The Architecture

1
The Architecture

• A wireless ad-hoc distributed computing
  environment
• Harnesses and aggregates low computing power of
  geographically-concentrated mobile devices even
  sensors in sensor networks
• Suitable for execution of Cellular Automata -
  based applications/ simulations
• Provides a bounded region of euclidean space to
  the application a virtual lattice V

Mobile phones
Berkeley Mote sensors
2

• A fixed immobile node I forms the origin of the
  lattice
• Nodes calculate their location relative to I
  (using algorithms in 1 )

• Based on location, they now form a 2-dimensional,
  physical lattice P
• P is logically re-arranged to form a virtual
  lattice V with dimension, size, etc. based on
  application requirements

Lattice origin I
Participant nodes

• The application is aware only of V P is
  transparent
• Accurate timing of communication is often
  critical to the simulation
• A neighbor in V is not necessarily a neighbor in
  P thus messages to neighbors in V may not reach
  them simultaneously, causing erroneous simulation
  results

3

• The communication sub-system ensures all messages
  are processed by nodes only after the maximum
  possible propagation time resolving the timing
  issue

• Upon completion of lattice formation, the
  application execution is initiated
• Mobility of participating devices and device
  failure can lead to the development of holes in
  the lattice

• Formation of lattice(s) in WAdL
• Unorganized mobile nodes
• A physical lattice - L is formed
• L is logically re-mapped to form a 3-D virtual
  lattice - V

4

• Strategies helpful in tackling node mobility /
  failure
• Neighbors working for failed / moving devices
• Multiple devices responsible for a lattice vertex
  performing tasks in parallel so that one of the
  backup devices take over when the primary device
  fails

• Physical obstructions might prevent direct
  communication between neighbors in P
• Use of a simple routing mechanism - utilizing
  devices adjacent to the obstruction, can help
  resolve this issue.

5
Related Work

• Many physical phenomena have complex analytical
  solutions - Analog models can be used to
  predict their behavior

Operation of Cellular Automata

• Some analog simulations can be modeled using
  Cellular Automata (CA)
• CA are dynamic - discrete in space and time
• Behavior completely specified in terms of local
  relations
• Lattice Computer can execute CA-based simulations
• Low computational demand processing elements
• Represents euclidean space where phenomenon
  unfolds

CA used in modeling a snowflake
6
W reless
Lattice
Vishakha Gupta
and Current affiliations (MSIN, CMU)
7
Ad-hoc
Computer
Mentor Dr. Anil M. Shende
(Roanoke College)
Gaurav Mathur, BITS-Pilani, India (Intel,
India)
8
Usage Scenarios

• Extremely cheap computing grids can be formed
  using clusters of cheap Mote-like devices /
  sensors
• Message routing in a wireless network
• Providing load-balancing and/or fault tolerance
  in a wireless network
• Some applications might need a structured network
  WAdL can help provide structure to an
  otherwise unstructured network

9
The Application

• We demonstrate an application based on simplified
  CFD model
• Computes the ideal lift and drag on an airplane
  wing
• Virtual wing flies in the virtual lattice
  generated by WAdL

Aerofoil and direction of lift and drag
Virtual flight of the simulated wing
10
Simulation Results

• Obtained simulation results are identical to
  analytical results
• Uses minimal network bandwidth causing
  negligible disruption to existing network traffic

Change in Lift generated by the Virtual Wing due
to Decreasing Density in V (plotted from
simulation data)
Bandwidth Utilization in WAdL with 1000 nodes
11
Future Work

• Linking multiple, geographically remote WAdLs
  together to form a single WAdL
  providing more euclidean space for
  simulation
• Routing messages around physical obstructions in
  a WAdL
• Using a WAdL for routing and addressing network
  congestion in a wireless setting
• Distributed clock synchronization

12
References
1 Anil M. Shende, Vishakha Gupta, Gaurav
Mathur. Lattice formation in a Wireless Ad-hoc
Lattice computer (WAdL). AlgorithmS for Wireless
and mobile Networks (A-SWAN), August 2004. 2 D.
S. Rajan, J. Case, A. M. Shende. Optimally
representing euclidean space discretely for
analogically simulating physical phenomena. In
Foundations of Software Technology and
Theoretical Computer Science, December 1990.
(Lecture Notes in Computer Science) 3 Donald
Greenspan. Deterministic Computer Physics.
International Jounal of Theoretical Physics,
1982. 4 L. Wilson A. Wadaa, S. Olariu. On
training a sensor network. In Proceedings of the
International Parallel Distributed Processing
Symposium, page 220, 2003. (Workshop on Mobile
Adhoc Networks) 5 C. L. Barrett, S. J.
Eidenbenz, L. Kroc, M. Marathe, J. P. Smith.
Parametric probabilistic sensor network
routing. Proceedings of the 2nd ACM
international conference on Wireless sensor
networks and applications, page 122-131,
2003. 6 Factual data for lift and drag on an
aerofoil.http//www.centennialoight.gov. 7
Network simulator 2 (ns-2). http//www.isi.edu/nsn
am/ns/.