Lecture 5: Topology Control

1
Lecture 5 Topology Control

• Anish Arora
• CIS788.11J
• Introduction to Wireless Sensor Networks
• Material uses slides from Paolo Santi and Alberto
  Cerpa

2
Problems affected by link quality

• Topology Control
• Neighborhood Management
• Routing
• Time Synchronization
• Aggregation
• Application Management

3
References

• Topology Control tutorial, Mobihoc04, Paolo
  Santi
• SPAN, Benjie Chen, Kyle Jamieson, Robert
  Morris, Hari Balakrishnan, MIT
• GAF/CEC, Y. Xu, S. Bien, Y. Mori, J . Heidemann
  D. Estrin, USC/ISI UCLA
• ASCENT, Alberto Cerpa and Deborah Estrin, UCLA
• GS3 Scalable Self-configuration and
  Self-healing in Wireless Networks, PODC 2002,
  Hongwei Zhang, Anish Arora
• M. Demirbas, A.Arora, V.Mittal, FLOC A Fast
  Local Clustering Service for Wireless Sensor
  Networks DIWANS 2004

4
Why Control Communications Topology

• High density deployment is common
• Even with minimal sensor coverage, we get a high
  density communication network (radio range gt
  typical sensor range)
• Energy constraints
• When not easily replenished
• Power usage
• Observation radios consume about the same power
  in idle state than Tx and Rx state
• Chicken egg problem to save energy, radios
  must be turned off (not simply reduce packet
  transmissions) but if radios are turned off,
  nodes cannot receive messages

5
Problem Statement(s)

• Find an MCDS, i.e. a minimum subset of nodes that
  is both
• Set cover
• Connected
• Choose a transmit-power level whereby network is
  connected
• per node or same for all nodes
• with per node there is the issue of avoiding
  asymmetric links
• cone-based algorithm
• node u transmits with the minimum power ?u s.t.
  there is at least one neighbor in every cone of
  angle x centered at u
• k-neighbors algorithm
• each node chooses nearest k neighbors for its
  subgraph
• k is chosen s.t. the graph generated is connected
  w.h.p.

6
Problem Statement(s)

• Find a minimum subset of nodes that provides some
  density
• in each geographic region ? connectivity
• well look at the examples of GAF, SPAN, GS3,
  ASCENT
• Given a connected graph G, find a subgraph G
  which can route messages between nodes in
  energy-efficient way
• both unicast and broadcast spanners
• reduces interference as well
• Sub-problems
• Prune asymmetric links
• Tolerate node perturbations
• Load balance

7
Where should TC be positioned in the protocol
stack?

• No clear answer in the literature
• One view
• Routing Layer
• TC Layer
• MAC Layer
• Routing protocol may trigger TC execution (to get
  better routes)
• Routing (structure) involves only active nodes
• MAC protocol may trigger TC execution (if
  neighborhood changes)
• TC controls coarse-grain duty-cycling, MAC
  controls fine-grain
• Mode changes need to be coordinated to avoid
  conflicts

8
Assumptions Radio/MAC

• Circular or Isotropic Models GS3
• Grid-based connectivity GAF, GS3
• Radio/MAC dependencies
• 802.11 Power Saving mode Span
• Promiscuous mode ASCENT, CEC

9
Assumptions Neighbor Information

• Locality
• 1-hop neighbor GS3, ASCENT
• n-hop neighbor (with various n gt 1) GAF, CEC,
  Span
• Dependency on routing GS3, Span
• Measurement-based ASCENT, CEC

10
Properties Reactivity to dynamics load
balancing

• Local re-calculation of state GS3
• Global re-calculation of state Span
• Local recovery GS3, GAF, CEC, ASCENT
• Explicit load balancing mechanisms GS3, Span,
  GAF, CEC

11
SPAN

• Goal preserve fairness and capacity still
  provide energy savings
• SPAN elects coordinators from all nodes to
  create backbone topology
• Other nodes remain in power-saving mode and
  periodically check if they should become
  coordinators
• Tries to minimize of coordinators while
  preserving network capacity
• Depends on an ad-hoc routing protocol to get list
  of neighbors the connectivity matrix between
  them
• Runs above the MAC layer and alongside routing

