(Georgia Institute of Technology, 2008)
Bamba, Bhuvan; Liu, Ling; Iyengar, Arun; Yu, Philip S.
Spatial alarms are personalized location-based triggers
installed by mobile users to serve as a reminder of a
location of interest to be encountered in their future trips.
Unlike continuous spatial queries, spatial alarms do not require
immediate processing and periodic reevaluation upon installation.
Thus, a critical challenge for efficient processing of spatial
alarms is to determine when to evaluate each spatial alarm,
while ensuring the demanding requirements of high accuracy
and system scalability. In this paper, we compare alternative
approaches for evaluation of spatial alarms: periodic evaluation,
safe period-based processing and safe region-based processing.
We argue that the safe region-based approach provides highly
efficient processing of spatial alarms at the server. Furthermore,
it reduces wireless communication costs and energy consumption
on the client side by reducing the number of location
updates to be transmitted to the server without sacrificing
accuracy of spatial alarm evaluation. We develop safe region
computation techniques based on different heuristics, namely,
Maximum Perimeter Rectangular Safe Region (MPSR), Largest
Component Rectangles Safe Region (LCSR) and Bitmap Encoded
Safe Region (BSR) approach, and present an in-depth study on
trade-offs involved in the selection of an appropriate safe region
computation strategy. Our experimental evaluation shows that
the best optimization strategy requires an approach which adapts
to changing system load conditions and resource constraints, as
none of the safe region computation techniques outperforms the
others on all relevant evaluation metrics. Experimental evaluation
also validates our conjecture that safe region-based processing
offers close to optimal performance in terms of CPU load on the
server and wireless communication costs at the mobile clients.