ASCENT: Adaptive SelfConfiguring sEnsor Networks Topologies

1
ASCENT Adaptive Self-Configuring sEnsor Networks
Topologies
EECS 600 Advanced Network Research, Spring 2005
Shudong Jin January 24, 2005
2
Outline

• Motivation of ASCENT
• How ASCENT works, with examples
• Analysis/Simulation/Experimental results
• Several open issues

3
Motivation Power Management

• Wireless sensor networks utilize high density of
  nodes
• Redundancy allows network to remain functional
  without all nodes participating
• Intentionally limit the number of communicating
  nodes at a given time, to save total network
  energy and energy usage of active nodes
• Dense sensor nodes self-configure to establish a
  topology
• Large number of nodes precludes manual
  configuration
• Environmental dynamics precludes design-time
  pre-configuration
• This paper attacks the self-configuration
  problem.
• Section II, a discussion on sensor network
  scenario (constraints, objectives, and
  assumptions)

4
Basic Ideas of ASCENT

• Each node assesses its connectivity and adapts
  its participation in the multi-hop network
  topology based on the measured operating region
  (why local?).
• Signals when it detects high message loss,
  requesting additional nodes in the region to join
  the network in order to relay messages.
• Reduces its duty cycle if it detects high message
  losses due to collisions.
• Probes the local communication environment and
  does not join the multi-hop routing
  infrastructure until it is helpful to do so.

5
ASCENT and routing

• ASCENT determines topology ASCENT simply decides
  which nodes should join the routing
  infrastructure.
• It runs above the link and MAC layer, and below
  the routing layer.
• ASCENT is not a routing or data dissemination
  protocol.

6
General Process
Communication Gap
Active Neighbor
Passive Neighbor
7
Node States

• Active
• Forwards data and routes packets
• Passive
• Monitors network for neighbors and data loss
  rates
• Periodically checks if necessary to become active
  
• Test
• Sends neighbor announcement message
• Monitors network for neighbors and data loss
  rates
• Forwards data and routes packets
• Sleep
• Turns radio off and goes to sleep

8
State Transition Diagram
After Tt
Test
Active

• Neighbors lt NT
• And
• Loss gt LT or
• Loss lt LT and Help

Neighbors gt NT (high ID for ties) Or Loss gt
previous Loss
After Tp
Passive
Sleep
After Ts
9
Parameter settings and tradeoffs

• NT Neighbor Threshold
• LT Loss Threshold
• Tt Test state timer
• Tp Passive state timer
• Ts Sleep state timer

10
Determining Data Loss and of Neighbors

• Each node adds a sequence number to each packet
• Sequence number allows node to detect lost
  packets from each neighbor
• Requires every node to maintain state about nodes
  that have sent a packet that has been received by
  the node
• Number of active neighbors (N) is the number of
  neighbors with link packet loss smaller than the
  neighbor loss threshold (NLS)
• NLS 1 (1/N)
• If N 4, NLS 75
• If N 10, NLS 90
• Intuitively, the more neighbors there are, the
  more likely that collisions will occur so the
  neighbor loss threshold is increased (why 1-1/N?
  not justified).

Seq No. Regular Packet Data
11
Analysis probability of collision
CSMA with a random back-off, chosen from S slots
12
Analysis - delay
13
Analysis energy saving
a Tp/Ts ß power in sleep power in idle
14
Simulation/Experimental Setup

• Discrete event driven simulator collisions occur
  if two or more events received within contention
  period
• Transmission success probabilities
• 0-A 1
• A-B linearly decreasing from 1 to 0
• B- 0
• Routing protocol directed diffusion
• TtTpTs 2530, NT 4, LT 20
• Uncongested network
• Energy model TxRxIdleSleep on the order of
  1001001001

15
Simulation/Experimental Setup (cont.)

• Node density is determined by the number of nodes
  between two nodes that are just within radio
  range of each other
• Sources and sinks are placed on opposite sides of
  the network
• It is not explained how many nodes are along each
  side (i.e. if the layout is more like a line or a
  square)

16
Packet Loss versus Node Density
Analysis, simulation, and experimental
results Packet loss withASCENT is not
affectedby node density.
17
Event Delivery Ratio versus Node Density
Experiment three hops. Simulation six
hops.Event delivery ratioonly requires
onemessage to reachsink, which is highlylikely
with diffusionrouting. ASCENT allows
highdelivery unaffected bynode density
18
Network Lifetime versus Node Density
Network lifetime isdefined as the time for90
of the nodesalong the source-to-sink path to
run out ofenergy. ASCENT greatlyextends the
networklifetime. Network lifetime is
nearlythree times with ASCENTand a node density
of 40. (why not linear?)
19
ASCENT Summary

• ASCENT runs above the link and MAC layer and
  below the routing layer
• Independence of ASCENT from link and routing
  layers allows possibility of finding best fit for
  given application
• Establish a small number of nodes to maintain
  good connectivity and allow other nodes to sleep
• Sleeping nodes occasionally awaken to see if
  network changes require nodes to participate in
  forwarding data and routing packets

20
Open issues for discussion

• Many, just to show a few.

21
Issues with Help Messages

• Broadcasting help message will cause nodes to go
  from passive to test states needlessly in areas
  where help is not needed, increasing collisions
  in that area temporarily

Passive Neighbor
Active Neighbor
Neighbor sending help message
22
Issues from Aging of System
After a certain amount of time, active nodes will
eventually die off.Neighboring active nodes must
detect this loss and issue help messageor
neighboring passive nodes must detect loss and
switch to teststate.
23
Issues from Aging of System (cont.)
Ideally, one node becomes activewhen only one
node stops working.All active nodes have four
neighbors. Possible (more than likely?)
thatmultiple nodes will become active.Already
active nodes will now havemore than four active
neighbors.
24
Issues from Aging of System (cont.)

• Decision to become active is based only on nodes
  own (very local) observations and not through
  information exchange with neighboring nodes
• Nodes may therefore decide to become active
  unnecessarily
• Lifetime of network could be extended with more
  information before nodes decide to become active.
• More communication may then be required
• But, just a little local connectivity information
  may help.

25
Issues with Active State Transition

• Should there be a state transition away from the
  active state?
• Better to drain a node completely or to drain an
  area of nodes more slowly?
• Spread energy usage to all nodes
• Different sets of active nodes (dynamic vs.
  static)
• Prevent partitioning for as long as possible
• Impact on routing protocols
• Routing setup more often
• Chance to find good routing sometimes rather than
  risk a long-lasting bad route

26
Issues with Neighbor Threshold Value

• Dependent upon application
• Assumes mostly homogeneous network
• Different nodes/sensors may need different
  numbers of the same kind of node/sensor as a
  neighbor
• E.g. only one temperature sensor needed in an
  area, but tracking sensors will need other such
  sensors in the area to also be active to usefully
  track object
• Satisfying all threshold values may force most
  nodes to be active

27
Other thoughts, and more?

• Dynamically change Ts and Tp depending on density
  of active and passive neighbors
• Ts should be smaller if node is in a sparse area
  (i.e. node is more likely to become active
  sooner).
• How to know number of passive neighbors?
• Routing protocol must be able to easily adapt to
  network changes
• Useful experiment would be to run ASCENT on a
  large network and examine which nodes become
  active
• Make decisions location-aware