12
Coordinator Election Announcement

• Rule if 2 neighbors of a non-coordinator node
  cannot reach each other (either directly or via 1
  or 2 coordinators), node becomes coordinator
• Announcement contention is resolved by delaying
  coordinator announcements with a randomized
  backoff delay
• delay ((1 Er/Em) (1 Ci/(Ni pairs))
  R)NiT
• Er/Em energy remaining/max energy
• Ni number of neighbors for node i
• Ci number of connected nodes through node i
• R Random0,1
• T RTT for small packet over wireless link

13
Coordinator Withdrawal

• Each coordinator periodically checks if it should
  withdraw as a coordinator
• A node withdraws as coordinator if each pair of
  its neighbors can reach each other directly of
  via some other coordinators
• To ensure fairness, after a node has been a
  coordinator for some period of time, it withdraws
  if every pair of nodes can reach each other
  through other neighbors (even if they are not
  coordinators)
• After sending a withdraw message, the old
  coordinator remains active for a grace period
  to avoid routing loses until the new coordinator
  is elected

14
Performance Results
15
GAF/CEC Geographical Adaptive Fidelity

• Each node uses location information (provided by
  some orthogonal mechanism) to associate itself to
  a virtual grid
• All nodes in a virtual grid must be able to
  communicate to all nodes in an adjacent grid
• Assumes a deterministic radio range, a global
  coordinate system and global starting point for
  grid layout
• GAF is independent of the underlying ad-hoc
  routing protocol

16
Virtual Grid Size Determination

• r grid size, R deterministic radio range
• r2 (2r)2 ? R2
• r ? R/sqrt(5)

17
Parameters settings

• enat estimated node active time
• enlt estimated node lifetime
• Td,Ta, Ts discovery, active,
• and sleep timers
• Ta enlt/2
• Ts enat/2, enat
• Node ranking
• Active gt discovery (only one node active per
  grid)
• Same state, higher enlt --gt higher rank (longer
  expected time first)
• Node ids to break ties

18
Performance Results
19
CEC

• Cluster-based Energy Conservation
• Nodes are organized into overlapping clusters
• A cluster is defined as a subset of nodes that
  are mutually reachable in at most 2 hops

20
Cluster Formation

• Cluster-head Selection longest lifetime of all
  its neighbors (breaking ties by node id)
• Gateway Node Selection
• primary gateways have higher priority
• gateways with more cluster-head neighbors have
  higher priority
• gateways with longer lifetime have higher
  priority

21
Network Lifetime
22
Challenges for local healing of solid-disc
clustering

• Equi-radius solid-disc clustering with bounded
  overlaps is not achievable in a distributed and
  local manner

23
FLOC protocol

• Solid-disc clustering with bounded overlaps is
  achievable in a distributed and local manner for
  approximately equal radius
• Stretch factor, m2, produces partitioning that
  respects solid-disc
• Each clusterhead has all the nodes within unit
  radius of itself as members, and is allowed to
  have nodes up to m away of itself
• FLOC is locally self-healing, for m2
• Faults and changes are contained within the
  respective cluster or within the immediate
  neighboring clusters

24
FLOC program

• By taking unit distance to be reliable comm.
  radius m be maximum comm. radius, FLOC
• exploits the double-band nature of wireless
  radio-model
• achieves communication- and energy-efficient
  clustering
• FLOC achieves clustering in O(1) time regardless
  of the size of the network
• Time, T, depends only on the density of nodes
  is constant
• Through simulations and implementations, we
  suggest a suitable value for T for achieving fast
  clustering without compromising the quality of
  resulting clusters

25
Model

• Geometric network, e.g., 2-D coordinate plane
• Radio model is double-band
• Reliable communication within unit distance
  in-band
• Unreliable communication within 1 lt d lt m
  out-band
• Nodes have i-band/ o-band estimation capability
• RSSI-based using signal-strength as indicator of
  distance
• Statistics-based using average link quality as an
  indicator
• Fault model
• Fail-stop and crash
• New nodes can join the network

26
Problem statement

• A distributed, local, scalable, and
  self-stabilizing clustering program, FLOC, to
  construct network partitions such that
• a unique node is designated as a leader of each
  cluster
• all nodes in the i-band of each leader belong to
  that cluster
• maximum distance of a node from its leader is m
• each node belongs to a cluster
• no node belongs to multiple clusters

