Association for Computing Machinery (ACM)
PEARC '19: Proceedings of the Practice and Experience in Advanced Research Computing (July 28 – August 1, 2019)
We are in the era of Spatial Big Data. Due to the developments of topographic techniques, clear satellite imagery, and various means for collecting information, geospatial datasets are growing in volume, complexity and heterogeneity. For example, OpenStreetMap data for the whole world is about 1 TB and NASA world climate datasets are about 17 TB. Spatial data volume and variety makes spatial computations both data-intensive and compute-intensive. Due to the irregular distribution of spatial data, domain decomposition becomes challenging. In this work, we present spatial data partitioning technique that takes into account spatial join cost. In addition, we present spatial join computation using Asynchronous Dynamic Load Balancing (ADLB) library. ADLB is a software library designed to help rapidly build scalable parallel programs using MPI. We evaluated the performance of ADLB-based MPI-GIS implementation. In our existing work, spatial data movement cost from ADLB server to worker MPI processes limited the scalability of MPI-GIS.
Yang, Jie; Paudel, Anmol; and Puri, Satish, "Spatial Data Decomposition and Load Balancing on HPC Platforms" (2019). Computer Science Faculty Research and Publications. 40.
ADA Accessible Version
Accepted version. Published as part of the proceedings of the conference, PEARC '19: Proceedings of the Practice and Experience in Advanced Research Computing (July 28 – August 1, 2019): 1-4. DOI. © 2019 Association for Computing Machinery (ACM). Used with permission.