Routing in Sensor Networks

1
Routing in Sensor Networks
2
Why Routing?

• Routing means carrying data packets from a source
  node to a destination node (usually called sinks
  in sensor networks terminology)
• Such routing paths helps to create
  energy-efficient data dissemination paths between
  sources (sensors) and sinks (global processing
  unit or human interface devices)
• Two kinds of routing single-path and multi-path
• Since energy efficiency is the most essential
  factor, routing algorithms must be robust to
  failures and secure against the compromised and
  malicious nodes to ensure data delivery without
  impacting the lifetime of the network

3
Algorithm for Robust Routing in Volatile
Environments (ARRIVE) Karlof, 2002

• Probabilistic algorithm and makes packet
  forwarding decisions based on localized
  information
• Based on a tree-like topology rooted at the sink
  of the network
• Uses forward approach to contribute to end-to-end
  reliability
• Avoids packet loss by sending multiple packets of
  the single event
• Three sources of packets loss expected
• Isolated link
• Patterned node failures
• Malicious or misbehaving nodes

4
ARRIVE Karlof, 2002

• Terminology
• Event Identified by SourceID, EventID
• Level Each node has unique level indicating
  distance from source to sink (in terms of hops)
• Parents Nodes one level closer to the sink
• Neighbors Nodes on the same level and be able
  hear each other
• Push Push packet to one of the neighbors
• Forward Forward packet to one of the parents
• Forwarding Probability Included in the packet
  header and used to probabilistically select
  whether to push or forward
• Reputation History Each node keeps this
  information for each of its parents and neighbors
  
• Convergence Prevents multiple packets of the
  same event being sent to same source of failure

5
ARRIVE Karlof, 2002

• Achieves diversity in paths in two ways
• Upon receiving a packet, the next hop is selected
  probabilistically based on link reliability and
  node reputation
• When more than two or more packets of the same
  event are processed, these packets are ensured to
  follow different outgoing links
• Takes advantage of passive participation and
  needs to be used cautiously
• Each nodes keeps the following information
• Level
• Neighbors list
• Parents list
• Reputation history of neighbors and parents
• Convergence history of specific events

6
ARRIVE Karlof, 2002

• Assumptions
• The networks is assumed to be dense enough that
  sufficient multiplicity of paths between sources
  and sink for algorithm to perform well
• The network is almost considered as a static
  network
• Sensors are considered to have a low per-node
  cost
• Routes used by the packets are unlikely to be
  optimal due to the probabilistic nature of the
  algorithm
• Messages flow from nodes to sink, not the other
  way around
• There is only one sink available
• Performance Metrics
• Event delivery ratio
• Three other metrics measuring the cost of
  deploying ARRIVE

7
ARRIVE Karlof, 2002

• Algorithm Description
• Bread first search rooted at sink is used to
  initialize level, parents, neighbors state
  information at each node
• When a nodes hears a packets, it checks to see if
  the packet is addressed for it
• If so, threshold processing takes place. Nodes
  are filtered by their reputation and convergence
  history of the neighbors and parents
• A decision needs to be made to either to choose
  to forward the packet to a parent or push it to
  one of its neighbors with the probability value
  found in the packet header. This is randomly
  determined by the forwarding probability function
  Pr(f).
• Each node is weighed by their reputation. The
  destination is randomly selected from the rest of
  the nodes (since bad reputation nodes are
  eliminated)
• If the the packet is forwarded to one of the
  parents, Pr(f) is not changed however, its value
  is increased

8
Algorithm for Robust Routing in Volatile
Environments (ARRIVE) Karlof, 2002
Figure 0 Overview of ARRIVE
Adapted from Karlof 2002
9
ARRIVE Karlof, 2002

• Advantages
• High end-to-end reliability
• Provides security by eliminating the compromised
  or malicious nodes
• By using multiple paths to forward the same
  event, the probability of the event reaching sink
  is increased and this also ensures to avoid
  packets being forwarded to the same broken link
• Reputation history assists in establishing a
  reliable path

10
ARRIVE Karlof, 2002

• Disadvantages
• Extra power consumption for inactive nodes
  (passive listeners) are not considered
• Better mechanism to take care problems caused by
  passive listening
• There is only one sink, packets are sent from
  sources to sink and not the other way around
• Sensor may have storage problems due to
  maintaining information about its neighbors and
  parents (reputation history)
• Maintaining multiple paths requires more
  resources

11
ARRIVE Karlof, 2002

• Suggestions/Improvements/Future Work
• Beneficial to measure how much passive listening
  affects energy use
• Nodes can be mobile during the simulation instead
  of static
• If there is a significant mobility, state
  information should be updated using a better
  mechanism than flooding
• Explanation of how multiple packets are generated
• How much redundant data is sufficient to optimize
  the network
• Energy-awareness needs to be taken into
  consideration
• Possibly use energy level parameter in the
  decision making
• Include probabilistic analysis of the algorithm
• Study the tradeoff the communication cost of
  ARRIVE vs. its robustness
• Consider load balancing issues such that nodes
  near the sink deplete their resources sooner than
  nodes farther away
• Lack of quantitative analysis of passive
  participation for security reasons
• Experiment with larger number of events

12
Rumor Routing Algorithm for Sensor
Networks Braginsky 2002

• Preliminaries
• Each node has its neighbor list, and an events
  table, with forwarding information to all the
  events it knows.
• After a node witnesses an event, an agent may be
  created, which is a long-lived packet and travels
  around the network. Each agent contains an events
  table, including the routing information for all
  events it knows.
• Since an event happens in a zone, composed of
  several or many nodes, its possible more than
  one agents are created from the zone and moving
  in the network.

