Given the vast amounts of data collected concurrently from a multitude of sources and distributed across a large number of stores, state-of-the-art techniques are proposed to manage this data and the computational resources that process it. The research proposed here addresses a vertical slice through these techniques, starting with sensor networks, emphasizing devices and their interoperability, progressing to computing grids, and finishing with multiagent systems. This area in the first instance provides architectural support for data fusion and data mining but also supports (in part indirectly through the other two areas) the other thrusts. We are implementing a sensor network that will exploit the unique access of the University of Alaska Southeast to both the ocean and glaciers for use by diverse scientific projects, to involve undergraduate students, and to be visible to the Juneau community. The network is intended to accommodate new sensors and various data rates and communication bandwidth needs. Grid computing promises to make data and processing resources available transparently and on a very large scale. Mining spatio-temporal patterns across multivariate stream data available on a grid, however, faces several fundamental challenges, such as how to apply a multitude of analysis components and how to locate and integrate the heterogeneous, geographically distributed sensors [Tham and Buyya, 2005]. In practice, data acquisition, fusion, and processing systems tend to have rigid control flow, which inhibits responsiveness; multiagent systems promise flexibility, but their overall behavior is virtually impossible to test since the separate threads of control can be interleaved in a multitude of ways.
[Tham and Buyya, 2005] Tham, C., Buyya, R., “SensorGrid: Integrating Sensor Networks and Grid Computing,” Technical Report, GRIDS-TR-2005-10, Grid Computing and Distributed Systems Laboratory, University of Melbourne, 2005.