TREST: A Hadoop Based Distributed Mobile Trajectory Retrieval System

Nowadays, mobile phones have prevailed in a great variety of applications, through which massive personal trajectories can be collected and analyzed to support many interesting location-based services. However, it is still challenging to efficiently store and retrieve this kind of spatio-temporal data, which has typical big-data feature with large size, streaming style and multiple data source. To address this issue, we develop a mobile trajectory retrieval system named TREST, which is based on the distributed Hadoop and HBase systems. TREST makes use of the horizontal expansion mechanism of Hadoop to store overwhelming spatio-temporal trajectories, and supports frequent incremental insertion of data stream. Meanwhile, TREST maps the spatio-temporal features of trajectories into the simple key-value schema of HBase to support fast retrieval. We also develop a prototype of TREST to manipulate the real mobile trajectory data set, which contains totally 104 million records collected by mobile service providers. Experiments on this data set show that TREST can efficiently support both Single-track and All-track retrieval within milliseconds on average.