13
Rumor Routing Algorithm for Sensor
Networks Braginsky 2002

• Algorithm Description
• When a node observes an event, it will add the
  event to its event table and may also create
  an agent
• An agent will travel in the network and its
  routing table will be updated if there is a
  shorter path to an event within the routing table
  of the node it is visiting
• In a similar way, the routing table of the
  currently visited node will be updated if its
  route to an event is more costly than the agents
• Any node may generate a query for a particular
  event. If it knows the route to the event, it
  will transmit the query. Otherwise, the query
  will be sent in a random direction, and this
  continues until the query reach a node which has
  a route to the event

14
Rumor Routing Algorithm for Sensor
Networks Braginsky 2002
Adapted from Braginsky 2002
15
Rumor Routing Algorithm for Sensor
Networks Braginsky 2002
Figure 3 The agent modifies the exist path
(left) to a more optimal one (right)
Adapted from Braginsky 2002
16
Rumor Routing Algorithm for Sensor
Networks Braginsky 2002

• Advantages
• Deliver queries to events in large networks with
  less average cumulative hops and lower energy
  requirements than simple flooding.
• The algorithm can handle node failure gracefully,
  degrading its delivery rate linearly with the
  number of failed nodes.

17
Rumor Routing Algorithm for Sensor
Networks Braginsky 2002

• Disadvantages
• The path found by the agent sometimes is not the
  shortest and could be unavailable if one of the
  links of the path is broken.
• The agent may carry lots of routing information
  of events even some events have disappeared.
• No hints from the paper about the number of
  agents which should be created from the event
  zone.
• Can the nodes make a query for an event if no
  such event is existing in the network?
• It seems that each event should have an id, but
  no information is provided in the paper.
• Optimization is not considered

18
A Stream Enabled Routing (SER) Protocol for
Sensor Networks Weilian, 2002

• SER protocol allows sources to choose the routes
  based on the instruction (or task) provided by
  the sinks
• An instruction is a predefined as an identifier
  value instead of attributes being assigned to
  task as in the case of directed diffusion
  Intanagonwiwat 2000. Therefore, only
  identifier is sent rather than the attribute
  list, resulting in memory conservation
• It takes into account the available energy of the
  sensor nodes, QoS requirements of the
  instruction, memory limitation of nodes, and the
  localized effect of dense nodes
• Sinks can give new instructions to the sources
  without establishing another path

19
SER Weilian, 2002

• Benefits of dynamic set-up of routes include
• Periodic updates of routes is not needed
• Adapts to failures and cope with topology changes
• No routing table is needed at each sensor node
• New sensor nodes can be added into the route
  selection
• The routes are determined based on the QoS
  requirements of sources
• Four types of communications are allowed
  one-to-one, one-to-many, many-to-one and
  many-to-many
• Four types of messages are used
• Scout message (S-message)
• Information message (I-message)
• Neighbor-neighbor message (N-message)
• Update message (U-message)

20
SER Weilian, 2002

• SER Overview
• SER has seven phases
• Source Discovery
• Route Selection
• Route Establishment
• Route Reconnection
• I-message Transmission
• Instruction Update
• Task Termination
• S-message is used during the source discovery to
  determine sources that will process the
  instruction (or task) specified in the S-message
• Sources decide the type and level of the routes
  needed by the instruction
• There are four types of routes, each with two
  levels (i.e., level-1 and level-2)
• The µ value is the radius of the level-2 routes

21
SER Weilian, 2002

• SER Overview
• Stream Identified by both the type and the level
  of route
• Each level-2 stream includes level-1 stream
• In level-2, the size of the radius µ of the
  stream can be determined based on QoS specified
  in the instruction
• Combination of types and levels creates different
  kinds of QoS for a stream
• After streams are chosen, the source sends
  N-message to establish the streams back to the
  sink
• The repairs of streams are accomplished via
  N-message and S-message
• Once the streams are accomplished, data travels
  from sources to the sink through either level-1
  or level-2 stream with I-message
• The sink can update the instruction (or task) at
  the sources through either level-1 or level-2
  stream using U-message
• Both sources and sink can terminate the streams
  using U-message

22
SER Weilian, 2002

• 1. Source Discovery
• A sink broadcasts an S-message to find routes
  from sink to source
• S-message contains the following fields

TID Task ID NAP Network Access Point (indicates
where the instruction is originated represents a
unique sink) LID Local ID (each node has a local
ID that is randomly selected from a set) NH
Number of hops from the sink AE Average energy
of a route
23
SER Weilian, 2002

• 1. Source Discovery
• TID has the following fields

LI Length indicator MT Message type (MT0
(S-message) MT1 (I-message) MT2 (U-message)
MT3 (N-message) INS Instruction (Maps
a numeric value to a specific instruction) TLOC
Targeted location

• When a sensor node receives an S-message, it
  determines if the instruction (INS) is intended
  for the node
• If the INS in the S-message is not intended for
  the node, the node stores the fields of S-message
  in a connection-tree (C-tree)

