GIT-CERCS-03-18
David Robinson, Karsten Schwan,
Dynamic Management of Cooperating Peers in Wireless Embedded Systems: An
Approach Driven by Information Quality
Data-intensive applications like remote sensing or
peer-to-peer communications are increasingly prevalent on
wireless-connected, pervasive systems platforms. Since
resource limitations make it infeasible to deliver all data to all
participants at all times, these applications' data streams must be managed
at runtime. The idea explored in this paper is to continuously manage
data streams in terms of the quality of the information received by end
users. Methods are developed that dynamically optimize application-relevant
metrics of information quality. These methods are realized in
peer-to-peer data delivery middleware, by continuously adapting
application-specific, source-resident data filters when delivering
information to end users. Results are presented for portable devices
operating across 802.11b-based wireless networks, using camera-captured
images as sample sensor data and using filters that implement a variety of
data downsampling methods, with consequent effects on information quality.
Experimental evaluations show that (1) dynamic management of information
quality is feasible and generates substantial overall quality gains compared
to non-managed approaches, and (2) that the complexity of multi-sensor,
multi-client quality management indicates the use of heuristic solution
algorithms specific to certain system configurations.