DIM for sensor networks

Overview

In wireless sensor networks, data or events will be named by attributes. Many of these attributes will have scalar values such as temperature, light levels, soil moisture conditions, to name a few. In these systems, one natural way to query for events of interest will be to use multi-dimensional range queries on these attributes. For example, scientists analyzing the growth of marine micro-organisms might be interested in events that occurred within certain temperature and light conditions: ``List all events that have temperatures between 50° F and 60° F , and light levels between 10 and 20''. Such range queries can be used in two distinct ways. They can help users efficiently drill-down their search for events of interest. The query described above illustrates this, where the scientist is presumably interested in discovering, and perhaps mapping the combined effect of temperature and light on the growth of marine micro-organisms. More importantly, they can be used by applications running within a sensor network for correlating events and triggering actions. For example, if in a habitat monitoring application, a bird alighting on its nest is indicated by a certain range of thermopile sensor readings, and a certain range of microphone readings, a multi-dimensional range query on those attributes enables higher confidence detection of the arrival of a flock of birds, and can trigger a system of cameras.

In traditional database systems, such range queries are supported using pre-computed indices. Indices trade-off some initial pre-computation cost to achieve a significantly more efficient querying capability. For sensor networks, a centralized index for multi-dimensional range queries may not be feasible for reasons such as energy-efficiency, network capacity, fault tolerance, etc. Rather, there will be situations when it is appropriate to build an in-network distributed data structure for efficiently answering multi-dimensional range queries.

In this project we are working on the design and implementation of a Distributed Index for Multi-dimensional data or DIM, as we call it. DIM leverages two key ideas: in-network data centric storage and locality-preserving geographic hashing.

Papers

Posters

Talks

Software

Contact

Please email to



Last Modified: 28 Feb 2006