27
Justification for stretch factor gt 2

• For m2 local healing is achieved a new node is
• either subsumed by one of the existing clusters,
• or allowed to form its own cluster without
  disturbing neighboring clusters

)
(
(
(
)
)
)
(
(
)
new cluster
28
Basic FLOC program

• Status variable at each node j
• idle j is not part of any cluster and j is not
  a candidate
• cand j wants to be a clusterhead, j is a
  candidate
• c_head j is a clusterhead, j.cluster_idj
• i_band j is an inner-band member of a
  clusterhead j.cluster_id a clusterhead itself is
  an i_band member
• o_band j is an outer-band member of j.cluster_id
• The effects of the 6 actions on the status
  variable

29
FLOC actions

• idle ? random wait time from 0T
  expired ? become a cand and bcast cand msg
• receiver of cand msg is within in-band ? its
  status is i_band ? receiver sends a conflict msg
  to the cand
• candidate hears a conflict msg ?
  candidate becomes o_band for respective cluster
• candidacy period ? expires ? cand
  becomes c_head, and bcasts c_head message
• idle ? c_head message is heard ?
  become i_band or o_band resp.
• receiver of c_head msg is within in-band ? is
  o_band ? receiver joins cluster as i_band

30
FLOC is fast

• Assumption atomicity condition of candidacy is
  observed by T
• Theorem Regardless of the network size FLOC
  produces the partitioning in T? time
• Proof
• An action is enabled at every node within at most
  T time
• Once an action is enabled at a node, the node is
  assigned a clusterhead within ? time
• Once a node is assigned to a clusterhead, this
  property cannot be violated
• action 6 makes a node change its clusterhead to
  become an i-band member, action 2 does not cause
  clusterhead to change

31
FLOC is locally-healing

• Node failures
• inherently robust to failure of non-clusterhead
  members
• clusterhead failure detected via lease
  mechanism, the orphaned nodes execute clustering
  ---see node additions
• Node additions
• either join existing cluster, or
• form a new cluster without disturbing immediate
  neighboring clusters

32
Extensions to basic FLOC algorithm

• Extended FLOC algorithm ensures that solid-disc
  property is satisfied even when atomicity of
  candidacy is violated occasionally
• Insight Bcast is an atomic operation
• Candidate that bcasts first locks the nodes in
  the vicinity for ? time
• Later candidates become idle again by dropping
  their candidacy when they find some of the nodes
  are locked
• 4 additional actions to implement this idea

33
Simulation for determining T

• Prowler, realistic wireless sensor network
  simulator
• MAC delay 25ms
• Tradeoffs in selection of T
• Short T leads to network contention, and hence,
  message losses
• Tradeoff between faster completion time and
  quality of clustering
• Scalability wrt network size
• T depends only on the node density
• In our experiments, the degree of each node is
  between 4-12
• a constant T is applicable for arbitrarily
  large-scale networks

34
Tradeoff in selection of T
35
Constant T regardless of network size
36
Implementation

• Mica2 mote platform, 5-by-5 grid
• Confirms simulation

37
Sample clustering with FLOC
38
GS3 Scalability via locality

• Locality is hard for some graph problems
• e.g., self-configuration and self-healing of
  routing tree
• An ideal goal for locality
  self-healing should be a function of the size of
  perturbation (in time, space, and energy)
• Locality depends on model

39
System model

• System
• multiple small nodes and one big node, on a
  plane
• node distribution
• density (? Rt s.t. with high probability,
• there are multiple nodes in
  any circular area of radius Rt)
• localization relative location between nodes can
  be estimated
• Perturbations
• dynamic nodes
• joins, leaves (deaths), state corruptions
• mobile nodes

40
Problem Geography-aware self-configuration

• Geographic radius of clusters is crucial
• for communication quality, energy dissipation,
  data aggregations applications
• Problem statement
• Given
• R ideal cell radius (R gt Rt)
• Construct a set of cells , connected via a head
  node in each cell s.t.
• radius of each cell is in R-c , Rc , where c
  f (Rt)
• each node belongs to only one cell
• cells and the connectivity graph over head nodes
  self-heal locally

41
Static networks

• An ideal case
• In reality no node may exist at some geometric
  centers (ILs), but, with high probability there
  are nodes no more than Rt away from any IL

