Methodology and optimization for implementing cluster-based parallel geospatial algorithms with a case study
- Additional Document Info
- View All
© 2019, Springer Science+Business Media, LLC, part of Springer Nature. Cluster-based parallel computing technology has been widely used in the geosciences. However, how to implement the corresponding parallel algorithm in a simple way, and how to make parallel algorithms more efficient and effective, are still of great value to this research area, especially to beginners who are new to parallel computing. In this research, a contour line generation algorithm is paralleled with a message passing interface-based parallel computing as a case study to illustrate the improvement and optimization methods for the aforementioned problems to high performance geo-computation newcomers. Through experiments it can be concluded that: (1) In order to implement parallel algorithms in a simple way, we adopt the single program/multiple data mode, which is a method that evenly distributes tasks. We make the processes independent by reading the input files simultaneously, while writing results in parallel and gathering them with the master. (2) Even small hotspots should also be considered in the optimization procedure in order to get an efficient parallel algorithm. (3) Implementing a suitable parallel algorithm should consider many factors, such as the state of the network, I/O throughput, and also the desired application and user requirements.
author list (cited authors)
Huang, F., Tie, B. o., Tao, J., Tan, X., & Ma, Y.