24
SER Weilian, 2002

• 1. Source Discovery
• C-tree is a logical tree which represents
  possible connections through the node. C-tree
  maintains the nodes neighbors that can
  participate in a routing back to sink
• DSP Downlink Sensor Problem (indicates if the
  downlink sensor node is having problem in routing
  I-message)
• NS Node Selected (indicates if the node is
  selected for routing)
• DLID LID of downlink sensor node (store the LID
  value of neighbor node which will route the
  I-message back to sink)
• ULID LID of uplink sensor node (U-message can be
  forwarded to sources from the sink or route
  reconnection is possible using N-message)
• A sensor node in an established route knows the
  LID values of both uplink and downlink nodes
• Initially, DLID and ULID values are not set and
  DSP and NS values are set to OFF
• Updated values of AE, NH and LID fields of
  S-message is broadcast to neighbors

25
SER Weilian, 2002

• 1. Source Discovery
• If sensor node receive the same S-message from
  its neighbors, it dismisses it
• The sources store S-message in a task-tree
  (T-tree)
• T-tree has XDLID values since source can select
  up to XLIDs to route I-message back to the sink
  based on QoS requirement
• The max value of x is the number of neighbor
  nodes
• Each DLID value corresponds to a DSP indicator
• In the T-tree, the leaf nodes has no ULID and NS
  indicator since sources are destination of
  S-message
• A source can receive xS-message since it has x
  neighbors
• Route associated with the first received
  S-message is considered shortest route
• Sources selects neighbor node to send I-message
  back to sink based on the QoS requirement of INS

26
SER Weilian, 2002

• 2. Route Selection
• Once the sources receive the S-message, they
  determine the QoS requirement of task in the
  S-message
• There are four types of streams for communication
  between sources and sinks and each stream can
  either be level-1 or level-2
• Type 1 Time Critical But Not Data Critical
• Type 2 Data Critical But Not Time Critical
• Type 3 Not Time and Data Critical
• Type 1 Data and Time Critical
• After the sources select the neighbor nodes, the
  sources broadcast an N-message to their neighbors
  indicating the level and size of the stream
• N-message contains the following fields

SLID Selected ID MES Message
27
SER Weilian, 2002

• 2. Route Selection
• If stream is level-1, µ 0 (width of the stream)
• At level-1, messages are routed back to the sink
  via hop-by-hop communication. Message are sent to
  only one node
• Level-2 stream contains level-1 stream which
  serves as a backbone in setting up level-2 stream
• The value of µ is the number of hops away from
  the nodes in the level-1 stream
• Messages can flow downhill to the sink or uphill
  to the sources by flooding through only the nodes
  that are part of the stream
• I-message flows downhill from sources to sink by
  using NH value stored in each node in the C-tree
• The nodes near to the sources have higher NH
  values
• U-message flows uphill from sink to sources by
  using the negative of NH value
• The nodes near to the sources have higher
  negative NH values

28
SER Weilian, 2002

• 3. Route Establishment
• N-message is used by a sensor node to inform
  neighbors about its local information
• The source sends an N-message to establish stream
  back to the sink
• Sensor nodes that are not part of a stream delete
  all data associated with N-message from C-tree
• If intermediate nodes between the sources and
  sinks have not received an N-message in response
  to S-message in a set time interval, the sensor
  node deletes the C-tree branch that is associated
  to S-message
• After the N-message arrives to the sink, the
  minimum delay or maximum average energy stream is
  established
• Sources can start sending I-messages to the sink
• I-message contains the following fields

FI Flow Indicator (message going
uphill or downhill) CNH Current number of
hops Payload Description
29
SER Weilian, 2002

• 4. I-message Transmission
• The neighbor nodes can determine if they need to
  route the I-message by the TID since each
  neighbor nodes maintain a C-tree
• When a source broadcasts I-message, it sends CNH
  field with the value from T-tree
• Intermediate nodes between sources and sink use
  C-tree
• FI and CNH fields are only used when the stream
  is level-2
• Each node only rebroadcasts once to avoid a node
  from broadcasting the same message over again
• After an I-message is received, the sensor nodes
  turns OFF the receiver for some amount of time if
  the sleep mode operation is ON such that the node
  can avoid listening neighbors broadcasting the
  same I-message
• C-tree indicates which instructions the sensor
  nodes need to route

30
SER Weilian, 2002

• 5. Route Reconnection
• If a sensor node is low on energy or there is too
  much noise around when transmitting at level-1,
  it can broadcast an N-message by setting up
  reconnect message indicator
• Once the neighbors receive N-message, they check
  their C-tree to decide if there are possible
  alternate routes
• N-message will be broadcasted until the alternate
  route is found
• Sudden Death of Route
• If the stream suddenly terminates, sink cannot
  get the I-messages
• The sink sends out a new S-message with higher
  QoS requirement version of the same instruction
  (higher QoS INS value)
• New streams can be found to avoid broken paths
• Multiple streams of level-2 can be setup between
  source and sink to improve robustness of
  I-message routing

31
SER Weilian, 2002

• 6. Instruction Update
• U-message allows sink to update its instruction
  to the sources
• The U-message from the sink to the sources flow
  uphill while it flows downhill from sources to
  the the sink when streams are level-2
• U-message contains the following fields