(IL Ideal Location)
42
How to find the set of cell heads

• Bottom-up ?
• hard to guarantee the placement and size of
  clusters
• Top-down w.r.t. big node
• use diffusing computation
• but, accumulation in deviation of head location
  from IL is a problem

i
43
Organizing neighboring clusters heads

• Deviation problem is handled locally
• instead of using real locations, node i uses its
  and its parents ILs
• i calculates the ILs of next band cells in its
  search region lt LD , RD gt
• big node lt0o , 360ogt
• other nodes lt-60o-a , 60oagt , where
  a ? Sin-1(Rt / R)
• for each IL, i ranks nodes within Rt radius of
  the IL (by ltD, Agt), and selects the highest
  ranked node as the corresponding cluster head

44
Summary static networks

• Cell structure is hexagonal
• cell radius
• Time taken to form the structure is ?(Db), where
  Db the maximum distance between the big node
  and the small nodes
• Scalability in self-configuration
• local coordination only with nodes within range
• local knowledge each node maintains info about a
  constant number of nearby nodes

45
Dynamic networks

• Dynamics include
• node join, leave (death), state corruption
• Common vs. rare
• common perturbations node density is preserved
• rare perturbations node density is destroyed
• Scalable self-healing is achieved via locality
  in
• intra-cell healing
• inter-cell healing
• sanity checking of state (invariants)

46
Local intra-cell healing

• Head shift
• upon head leaving (death)
• local in a radius of Rt
• Cell shift
• upon the death of all the nodes in an area of
  radius Rt
• local in a radius of R
• independent but consistent shift at individual
  cells ? sliding of the global head level
  structure

47
Local inter-cell healing sanity checking

• Local inter-cell healing
• upon failure of intra-cell healing at head j,
• first, the parent of j tries to find a new head
  j
• if that fails, the children of j find new parents
• Local sanity checking of state invariants
• upon detecting violation of the hexagonality
  property,
• node corrects itself after checking with its
  neighbors
• when state perturbation includes several nodes,
  the perturbed region corrects itself from the
  outside going in, and all nodes are corrected
  within time proportional to size of perturbed
  region

48
Summary dynamic networks

• Cell radius
• for cells not adjoining any gap
• for cells adjoining a gap
• Head tree is now minimum distance tree rooted at
  the big node
• Stabilization time from perturbed state ?(Dp),
  where Dp diameter of the continuously perturbed
  area

49
Summary dynamic networks (contd.)

• Scalability in self-healing
• local fault-containment and healing
• local knowledge
• Local healing and fault-containment enables
• stable cell structure
• lengthened lifetime ?(nc) , where nc the
  number of nodes in a cell

50
Related work

• Cellular hexagon structure (Mac Donald 79)
• Preconfigured not considering self-healing
• LEACH (Heinzelman et al 00)
• No guarantee about the placement and size of
  clusters
• Perturbations dealt with by globally repeating
  the whole clustering process
• Logical-radius based clustering (in Banerjee 01)
• non-local cluster maintenance, and no
  consideration of state corruption
• only logical radius ? long links and link
  asymmetry are possible
• multiple rounds of diffusion

51
ASCENT

• Adaptive Self-Configuring sEnsor Networks
  Topologies
• Observation different applications may require
  the underlying topology to have different
  characteristics. For example
• Minimal
• Homogeneous with a certain degree of connectivity
• Heterogeneous with different degrees of
  connectivity in different regions
• Examples of these different regions may be
• Along a data flow path
• Avoiding a data flow path
• In the border of an event of interest
• Input application tolerance specified in terms
  of acceptable loss rate at any node

52
Model

• Adapt to empirical measurements of link quality
  each node assesses its connectivity adapts its
  participation in the multi-hop topology based on
  the measured operating region
• Assumptions ASCENT needs to
• turn off the radio (sleep state)
• turn the NIC/MAC in promiscuous mode (passive
  state)
• ASCENT runs on top of MAC and below routing does
  not uses any information gathered by routing

53
ASCENT Basics

• Node state active or passive
• Active nodes are in topology forward data
  packets (using orthogonal routing mechanism that
  runs on topology)
• Passive nodes can sleep or collect network
  measurements
• Each node measures of neighbors and packet loss
  locally
• Each node then decides to join the network
  topology or to adapt (e.g. reducing its duty
  cycle to save energy)

