Parameter Sweep

Imagine that you have an executable file that can be called with options from the command line. The executable could contain a complex model from any discipline (e.g., finance, engineering, computational biology), and the options could be parameters passed to that model. In many disciplines, it is useful to run such models many times, with many different values for the various input parameters, and to combine the results to study the effects of those parameter changes. This type of analysis is called a "parameter sweep," and such analyses can be very demanding computationally. A natural way to break up this analysis is to define a work unit as a single run of the model, with a particular set of parameters. Without the Digipede Network, however, the process of farming out hundreds or thousands of model runs to different computers can be daunting.

The Digipede Network is particularly well suited for this type of analysis. From the Digipede Workbench, you can specify how to vary the various parameters through a single simple screen. You can pass different parameters to your model as literals, ranges, random variables, individual records in input files, or any combination of the above. When you submit your job, you don't need to worry about which computers will execute each model run - the Digipede Network handles that for you, taking full advantage of all available computing power on your network. You don't need to worry about moving input or output files; the Digipede Network uses intelligent data management, including versioning, distribution, and caching of datasets, to make sure the right information is available for each model run.