NINS New INS

• 7. Task Termination
• A task at the sources are terminated in two ways
• Sources have finished the task associated with
  the instruction given by sink
• U-message with the task completed instruction
  indicator is broadcast by sources
• Sink decides to terminate the instruction
• U-message with the task termination instruction
  indicator is set by the sink
• The streams are torn down by removing C-tree
  braches at the intermediate nodes and T-tree at
  the sources

32
SER Weilian, 2002

• Advantages
• QoS requirements of the instruction is considered
• Average energy of the routes are taken into
  consideration in routing
• Robustness is achieved through selection of
  level-1 and level-2 streams
• After the route is established, sink can give new
  instructions to the sources without setting up
  another route
• Four types of communication is supported
  one-to-one, one-to-many, many-to-one, and
  many-to-many

• Disadvantages
• Storage and computation cost at the nodes
• Loops can form in level-2 streams
• How to set the value of µ

33
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Introduction
• This paper proposes a multipath routing to
  increase resilience to node failure
• Even though multipath routing techniques have
  been discussed in the literature, application of
  multipath routing to sensor networks that permit
  data-centric routing with localized path setup
  has not been studied too much
• Two different approaches to construct multipaths
  between two nodes have been considered
• Classical node-joint multipath adopted by prior
  work, where the alternate paths do not intersect
  the original path or each other. The disjoint
  property ensures that when k alternate paths are
  constructed, no set of k node failures can
  eliminate all the paths
• Another approach is building many braided paths.
  With this approach, there are usually no
  completely disjoint paths, instead, there may be
  many partially disjoint alternate paths

34
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Two issues are addressed
• The paper defines localized algorithms for the
  construction of alternate paths. In order to
  achieve robustness and energy-efficiency, sensor
  network data dissemination mechanisms use
  localized decisions for path setup and recovery
  from failure
• The relative performance of disjoint and braided
  multipaths are evaluated using two metrics
  resilience and maintenance overhead. The
  resilience of a scheme measures the possibility
  of when the shortest path has fails, an alternate
  path is available between source and sink. The
  maintenance overhead of a scheme is a measure of
  the energy required to maintain these alternate
  paths using period keep-alives. There is a
  tradeoff between these two metrics becoming
  more resilient generally consumes more energy

35
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Disjoint and Braided Multipaths
• A. Sensor Networks
• Due to compact form factor of the sensor nodes
  and their low cost, packed cluster of sensor
  nodes can be densely deployed, near the phenomena
  to be sensed.
• The advantage would be to still obtain high SNR
  (signal generated by any physical phenomena
  attenuates with distance) with cheap sensors.
• Additionally, an individual sensor may not have
  to frequently perform multi-target resolution
  (i.e., distinguish between different targets such
  as individuals and vehicles)
• Such multi-target resolution can involve complex
  deconvolution algorithms requiring non-trivial
  processing capability Pottie 2000

36
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Disjoint and Braided Multipaths
• A. Sensor Networks
• Three criteria drive the design of large-scale
  sensor networks
• Scalability
• These networks may involve thousands of nodes
• Energy-efficiency
• Wireless communication can have much higher
  energy cost than computation Pottie 2000
• Robustness
• To environmental effects and link failures

37
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Disjoint and Braided Multipaths
• B. The Problem
• Previous work, directed diffusion Intanagonwiwat
  2000 constructs dissemination paths from
  multiple sinks to multiple sources. Here,
  multipath dissemination from a single source to a
  single sink is considered
• The solution which constructs energy-efficient
  paths on-demand works as follows
• A source of sensory data periodically broadcasts
  (at a low rate) about events describing the
  detection of the external phenomena that is being
  sensed
• Upon receiving multiple copies of these events,
  the sink sends a reinforcement message to one of
  its neighbors stating that it prefers to receive
  notification of detection of events at a higher
  frequency from this neighbor (Figure 4(a))

38
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Disjoint and Braided Multipaths
• B. The Problem
• This enforcement message is propagated to the
  source via hop-by-hop. Each node makes and
  independent, local decision about which of its
  neighbors it chooses to forward the reinforcement
  (Figure 4(b)). As it propagates, the
  reinforcement message implicitly sets up a data
  path in the reverse direction. At each node, the
  reinforcement message sets up state that forwards
  matching data towards the previous hop
• When a node in the reinforcement path fails
  (Figure 4(c)), the sink detects an absence of
  detection events and reinitiates reinforcement.
  The sink must periodically send reinforcement
  messages

39
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002
Disjoint and Braided Multipaths B. The Problem
Figure 4 A Simplified schematic for Directed
Diffusion
Adapted from Ganesan 2002
40
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Disjoint and Braided Multipaths
• B. The Problem
• The main problems considered in this paper are
• For energy-efficiency reasons, paths are
  constructed on-demand rather than proactively
• For robustness reasons, a periodic low-rate
  flooding scheme notifies the sink and other nodes
  of available alternate paths. The periodicity of
  flooding determines the temporal accuracy of
  alternate path characteristics
• The major drawback of this scheme, from
  energy-efficiency point of view, is the periodic
  flooding of low-rate events

41
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Disjoint and Braided Multipaths
• B. The Problem
• This paper considers mechanisms that allow
  restoration of paths from source to sink without
  the periodic flooding
• These mechanisms are based on some observations
  while setting up the path between a source and
  sink, it may be possible to set up and maintain
  alternate paths in advance (with some extra
  energy) to minimize the possibility of having to
  invoke data flooding for alternate path discovery

