Parallel In-Memory Distance Threshold Queries on Trajectory Databases
Department of Information and Computer Sciences and NASA Astrobiology Institute, University of Hawai’i, Honolulu, HI, U.S.A.
The Sixth International Conference on Advances in Databases, Knowledge, and Data Applications (DBKDA 2014), 2014
@inproceedings{gowanlock2014parallel,
title={Parallel In-Memory Distance Threshold Queries on Trajectory Databases},
author={Gowanlock, Michael and Casanova, Henri and Schanzenbach, David},
booktitle={DBKDA 2014, The Sixth International Conference on Advances in Databases, Knowledge, and Data Applications},
pages={80–83},
year={2014}
}
Spatiotemporal databases are utilized in many applications to store the trajectories of moving objects. In this context, we focus on in-memory distance threshold queries that return all trajectories found within a distance d of a fixed or moving object over a time interval. We present performance results for a sequential query processing algorithm that uses an in-memory R-tree index, and we find that decreasing index resolution improves query response time. We then develop a simple multithreaded implementation and find that high parallel efficiency (78%-90%) can be achieved in a shared memory environment for a set of queries on a real-world dataset. Finally, we show that a GPGPU approach can achieve a speedup over 3.3 when compared to the multithreaded implementation. This speedup is obtained by abandoning the use of an index-tree altogether. This is an interesting result since index-trees have been the cornerstone of efficiently processing spatiotemporal queries.
April 22, 2014 by hgpu
