooding, and b) centrally                   DIFFUSION
    computed tree (“ideal”)

•   Key metrics:
       total energy consumed per
        packet delivered (indication
        of network life time)
       average pkt delay
                                               CENTRALIZED DIFFUSION


                        Rumor Routing

•   Designed for query/event ratios between query and event

•   Motivation
       Sometimes a non-optimal route is satisfactory

•   Advantages
       Tunable best effort delivery
       Tunable for a range of query/event ratios

•   Disadvantages
       Optimal parameters depend heavily on topology (but can be
        adaptively tuned)
       Does not guarantee delivery
Rumor Routing

                    Basis for Algorithm

•   Observation: Two lines in a
    bounded rectangle have a
    69% chance of intersecting

•   Create a set of straight line
    gradients from event, then
    send query along a random
    straight line from source

•   Thought: Can this bound be
    proved for a random walk .
    What is this bound if the line
    is not really straight?
                      Creating Paths

•   Nodes that observe an event
    send out agents which leave
    routing info to the event as
    state in nodes

•   Agents attempt to travel in a
    straight line

•   If an agent crosses a path to
    another event, it begins to
    build the path to both

•   Agent also optimizes paths if
    they find shorter ones
                         Algorithm Basics

•   All nodes maintain a neighbor list

•   Nodes also maintain a event table
     When it observes an event, the event is added with distance 0

•   Agents
     Packets that carry local event info across the network
     Aggregate events as they go


                        Agent Path

•   Agent tries to travel in a “somewhat” straight path
     Maintains a list of recently seen nodes

     When it arrives at a node adds the node’s neighbors to the
     For the next tries to find a node not in the recently seen list

     Avoids loops

     -important to find a path regardless of “quality”

                     Following Paths

•   A query originates from source, and is forwarded along
    until it reaches it’s TTL

•   Forwarding Rules:
     If a node has seen the query before, it is sent to a random
     If a node has a route to the event, forward to neighbor along
       the route
     Otherwise, forward to random neighbor using straightening

                   Fault Tolerance

•   After agents propagated paths to events, some nodes
    were disabled

•   Delivery probability degraded linearly up to 20% node
    failure, then dropped sharply

•   Both random and clustered failure were simulated with
    similar results

               Reliable Data Transport

•   Transport layer design is difficult because of application-
    specific nature of sensor networks

•   Networking layers tend to become fused (particularly
    transport and application)

•   Goal: design customizable transport layer

•   Provide the primitives for reliable transport


Shared By: