Parallelization of Geospatial Models
I wish Microsoft would provide more info on the Windows Azure Fabric Controller. This is the technology that allows a large cluster of computers to be harnessed by an application. In addition to documentation, don’t think they’ve done a very good job selling the benefits of parallelization. Here are three spatial modeling areas where I think parallel processing would make life easier. These are the sorts of models that you typically run over night or over the weekend. Before the geospatial community dismisses the cloud as hype, I think we really need to look at the potential benefits.
Trip Assigment is used by planners wishing to analyze how changes to a street network impact traffic flows. Here is a paper mentioning of how parallelization can benefit this activity.
“The extraordinary time savings of this method was limited only by the amount of hardware available and the granularity of the model steps. In practice, full runs shortened from 36 hours to 9.”
Non-steady state river flow modeling takes a lot of CPU. Looks like DHI has parallelized MIKE 21.
“Particular noteworthy is the introduction of parallel processing in the FM series of MIKE 21.”
When parallelization works, adding CPUs becomes cost effective. Intel realizes this and has sponsored some interesting work in using Hadoop for ground motion modeling. They were able to reduce the model runs from 36 to two hours.