A distributed workload-aware approach to partitioning geospatial big data for cybergis analytics

Numerous applications and scientific domains have contributed to tremendous growth of geospatial data during the past several decades. To resolve the volume and velocity of such big data, distributed system approaches have been extensively studied to partition data for scalable analytics and associated applications. However, previous work on partitioning large geospatial data focuses on bulk-ingestion and static partitioning, hence is unable to handle dynamic variability in both data and computation that are particularly common for streaming data. To eliminate this limitation, this thesis holistically addresses computational intensity and dynamic data workload to achieve optimal data partitioning for scalable geospatial applications. Specifically, novel data partitioning algorithms have been developed to support scalable geospatial and temporal data management with new data models designed to represent dynamic data workload. Optimal partitions are realized by formulating a fine-grain spatial optimization problem that is solved using an evolutionary algorithm with spatially explicit operations. As an overarching approach to integrating the algorithms, data models and spatial optimization problem solving, GeoBalance is established as a workload-aware framework for supporting scalable cyberGIS (i.e. geographic information science and systems based on advanced cyberinfrastructure) analytics