(Georgia Institute of Technology, 2011)
Doo, Myungcheol; Liu, Ling
One of the most challenging problems in scaling spatial alarm processing is to compute alarm free regions
(AFR) such that mobile objects traveling within an AFR can safely hibernate the alarm evaluation process
until approaching the nearest alarm of interest. In this paper we argue that maintaining an index of both
spatial alarms and empty regions (AFR in the context of spatial alarm processing) is critical for scalable
processing of spatial alarms. Unfortunately, conventional spatial indexing methods, such as R-tree family,
k-d tree, Quadtree, and Grid, are not well suited to index empty regions. We present Mondrian Tree - a
region partitioning tree for indexing both spatial alarms and alarm free regions. We first introduce the Mondrian
tree indexing algorithms, including index construction, search, and maintenance. Then we describe a
suite of Mondrian tree optimizations to further enhance the performance of spatial alarm processing. Our
experimental evaluation shows that the Mondrian tree index outperforms traditional index methods, such
as R-tree, Grid, Quadtree, and k-d tree, for spatial alarm processing.