Termite: Emergent AdHoc Networking

1
Termite Emergent Ad-Hoc Networking

• Martin Roth
• Wireless Intelligent Systems Laboratory
• Cornell University
• Ithaca, New York, USA
• Conf.
• M. Roth, S. Wicker, Termite Ad-Hoc Networking
  with Stigmergy, IEEE 2003 Global Communications
  Conference (Globecom 2003), December 2003.
• M. Roth, S.Wicker, Termite Emergent Ad-Hoc
  Networking, The Second Mediterranean Workshop on
  Ad-Hoc Networks, Medhia, Tunisia, 2003
• Book.
• M. Roth, S. Wicker, Termite A Swarm Intelligence
  Routing Algorithm for Mobile Wireless Ad-Hoc
  Networks, Springer SCI Series Swarm Intelligence
  and Data Mining, 2005
• Review by
• Paskorn C, _at_ DSSG

2
Termite

• Termite is a routing algorithm for mobile
  wireless ad-hoc networks based on principle of
  swarm intelligence.

3
Ad Hoc Network

• Authors interest is how to route information in
  mobile wireless ad hoc networks
• Nodes in the network
• Move about freely
• Communicate via a wireless channel
• No predefined network structure exists
• Topology is dynamic
• How can information be moved through the network
  if the paths are continuously changing?

4
Swarm Intelligence

• A model of natural swarm behavior
• System of many independent agents obeying
• Positive Feedback
• Reinforce good solutions
• Negative Feedback
• Remove old solutions
• Randomness
• Allows many solutions to be simultaneously tested
• Multiple Interactions
• Spreads local information
• Generates global behavior from local interactions

5
Stigmergy and Emergence

• Stigmergy
• Communication between individuals through a
  common environment
• Pheromone gradients towards hills
• Emergence
• Characteristic of a system in which unforeseen
  properties arise from the interaction of many
  individuals
• System properties are not individual properties
• Whole is greater than the sum of the parts
• A critical number of termites is needed to build
  a hill

6
Termite Hills

• Given a surface, some pebbles, and a few
  termites
• Termites walk randomly over the surface picking
  up pebbles and dropping them near others
• Pebbles are infused with pheromone
• Termites are attracted to pheromone
• Positive feedback for good solutions
• Pheromone evaporates
• Negative feedback
• Prevents bad solutions from remaining in the
  collective memory

7
Termite - termite Analogy

• What is the analogy between Termite and termites?
• Packets Termites
• Nodes Hills
• Pheromone Table Environment
• Packets communicate with each other through the
  environment
• Data Packets follow the pheromone trail to their
  destination while laying a trail to their source

8
Making the Analogy Fit

• Assumptions needed to make Termite work
• Links between nodes are symmetric
• If a node A can communicate with a node B, B can
  communicate with A
• Communication patterns across the network are
  uniform
• A is equally likely to send a message to any
  other node in the network

9
Neighbor Management
D
A
B
S
F
C
E

• Each node maintains a list of neighbors
• A, B, C
• Neighbors are lost when they cannot be contacted
• Transmission to neighbor fails
• Transmissions from neighbor are never heard again

10
Pheromone Table

• Each entry contains Pn,d (amount of pheromones)
• n is neighbor index d is destination
  index
• Pn,d ( amount of pheromones form node d on the
  link with neighbor n. )

D
A
B
S
n
F
C
E
G
d
11
Pheromone Update Pn,d

• When packet arrives at a node, the pheromones for
  the source of the packet is incremented by a
  constant ( 1)

Bs Pheromones Table
D
A
B
Hop p
C
S
12
Pheromone Decay

• Each Pn,d is periodically multiplied be the decay
  factor e-? , ?gt0
• Decay Period 1 secod
• If pheromone for a particular node decays, then
  the corresponding row or column is removed from
  the pheromone table

