200 likes | 289 Views
Motivating Sensor Network Research: The Applications and Computer Science Issues. Prabal Dutta and David Chu. What Makes Good Application-Led Research?. Richard Sharp and Kasim Rehman. Perspectives. “Applications are of course the whole point of ubiquitous computing” Mark Weiser [Wei93]
E N D
Motivating Sensor Network Research: The Applications and Computer Science Issues Prabal Dutta and David Chu
What Makes Good Application-Led Research? Richard Sharp and Kasim Rehman
Perspectives • “Applications are of course the whole point of ubiquitous computing” • Mark Weiser [Wei93] • “We need to increase the applications deployed to books written ratio in sensor networks” • Deborah Estrin [Personal Communications] • “In the future, increasing proportion of computer science research will be application-driven” • Eric Brewer and Mike Franklin [CS262A-Fa04]
Defining Application-Led Research • Application-Led Research • Driven by domain problem • Evaluated by quantifying benefits brought to domain • Technology-Led Research • Not necessarily motivated by potential domain benefits • Interesting or challenging from a technical perspective • Research Goals Should (do you agree?) • Identify users’ problems and application requirements • Provide infrastructure developers with application requirements • Validate technology and provides insights into its use
Selecting Applications • Will this change the way people think? • If nothing changes after your research, what’s the point? • Must make an impact on computer science • Just impacting biology or civil engineering is not enough • Starting from scratch can make this more difficult or easier • If system building, what will you learn from it? • There must be an important question in there! • Identify and attack “severe and persistent problems” • Avoid trivial “proof-of-concept” research projects • Team up with domain experts when selecting problems • Make sure there’s a concept and it’s worth proving
Implementing Applications • To start from scratch or not? • Benefits? • Drawbacks? • Is building reusable infrastructure worth it? • Research community values novelty over good engineering • Research community doesn’t value implementation as research • Do you agree? • Reframe the question: What are your options? (Aside) • Your efforts can be directed structurally or strategically • Structural: change the community so that it values infrastructure • Strategic: pick the right topic, and your work will be broadly used (and well referenced)
Evaluating Applications • Small, lab-scale evaluations • Useful: in the early stages of design • Insufficient: impossible to understand the impact of • Environment on technology • Technology on environment • NEST FE Provides some good examples • Applications are evaluated only against themselves • Self-evaluation is insufficient • Requires applications, infrastructure, and data to be shared • Is this a good idea? • Is it done in other fields?
Recommendations • Choose applications carefully • Address severe persistent problems; avoid trivial ones • Share technical infrastructure • Design reusable SW/HW; publicly release code • Evaluate applications in realistic environments • Only way to investigate interactions between tech/env/users • “The real world is it’s own best model” – Rodney Brooks • Perform comparative evaluations • Release data sets from field trials; allows other to analyze
Allen Newell’s Research Style • Good science responds to real problems • Don’t pick fantasy problems; there are too many real ones • Good science is in the details • Takes the form of a working model • Includes detailed analysis or implemented models • Good science makes a difference • Measure of contribution is in • How it solves real problems • Shapes the work of others
Some Computer Science Issues in Ubiquitous Computing Mark Weiser
Are We There Yet? • Hundreds of Tabs? • Tens of Pads? • One or two Boards?
Did Their Work Have Impact? • Yes! Due to emphasis on computer science issues: “The fruitfulness of ubiquitous computing for new computer science problems justified our belief in the…framework” • Issues like • Hardware components • Low power (P=C*V^2*f gives lots of degrees of freedom) • Wireless (custom radios (SS/FSK/EM-NF bits/sec/meter^3 metric) • Pens (how do you write on walls?) • Network Protocols • Wireless media access (MACA: RTS/CTS) • Gigabit networks (lot’s of little devices create a lot of traffic) • Real-time protocols (IP telephony) • Mobile communications
Connecting the Physical World with Pervasive Networks Deborah Estrin, David Culler, Kris Pister, Gaurav Sukhatme
Goals • Goal: to measure the physical world • Across large spaces • Over long periods of time • Using multiple sensing modalities • In remote, and largely inaccessible locations “The physical world is a partially observable, dynamic system, and the sensors and actuators are physical devices with inherent accuracy and precision limits.”
Challenges • Immense scale of distributed systems elements • Vast numbers of devices • Fidelity • Limited physical access • Embedded in the environment • Remote, expensive, or difficult to access • Wireless communications • Energy harvesting or very moderated energy consumption • Extreme dynamics • Temperature, humidity, pressure, grass height, … • Passive vigilance to a flurry of activity in seconds
Challenge: Immense Scale NEST FE: 557 Trio Nodes, Self-powered, self-maintaining, GPS ground truth, multiple subsets
Challenge: Limited Physical Access Redwoods to appear Sensys 05
Challenge: Extreme Dynamics ExScal • Border Control • Detect border crossing • Classify target types and counts • Convoy Protection • Detect roadside movement • Classify behavior as anomalous • Track dismount movements off-road • Pipeline Protection • Detect trespassing • Classify target types and counts • Track movement in restricted area