42
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Multipath Routing
• The multipath routing allows the establishment of
  multiple paths between source and destination
• The reasons for classical multipath routing are
• Load balancing -- traffic between a
  source-destination pair is divided across
  multiple (partially or completely) disjoint paths
• Increase the possibility of reliable data
  delivery
• In these approaches, the multiple copies of data
  are sent along different paths
• Both of these reasons of classical multipath are
  still applicable in wireless sensor networks
• load balancing can distribute energy utilization
  across nodes in the network, possibly resulting
  in longer lifetimes
• duplicate data delivery along multiple paths
  allows more accurate tracking in surveillance
  application with an additional energy usage

43
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Multipath Routing
• The focus of this paper is to find alternate
  paths between source and sink with the help of
  multipath routing
• The rationale of using multipath can be described
  as
• The goal of localized reinforcement based
  mechanisms is to empirically (i.e., by measuring
  short-time traffic characteristics) establish
  best path (i.e., low latency, low packet loss,
  etc).
• Primary path is used to represent this best path.
  From the applications perspective, the goal is
  to deliver data along this primary path.
• To recover from failure of this primary path, a
  small number of alternative paths are constructed
  and maintained without using period flooding in
  case the primary path fails
• When the primary path is set up, the network also
  sets up the multipaths along which data is sent
  at a low-rate

44
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Multipath Routing
• This low-rate data represent energy expended for
  maintaining multipaths and the term maintenance
  overhead denotes this energy. The low-rate data
  represent keep-alives on the alternate paths
• When failure is detected on the primary path,
  nodes can rapidly reinforce an alternate path
  without initiating route discovery thru flooding
• In case of the failures on the primary path and
  on all the alternate paths simultaneously, the
  source or sink initiates network-wide flooding of
  data to re-establish the multipath
• This paper considers two possible multipath
  routing
• Disjoint
• Braided

45
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Multipath Routing
• This paper addresses two issues
• It is not obvious what localized mechanisms may
  be used to construct disjoint and braided paths
• Disjoint and braided multipath trade energy for
  resilience differently
• Disjoint Multipaths
• Construct a small number of alternate paths that
  are node-disjoint with the primary path and with
  each other
• These alternate paths are independent of the
  failures on the primary path however, they can
  be less desirable (i.e., high latency, low
  throughput, etc)

46
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Disjoint Multipaths
• The definition of k node-disjoint multipath is
• Construct the primary path P between source and
  sink
• The first alternate disjoint path P1 is the best
  path node-joint with P
• The second alternate disjoint path P2 is the best
  path that is a node-disjoint with P and P1, and
  so on
• This definition assumes global knowledge of
  topology and network characteristics there this
  is called idealized algorithm for constructing
  disjoint paths and the corresponding multipath is
  called idealized k-disjoint multipath
• The question is how can we obtain node disjoint
  multipaths using local information only without
  global topology information?

47
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Disjoint Multipaths
• Assume that some low-rate sample have initially
  been flooded throughout the network (Figure 5(a))
• The sink has some empirical information about
  which of its neighbors can provide it with the
  highest quality of data (lowest loss or lowest
  latency)

(a) Low-rate samples
Figure 5 Construction of Localized Disjoint Paths
Adapted from Ganesan 2002
48
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Disjoint Multipaths
• To this most preferred neighbor, it sends out a
  primary-path reinforcement (Figure 5(b))
• Similar to basic directed diffusion scheme, that
  neighbor locally determines its most preferred
  neighbor in the direction of the source and so on

(b) Primary-path P
Figure 5 Construction of Localized Disjoint Paths
Adapted from Ganesan 2002
49
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Disjoint Multipaths
• After it starts receiving data along the primary
  path or after sending the primary-path
  reinforcement, the sink sends an alternate path
  reinforcement to its next most preferred
  neighbor. This neighbor A propagates the
  alternate path reinforcement to its most
  preferred neighbor B in the direction of the
  source.

(c) Alternate-path Negative Reinforcement
Figure 5 Construction of Localized Disjoint Paths
Adapted from Ganesan 2002
50
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Disjoint Multipaths
• If B can determine from local state that it is
  already on the primary path between source and
  sink, it sends a negative reinforcement to A
  (Figure 5(c))
• A then selects its next best preferred neighbor
  otherwise, B propagates the alternate path
  reinforcement to its most preferred neighbor and
  so on (Figure 5(d))

(d) Alternate-path P1
Figure 5 Construction of Localized Disjoint Paths
Adapted from Ganesan 2002
51
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Disjoint Multipaths
• Nodes other than sink do not originate alternate
  path reinforcements
• This mechanism described can be extended to
  construct k disjoint multipaths, by sending out k
  alternate path reinforcement from the sink, each
  separated from the next by a small delay
• In this case, each node is constrained to receive
  only one reinforcement of either type-primary
  path, or alternate path
• If it receives more than one reinforcement, the
  node negatively reinforces these, ensuring
  disjointed-ness
• This is called localized disjoint multipaths
• In the idealized algorithm, the first alternate
  path is the primary path which is node-disjoint
  with the primary path. Since localized
  construction has only local knowledge of
  alternative paths, its search procedure may
  discover longer alternate paths (Figure 5(e))