13
Pheromones Bound

• Pheromones Ceiling
• Pheromones Floor
• Initial Pheromones
• If packet is received from an unknown source, a
  new entry for that node is created in the
  Pheromones Table
• New Pn,d initial pheromones value
• Pheromones Floor lt Pn,d lt Pheromones Ceiling

14
Route Selection

• Packet from Node B
• Destination is D
• This packet is routed randomly based on the
  pheromone present on the neighbor links of S
• Packet will not be forwarded back to B
• If only one neighbor, the packet is dropped

D
A
B
S
C
15
Probability

• The transformation of pheromone for d on link n (
  Pn,d ) into the probability (pn,d) that the
  packet will be forwarded to n
• K pheromone bias (if K large, we needs
  pheromone to be large to have effect)
• F pheromone exponent (amplifier)
• emphasize difference between links

16
Routing Example
A packet should be forwarded from S to D. K 0 F
2
D
A
Pheromone 53.27 Prob 0.77
Destination Address D
B
Neighbor Address A Pheromone 10
Pheromone 2.60 Prob 0.19
Neighbor Address B Pheromone 5
Pheromone 1.03 Prob 0.03
Neighbor Address C Pheromone 2
C
17
Packet Design

• Source address
• Destination address
• Previous hop address
• Next hop address
• Message identification (not use)
• TTL ( time to live) (eliminate loop)

18
Route Packet

• There are 5 type of packets
• Data Packets
• Route Request Packets
• Route Reply Packets
• Hello Packets
• Seed Packets

Control packet
19
Data packets

• Contain data.
• Data packet is routed in network
• If node does not know how to forward packet
• ( No destination in pheromone table)
• Data packet is stored and route request packet is
  issued.
• If reply is not received within time period, the
  data packet is dropped

20
Route Management

• Route Request (RREQ)
• Random walk to find flow from destination
• Not flooding
• Route Reply (RREP)
• Routed to the source according to theForwarding
  Equation
• Same as data packets
• Control packets are forwarded with high priority

21
Example
Source S Destination D
22
Route Management (Cont.)

• Hello Packets
• are used to search for neighbors when a node
  become isolated (pheromone table is empty). Hello
  packets are broadcast a a regular interval until
  reply is received. So the node can create
  pheromone table
• Seed Packets , are used to initial pheromone
  table.
• Seeds make random walk through network.
• It is useful for reducing hop count

23
Simulation

• Termite has been simulated using
• Opnet Technologies Inc
• www.opnet.com
• 100 nodes
• 600 seconds
• 64 byte data packet
• Random destination
• Each node send 2 data packets per sec

24
Random Way Point
50m
50m
25
Data Goodput and Overhead

• Data Goodputfalls
• Control Overhead
• Control packet /data transmitted (not include
  retransmitt)
• increases
• Bandwidth Overhead constant
• Bits of control packet / all data ( include
  retransmitted)

Data good put
Control ovehead
Bandwidth overhead
26

• The result that bandwidth overhead is constant is
  a benefit of Termite
• Because there is no flooding
• Only two RREQ packet are send

27
Path Length

• Data Hops grow large
• Route Request and Route Reply Hops
• remain short

Why?
28
The reason..Why data hops grow large.

• Average next hop probability drops quickly

29
The reason..Why Route Request hops remain short.

• Requested nodes remain within two hop distance

Control packet do not have to travel far to meet
their target
30
Final Remarks

• The principles of swarm intelligence are used to
  create a simple approach to ad-hoc routing
• Control Paths are consistently small
• Termite is able to maintain nearly constant
  overhead regardless of node mobility

31
Thank you

• Question ?

32
Stigmergy and Emergence

• Stigmergy
• Communication between individuals through a
  common environment
• Pheromone gradients towards hills
• Emergence
• Characteristic of a system in which unforeseen
  properties arise from the interaction of many
  individuals
• System properties are not individual properties
• Whole is greater than the sum of the parts
• A critical number of termites is needed to build
  a hill