A Distributed and Scalable Framework for Low-Latency Continuous Trajectory Stream Processing

In recent times, the widespread adoption of GPS-enabled devices has resulted in an exponential surge in the amount of spatio-temporal data streams. These data streams, often referred to as trajectory data streams, represent a sequence of spatial locations of mobile objects. As these streams continue to grow, so does the demand for its efficient processing and analysis. Various government agencies and businesses now require real-time or low-latency solutions for continuous query processing and interactive analysis of trajectory streams, driven by applications like real-time route guidance during disasters, tracking critically ill patients to prevent mishaps, ongoing road traffic management to avert accidents and congestion, and more. Trajectory streams, due to their volume and velocity, necessitate scalable Stream Processing Engines (SPEs) for effective handling. However, currently available state-of-the-art scalable spatial SPEs do not adequately support low-latency and continuous processing of trajectories. Conversely, while scalable SPEs do exist, they lack the necessary data structures, operators, and indices essential for processing trajectory streams. To bridge this gap, our work introduces TStream, a distributed, low-latency, and continuous trajectory stream processing system. TStream addresses this challenge by offering support for spatial data types, spatial indices, and spatial continuous queries such as range, nearest neighbor searches (kNN), and joins over spatial trajectory streams. To substantiate the efficacy of TStream, we present an extensive experimental study conducted on two real-world trajectory datasets.