MagicWeaver

An Agent Based Simulation Framework for Wireless

Sensor Networks

What Should You Take Away?

• Extremely High Level Overview of Sensor Networks

• Design Rationale of MagicWeaver• Capabilities of MagicWeaver• Research & Implementation – Where do

we draw the line?

Outline

• Sensor Networks– Overview of Sensor Networks– Typical Characteristics of Sensor Devices– Typical Application Areas– Research Areas

• MagicWeaver– Motivation– Design,Implementation– Discussion & Future Work– What I gained from it?

Sensor Nets Overview

• Caters to the environments in which humans operate

• Step towards “Everyday objects acting as computing devices”

• Adhoc Networks Vs Sensor Networks– Nature of Deployment - Planned– Degree of Device Constraints - Extreme– Functioning Mode - Unattended– Scale of Devices – Huge Numbers

Sensor Nets Overview– Infrastructure: Often Present– Task Definition: Well Defined– Location: Known– Replenishment: Often Not Possible– Applications: Towards an Environment

“Human-Centric to Human Supervised Computing”

Characteristics of Sensor Devices

• Array of Sensors: Temperature, Pressure, Light, Acoustic, Motion Detectors etc.

• Non-replinishable Energy - Currently• Less than 4 KB of memory• No Persistent Storage• Processing Capacity: 4 MHz to 20 MHz• Communication Radius: Few Tens of Feet

Berkeley Motes-weC Mini Mote• 19.1 Kbps • 20m Range• Light Sensing• Temperature Sensing• 4 MHz – 3.0 V• 8 Kb –Program

Memory• 512 b – Data Memory• Available from:

CrossBow Inc. $900 for a complete kit.

Application Areas

• Military Environments– Surveillance information/Reconnaissance

Missions

• Bio-medical Sensing– Monitoring of conditions inside human body

• Less Accessible environments like large Industrial Plants and Aircraft Interiors

• Smart Spaces – Smart Offices and Rooms, Rock Concerts

Research Areas in Sensor Nets• Hardware Research

– Practically deployable devices– Smart Dust

• Networking Research– Energy-aware routing protocols

• Data Management Research– Energy-aware data propagation schemes– Data Fusion and analysis

• Software Engineering Research– Platforms for building sensor applications

MagicWeaver• Why ?

– Thesis in Sensor Nets– Data Management – Key Focus– Research is recently spurred

• What ?– Simulation Framework For Sensor Networks– Multi-agent system modeling sensor nodes

and its actions– Software API for incorporating various

models

“Let’s Weave Magic”

Course of Execution

• Initial Work plan– To Understand various data propagation

schemes• SPIN, Directed Diffusion, Flooding and Gossipping

– To gain further insights into the domain

• System Built– Framework for Simulation– Models in Sensor Networks

“Final Link is Yet to be Established”

Related Work

• SensorSim– Closest it can get – NS-2 Mobility and Wireless Extensions +

Sensor Network Extensions

• SensorWare– Middleware for Mobile Code Support

• Slew of Work in USC/ISI, UCLA, Berkeley

Overview of MagicWeaver• Agents-Sensors – High Degree of

Correlation• Heterogeneity of Algorithms – Warrants a

Flexible framework• Abstracts Models pertinent in Sensor

Networks• Runtime Environment for execution under

the purview of “User-Defined” models.• Object-Oriented Model Definitions

Java + JADE

Towards a Framework

Agent Framework

Tasking & Task Model

Environment Model

Device &Cache Model

Data Propagation

Model

Motes, LRRRandom

Temp,PresMotion, Sound

Flooding

Location & Clustering

Model

(X,Y)-Distance Based

Base Statio

n Model

Data Format

Agent Components

Logger

Network Topology AgentBase Station Agent

Environmental Agent

Sensor Agent

Design of a Sensor Agent

DCLDPDL STL

Controller

Introspector

NT

Agent PlatformDCL: Device Constraint Layer STL: Sensor Tasking LayerDPDL: Data Propagation Definition LayerNT:Network Topology

Interaction Pattern

Get Locations(SA To NTA)

Get Locations(SA To NTA) Compute Neighbors, Assign

BS,EA (NTA To All)

Compute Neighbors, AssignBS,EA (NTA To All) Start Sensing

SA to EA

Start SensingSA to EA

Send Sensed /Forwarded Data

Send Sensed /Forwarded Data

If BS ConIf BS ConSend to BSSend to BS

Analyze DataAnalyze Data N

Sensor Agents Network View

Base Station AgentNetwork Topology Agent

Logger

Env Agents Sensor Agents

Facts Used in the Simulation

• Energy is spent while receiving, transmitting and waiting to receive wirelessly

• Energy is spent while doing the sensing task(s)

• Energy is spent while forwarding messages (agents acting as routers)

• Energy is spent while switching tasks • Only a few nodes have base station

connectivity

Utilities

• SensorView – Graphical Tool for displaying the view of the sensor network

• BaseStation View – Graphical Tool for displaying the data received at the base station.

• JADE Sniffer/RMA

Discussion

• Complete Modular Design• Complexity left to the Sensor Network

Designer• Runtime environment• Agent paradigm completely abstracted

out

Further Work • Implementation of data propagation schemes

– Gossiping, SPIN, Directed Diffusion

• Query Tool for Node Querying• Models of failure-prone nature of nodes and

loss-prone nature of wireless links• Dynamic switching of tasking policy• In-built support data analysis• Design of Test-bed for experimental evaluation

– On what basis do we compare the schemes implemented

– How do you actually carry out the simulation to get the results you want to measure

What Did I Gain?

• Sound understanding of the domain of sensor networks

• More useful insights into JADE based agent programming

• Prospective area of further work – Towards Masters thesis

“Complete Satisfaction”