A theoretical approach for using cyberinfrastructure (Wang & Armstrong, 2009)

In this paper, Wang and Armstrong (2009) propose a theoretical framework for solving the problems, which are about logic, generality, compatibility, in the GIScience practices of spatial domain decomposition and task scheduling. There are some correlations between these problems.

Wang and Armstrong firstly argue that we should move the focus from spatial data and its operations to computational intensity. It seems not acceptable for geographer to give up concerning about spatial characteristics of data and exploring methods to deal with them. However, in a higher level, it is necessary to remove the particularity of spatial data before domain decomposition and task scheduling. That said, transformations from spatial data to computational intensity are handled in a lower level. Otherwise, it is not able to have generic parallel computing solutions for geographic analysis, since domain decomposition will be changed as spatial characteristics change. Besides, the particularity of spatial data make its analysis rely on specific parallel computer architectures, which restricts the adaptability of solutions. Now that, I think the most critical issue is how to have a “good” computational domain representation. No matter object-based or field-based, we must lose some information when representing the physical world by a list of data. When we put grids on it, when we calculate computational intensity, or when we look for the homogeneity in quadtrees, there may be discords between the final representation and the ground truth. Even with the consideration of granularity, I still think this problem is worth further exploring. In section 6, Wang and Armstrong note that the framework is based on region quadtrees, which is a recognized way to deal with 2D surface, referring to a compu-band in this paper. However, quadtree can become quite unbalanced when the data are unevenly distributed. A lot of grids can be blank in certain situations, which can potentially waste computing and memory resources. There should be some ways to address this concern.


Comments are closed.