GPU-Enabled Visual Analytics Framework for Big Transportation Datasets

Transportation agencies rely on a variety of data sources for condition monitoring of their assets and making critical decisions such as infrastructure investments and project prioritization. Recent exponential increase in the volumes of these datasets has been causing significant information overload problems for data analysts; data curation process has increasingly become time consuming as legacy CPU-based systems are reaching their limits for processing and visualizing relevant trends in these massive datasets. There is a need for new tools that can consume these new datasets and provide analytics at rates resonant with the speed of human thought. The current paper proposes a new framework that allows for both multidimensional visualization and analytics to be carried seamlessly on large transportation datasets. The framework stores data in a massively parallel database and leverages the immense computational power available in graphical processing units (GPUs) to carry out data analytics and rendering on the fly via a Structured Query Language which interacts with the underlying GPU database. A front-end is designed for near-instant rendering of queried results on simple charts and maps to enable decision makers to drill down insights quickly. The framework is used to develop applications for analyzing big transportation datasets with over 100 million rows. Performance benchmarking experiments conducted showed that the methodology developed is able to provide real-time visual updates for big data in less than 100 ms. The performance of the developed framework was also compared with CPU-based visual analytics platforms such as Tableau and D3.