54
State Transitions
NT neighbor threshold LT loss threshold Tx
state timer values (x p passive, s sleep, t
test)
55
Details

• Each node adds a sequence number per packet (for
  loss detection)
• Neighbor estimator based on a neighbor loss
  threshold (NLT) 1 1/N (N number of neighbors
  in the previous cycle)
• Neighbor threshold value (NT) determines the
  average degree of connectivity in the network
• Loss threshold determines the maximum data loss
  application can tolerate
• Relation between Tp/Ts (passive sleep timers)
  determines amount of energy savings and
  convergence time

56
Performance Results
Energy Savings (normalized to the Active case,
all nodes turn on) as a function of density.
ASCENT provides energy savings up to 5.5 times
for high density scenarios
57
ASCENT Energy Savings Analysis
NT neighbor threshold Tp passive state
timer Ts sleep state timer Sleep power radio
off Idle power radio on ? Tp/Ts ? Sleep/Idle
0.004
58
Adaptive Timers

• For any given probability target Pk and given
  the of passive nodes in the area, nodes can
  calculate the optimal relation between the
  passive and sleep timers (?)
• Larger the Pk target, the larger the alpha for
  any given density
• Larger the k, the larger the alpha (although it
  grows slowly)

59
Adaptive vs Fixed Timers
60
Interaction with routing

• SPAN gets the connectivity matrix and the
  neighbor list from a routing protocol. In
  addition, it needs to modify the route lookup
  process such that routing uses only coordinators
  to route packets
• GAF/CEC and ASCENT do not depend on the routing
  mechanisms, nor they need to modify them

61
Other issues

• None of the previous schemes required any
  particular MAC
• It would be interesting to study the synergy
  effect that these schemes may have with energy
  savings MACs (S-MAC, UCB MAC, etc)
• Also interesting to study the synergy effect with
  high-level energy savings mechanisms (STEM)

62
Energy Savings

• SPAN (2), GAF(3), CEC(3.5), and ASCENT (5.5)
  achieve comparable energy savings, albeit their
  goals are different
• SPAN and ASCENT have sublinear energy savings as
  a function of density (because they periodically
  check the status of the network). GAF/CEC and
  ASCENT (w/adaptive timers) have linear energy
  savings as a function of density
• Further studies are required to make more
  conclusive statements

63
Topology Control from a Sensing Perspective

• So far we have considered only the communications
  perspective
• Sensing coverage model
• typically unit disk sensing
• note depends on object being sensed
• Node deployment model
• deterministic with (no failures or with isolated
  failures)
• approximated by a pdf or is random (as a result
  of rampant errors)
• Coverage requirements
• Point coverage (deterministic or probabilistic
  guarantee)
• Barrier coverage (deterministic or probabilistic
  guarantee)
• Worst-case coverage least exposed path
• Tracking coverage any uncovered path has length
  at most l

64
Sensing Coverage References

• Survey
• Coverage in Wireless Sensor Network, Mihaela
  Cardei, Jie Wu
• For 1-coverage
• Pater Hall, "An Introduction to the Theory of
  Coverage Processes, 1988
• For k-coverage
• Santosh Kumar and Balogh, Mobisys 2004
• For k-coverage poisson deployment
• Honghai Zhang and Jennifer Hou, Mobihoc 2004

65
Coverage Results

• 1-point coverage with deterministic placement
• hexagonal layout is optimal
• k-point coverage with deterministic placement
• question of optimal placement is open
• k-point probabilistic coverage
•  
• almost always k-coverage for poisson deployment
• n?r2 ln(n) k ln(ln(n)) (error term)
• where n is sensors and r is sensing radius
• almost always k-coverage for random uniform
  deployment
• has essentially same result

66
Coverage Algorithms

• Checking whether network is not suitably covered
• point coverage violation check is possible
  locally
• Maintaining coverage via sleep-wakeup
• optimal scheme is NP-Complete, if deployment
  unknown
• (so heuristics used)
• random independent scheduling, if deployment
  uniformly random
• sentry rotation between redundant nodes in each
  cluster/region

67
Both Communication and Sensing Topology Control

• Relation between sensing radius and communication
  radius
• If Comm radius 2 x Sensing radius
• then (k-coverage ? k-connectivity)