52
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Disjoint Multipaths
• In figure 5(e), the sink reinforces A in
  preference to X, although X is on a shorter
  alternate path. Since the sink hears events
  earlier from A, but does not consider that these
  are forwarded to A by B which is on the primary
  path
• The idealized algorithm would choose Q as the
  alternate disjoint path

(e) Caveat
Figure 5 Construction of Localized Disjoint Paths
Adapted from Ganesan 2002
53
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Braided Multipaths
• Alternate paths in a braid are partially disjoint
  from the primary path, not completely
  node-disjoint
• The definition of braided multipath can be seen
  in Figure 6
• For each node on the primary path, find the best
  path from source to sink that does not contain
  that node and this path may not be completely
  node-disjoint with the primary path

Figure 6 Idealized Braid
Adapted from Ganesan 2002
54
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Braided Multipaths
• The resulting set of paths along with the primary
  path is called the idealized braided multipath
• The links representing a braid either lie on the
  primary path or close to it
• The localized technique for constructing braids
  is similar to idealized algorithm for disjoint
  multipath however, the local rules are a bit
  different sink also sends an alternate path
  reinforcement to its next preferred neighbor
  (i.e., node B) apart from sending the primary
  path reinforcement message to its most preferred
  neighbor (i.e., A)
• As before, node A propagates the primary path
  reinforcement to its most preferred neighbor and
  so on
• In addition, A (and recursively each other node
  on the primary path) originates an alternate path
  reinforcement to its most preferred neighbor
  each node tries to route around its near neighbor
  on the primary path towards the source

Adapted from Ganesan 2002
55
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Braided Multipaths
• When a node not on the primary path receives an
  alternate path reinforcement, it propagates it
  towards its most preferred neighbor however,
  when a node already on the primary path receives
  an alternate path reinforcement, it does not
  propagate it further
• Figure 7 shows a localized braid constructed
  using the algorithm described

Source
Sink
Figure 7 Localized Braid
Adapted from Ganesan 2002
56
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Braided Multipaths
• In Figure 7, nk1 sends sends an alternate
  reinforcement to route around nk that passes
  through ai and ai-1 before rejoining the primary
  path at nk-2
• An alternate path reinforcement sent out by nk1
  can follow any sequence of nodes can be
  completely disjoint from the rest of the primary
  path and join the primary at nk
• The localized braid is different from the
  idealized braid idealized construction
  algorithm does not prevent an alternate path from
  being chosen which is completely node-disjoint
  with the primary path

Adapted from Ganesan 2002
57
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Qualitative Comparison
• The energy cost of alternate disjoint paths
  depends on the network density, i.e., at low
  network densities, alternate disjoint paths are
  longer and more costly than the primary path. In
  addition for a larger k, the energy used to
  maintain k-disjoint paths is high
• On the other hand, at high densities, possibility
  of finding node-disjoint alternate paths of
  shorter length increases, thus reducing energy
  used in maintenance
• In the idealized braid, an alternate path routes
  around a single primary path node and the energy
  consumption of an alternate path in the braid is
  similar to the primary path and not much
  dependent on the density
• At lower densities, the difference in energy
  consumption for multipath maintenance between
  disjoint multipath and braided multipath is high
  the difference decreases with increasing density

Adapted from Ganesan 2002
58
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Qualitative Comparison
• The paper considers two kinds of failure
  isolated and patterned
• Isolated failures model independent node failure
  while patterned failures model geographically
  correlated failure
• In disjoint paths, any number of nodes can fail
  on the primary path without impacting the
  alternate path however, the failure of a single
  node on each alternate path means the failure of
  the multipath
• In braided multipaths, the various alternate
  paths are not independent, and a combination of
  failures on the primary path could impact all
  alternate paths however, the number of distinct
  alternate paths through a braid is higher than
  the number of nodes in its primary path
• Patterned failures affect disjoint and braided
  paths differently

Adapted from Ganesan 2002
59
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Qualitative Comparison
• A failure pattern that affects the primary path
  is likely to affect alternate paths near primary
  path more than paths that are distant
• Since braiding encourages geographically closer
  alternate paths, disjoint multipaths are likely
  to be more resilient to pattern failures than
  braided multipaths
• The following questions have been explored using
  simulation
• How much additional energy must one expend in in
  order to increase resilience by a fixed amount?
• How does the energy/resilience tradeoff vary with
  density or with the extent and frequency of
  patterned failures?
• How closely do the localized schemes approximate
  their idealized counterparts?

