ATLAS

Abstract

A majority of the data generated in several domains is geotagged. These data also have a chronological component associated with them. Pervasive data generation and collection efforts have led to an increase in data volumes. These data hold the potential to unlock valuable insights. To facilitate such knowledge extraction in a timely manner, the underlying file system must satisfy several objectives. In this study, we present Atlas, a distributed file system designed specifically for spatiotemporal data. Atlas includes several capabilities that are suited for performing large-scale analyses: aligning dispersion with data access patterns, load balancing storage, and facilitating interoperation with analytical engines such as Hadoop and Spark. Our empirical benchmarks profile several aspects of Atlas, and demonstrate the suitability of our methodology.