Distributed Viewshed Analysis An Evaluation of Distribution Frameworks for Geospatial Information Systems

Abstract

Viewshed analysis is the process of computing what areas of a terrain are visible from a certain observation point. In this thesis we evaluated the performance of these computations on cloud clusters using the distribution framework Apache Spark. We implemented three commonly used viewshed algorithms; R3 which is slow but highly accurate as well as R2 and van Kreveld which are faster but less accurate. Two versions of each algorithm were implemented, one to run on a single multi-core machine and one to run on a server cluster using Spark. We compared the accuracy and running time of the different algorithms in order to determine when to use the different algorithms. Our results show that viewshed analysis does not perform well when implemented using Spark if real-time results are required. In fact the faster algorithms performed consistently worse on the cluster, even for very large input data. For the highly accurate, but slow, R3 algorithm we were able to achieve a 1.6x speedup using the distribution framework.