Adapted from Ganesan 2002
60
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Conclusions
• This paper the use of multipath routing for
  energy-efficient recovery from node failures in
  wireless sensor networks
• When a small number of multipaths are available,
  failures on the primary path can usually be
  recovered without invoking network-wide flooding
  for path discovery since flooding can reduce
  network lifetimes
• Two kinds of multipath designs are proposed and
  evaluated the classical node-disjoint multipath,
  and a novel braided multipath that consists of
  partially disjoint alternate paths
• The energy/resilience tradeoffs of these
  mechanisms both for independent and
  geographically-correlated failures are explored

Adapted from Ganesan 2002
61
Highly-Resilient, Energy-Efficient Multipath
Routing in Wireless Sensor Networks Ganesan,
2002

• Conclusions
• The findings of the study are as follows
• For a disjoint multipath configuration whose
  patterned failure resilience is comparable to
  that of braided multipaths, the braided
  multipaths have about 50 higher resilience to
  isolated failures and a third of the overhead for
  alternate path maintenance
• It is harder to design localized
  energy-efficient mechanisms for constructing
  disjoint alternate paths, since the localized
  algorithms do not have the information to find
  low latency disjoint paths
• Finally, increasing the number of disjoint paths
  also increase the resilience of disjoint
  multipaths but with a higher energy cost

Adapted from Ganesan 2002
62
Meshed multipath routing (M-MPR) with selective
forwarding an efficient strategy in wireless
sensor networks De, 2003

• Introduction
• This paper presents a meshed multipath routing
  (M-MPR) with selective forwarding (SF) of packets
  and end-to-end forward error correction (FEC)
  coding in wireless sensor networks
• Two ways of effecting disjoint multipath routing
  (MPR) are as follows
• Disjoint (or split) MPR (D-MPR) with selective
  forwarding (SF)
• Each packet is sent along different disjoint
  routes and the decision of path selection is made
  by the source on packet-by-packet basis
• D-MPR with packet replication (PR) (or limited
  flooding)
• Multiple copies of a data packet are transmitted
  simultaneously along multiple disjoint routes
  from a source to a destination

63
M-MPR De, 2003

• Introduction
• When A forward error correction (FEC) is used,
  D-MPR with PR approach requires minimum code
  length (least error correction overhead)
  however, it may have inefficient resource
  utilization
• The D-MPR with SF approach relies on the end node
  (i.e., source node) to choose routes for each
  packet
• The end node may not have up-to-date route
  information to make routing decisions and also it
  is not realistic to exchange whole network
  information among all the nodes
• The scheme proposed in this paper allows some
  intermediate nodes to have more than one
  forwarding direction to a given destination
• Selective forwarding of packets (SF) is proposed
  where the forwarding decision is made
  dynamically, hop-by-hop, based on the conditions
  of downstream forwarding nodes
• End-to-end FEC is used to avoid
  acknowledgement-based retransmission

64
M-MPR De, 2003

• Meshed Multipath Routing
• A. Multipath searching
• In remote sensing applications, the sensor nodes
  need to communicate with a common monitoring or
  control center can be called a clusterhead or a
  controller node
• In such applications, the sensors are mostly
  stationary and their location information can
  transmitted during the initial deployment phase
  by standard trilateration approach using other
  GPS-capable nodes or by the directional beaconing
  approach
• The controller node is location-aware and
  distribute its location information to other
  nodes via broadcast or beaconing
• With the above considerations in mind, a meshed
  multipath is set up in three steps (i) acquiring
  neighborhood information, (ii) route discovery,
  and (iii) route reply

65
M-MPR De, 2003

• Meshed Multipath Routing
• A. Multipath searching
• Acquiring Neighborhood Information
• Each active node broadcasts its ID, residual
  battery power, and location information to local
  neighbors
• For each active neighbor i, a node maintains the
  following information in its database IDi,
  locationi, residual_poweri
• Since the sensor nodes are considered stationary,
  period update on neighborhood status is not
  needed unless the node is entering to sleep mode
  or has just woken up
• In this case, the nodes status is locally
  broadcast based on which of the neighborhood
  tables of nearby nodes are updated

66
M-MPR De, 2003

• Meshed Multipath Routing
• A. Multipath searching
• Route Discovery
• Each node attempts to form a meshed multipath
  based on the neighborhood database and location
  information of the controller node
• Up to now, the intermediate node is allowed to
  accept multiple discovery packets
• During source-to-destination route discovery
  process, at most two copies of a discovery packet
  are accepted by an intermediate node and the
  first arrived packet is forwarded to maximum two
  downstream neighbors nodes to ensure the
  reduction of the receiver complexity and power
  consumption of a node (Figure 8(a))
• Maximum two forwarding node is chosen since this
  allows an alternate route with minimum possible
  extra control overhead

67
M-MPR De, 2003

• Meshed Multipath Routing
• A. Multipath searching
• Route Discovery
• The route packet has the following fields
  source_ID, source_location, intermediate_node_ID,
  next_node_ID1, next_node_ID2, destination_ID,
  destination_location, TTL where the IDs of
  forwarding nodes (next_node_IDi, i 1, 2),
  intermediate_node_ID, and TTL values are updated
  at each intermediate stage

Figure 8 (a) a source-to-destination meshed
multipath
68
M-MPR De, 2003

• Meshed Multipath Routing
• A. Multipath searching
• Route Discovery
• Each intermediate node maintains the following
  information in its routing database
  previous_node_IDi,, previous_node_IDn,
  next_node_ID1, next_node_ID2
• Since several nodes is targeting the same
  destination, an intermediate node can have more
  than two previous_node entries in its routing
  table although there will be no more than two
  next_nodes (Figure 8(b))
• The list of previous_node is bounded since the
  number of local neighbors are finite and no entry
  is created in the routing table for discovery
  packets coming from an upstream neighbor which is
  already listed in the list
• If an intermediate node that has already
  forwarded a discovery packet receives another
  discovery packet, it updates the previous_node
  list in its routing table and drops the packet

