|
|
|
Last week, we continued our discussion of ecosystem simulations
by looking at a research project which incorporated many of the
staples of modern scientific computing. This week, we continue
in the same vein by examining studies being conducted on
ecological systems using massively parallel computing techniques
at Charles Sturt University (CSU) in Australia. As you will see,
this ambitious project provides us with a lot of juicy food for
thought.
The ecosystem research at CSU is based on Geographic Information Systems (GIS) databases, which provide spatially important data ranging from local geographic systems to satellite images. Because of the wealth of data available, GIS databases have long been the target of simulation scientists, but they provide no temporal information to the researcher. As well, to adequately model a section of terrain requires more data manipulation than traditional software/hardware combinations can handle. Because of these limitations some additional pieces had to be constructed before the puzzle could be finished. In particular, the CSU researchers recognized the need for efficient discrete event cellular modeling. This basically consists of dividing the system into a set of localized "boxes," each with its own local interactions, as well as interaction terms among the boxes. As you might imagine, via creative implementation, this scheme lends itself well to parallelization. Indeed, it has been suggested that each "box" can be assigned to its own processing element, giving hope for great parallel speed-up. The targeted machine for this study, the Connection Machine 5, employs a data parallel architecture which is especially well-suited to this type of application. In addition to the GIS data bases and the parallel cellular modeling software, the researchers at CSU recognize the necessity of a scientific visualization tool to make the methods and results more accessible. They envision an interactive visualizer that will allow the user to see the results of a simulation as the calculation finishes. Then he will be able to adjust parameters and determine the effect of the changes. In addition to providing valuable insight into the physical nature of the system, such a tool will also allow the scientist to determine an acceptable ratio of accuracy-to-performance as the simulation cell sizes are adjusted. The CSU web site presents a concrete example of this software/ hardware scheme in action, as well as providing deeper discussion of the ideas touched on here. It's definitely worth visiting at: Go To Page: 1
The copyright of the article Massively Parallel Ecosystem Simulations in Scientific Computing is owned by . Permission to republish Massively Parallel Ecosystem Simulations in print or online must be granted by the author in writing.
|
|
|
|
|
|
|
|