69
M-MPR De, 2003

• Meshed Multipath Routing
• A. Multipath searching
• Route Discovery

Figure 8 (b) Meshed topology formed by
many-sources-to-a-destination routes
70
M-MPR De, 2003

• Meshed Multipath Routing
• A. Multipath searching
• Route Discovery
• Entry in the routing table at each node is
  maintained as a soft-state that is deleted after
  a time out unless a reply is received from a
  controller node
• Since most sensor applications are data-centric,
  delay differences (jitter) between packet
  arrivals is not a big concern
• No other resource reservation apart from storing
  and maintaining upstream and downstream nodes
  information is made during this phase
• Therefore, the route discovery phase can be
  considered as a topology construction process

71
M-MPR De, 2003

• Meshed Multipath Routing
• A. Multipath searching
• Route Reply
• Route reply message identifies the nodes
  comprised the meshed path
• The controller node, upon receiving the discovery
  packets from a single source, selects the first
  two and sends a route reply following the
  original links by the route discovery packets in
  reverse direction with the following fields
  source_ID, source_location, intermediate_node_ID,
  previous_node_ID1, previous_node_ID2
• Each intermediate node changes the states of its
  corresponding entries from soft to permanent for
  the duration of its active participation, updates
  the fields of the reply packet other than the
  source information and forwards the reply packet
  to its upstream node
• When forwarding the route reply message, the node
  does not require the knowledge of source
  information

72
M-MPR De, 2003

• Meshed Multipath Routing
• A. Multipath searching
• Route Reply
• In case of the discovery packets arriving to
  controller node from several sensor nodes,
  multicast reply is used
• If an intermediate node is out of service or goes
  to sleep mode, the upstream nodes select
  necessary neighbors to sustain connectivity
• Intermittent link breakage will not trigger
  reconfiguration of meshed multipath, instead it
  is handled using selective forwarding (SF)
• In the constructed meshed topology, the number of
  downstream links is no more than two, whereas the
  number of upstream nodes can be more
• As can be observed from Figure 8(b), node n has
  three upstream nodes a, b, and c and two
  downstream nodes x and y

73
M-MPR De, 2003

• Meshed Multipath Routing
• B. Multipath routing
• After the mesh multipath is constructed, the
  packets are forwarded to the destination via the
  meshed multipath using either packet replication
  (PR) or selective forwarding (SF)
• In PR, a source packet is copied along all
  possible paths to its destination
• A node receiving more than one correct copy of
  the packet from upstream nodes selects one
  successful packet to forward to the downstream
  nodes helps to reduce power consumption due to
  the transmission of multiple copies of the same
  packet
• In SF, if multiple downstream nodes are available
  either at the source or at an intermediate node,
  the packet is forwarded along only one downstream
  link based on local conditions if all outgoing
  links are good, random selection is made

74
M-MPR De, 2003

• Meshed Multipath Routing
• B. Multipath routing
• In addition to fault tolerance objective,
  selective forwarding approach along meshed
  multipath offers more efficiency than PR in terms
  of resource utilization and and congestion
  avoidance
• Generally, the signal transmitted by a sensor
  node is broadcast to all its neighbors
• The main difference between PR and SF is that in
  PR, the packet is intended for multiple neighbors
  in which each of them will receive and forward
  the packet whereas in SF, only one receiver will
  receive and forward the packet
• Due to broadcast nature, M-MPR requires less
  transmission energy than D-MPR MPR provides more
  flexibility in selective forwarding decisions
  than D-MPR, resulting in more successful packet
  delivery rate

75
References

• Braginsky 2002 D. Braginsky and D. Estrin,
  Rumor Routing Algorithm for Sensor Networks,
  First Workshop on Sensor Networks and
  Applications (WSNA), September 28, 2002, Atlanta,
  Georgia, pp. 22-31.
• De 2003 S. De, C. Qiao, and H. Wu, Meshed
  multipath routing with selective forwarding an
  efficient strategy in wireless sensor networks,
  Computer Networks, 43, Elsevier Journal, 2003,
  pp. 481-497.
• Ganesan 2002 D. Ganesan,  R. Govindan, S.
  Shenker and D. Estrin, Highly Resilient, Energy
  Efficient Multipath Routing in Wireless Sensor
  Networks,  Mobile Computing and Communications
  Review (MC2R) Vol 1., No. 2. 2002.
• Intanagonwiwat 2000 C. Intanagonwiwat, R.
  Govindan and D. Estrin, Directed Diffusion A
  Scalable and Robust Communication Paradigm for
  Sensor Networks, In Proceedings of the Sixth
  Annual International Conference on Mobile
  Computing and Networks (MobiCOM 2000), August
  2000, Boston, Massachusetts.
• Karlof 2002 C. Karlof, Y. Li and J. Polastre,
  ARRIVE Algorithm for Robust Routing in Volatile
  Environments, UC Berkley, Tech Report, May 2002.
• Pottie 2000 G. Pottie, W. Kaiser, L. Clare,
  and H. Marcy, Wireless Integrated Sensor
  Networks, Communications of the ACM, Vol. 43, No.
  5, pp. 51-58, May 2000
• Weilian 2002 W. Su and I. Akyildiz, A Stream
  Enabled Routing (SER) Protocol for Sensor
  Networks, Med-hoc-Net 2002, Sardegna, Italy,
  September 2002.