Many-Task Computing Tools for Multiscale Modeling

As computers have become more powerful, the simulations and data processing applications that use them have become resource hungry, and, at the same time more complex. The complexity of simulations has increased in terms of the number of dimensions (from 1D to 2D to 3D), the set of equations being simulated (from one equation, to multiple equations in one domain, to multiple equations in multiple domains), the number of time, length, or other scales being studied simultaneously, and the level of interaction between scales. In multiscale modeling, several applications are frequently assembled to perform increasingly complex and integrated multiscale analyses. Increasing the scientific insights and productivity from such multiscale methods often demands increasing the number of individual applications that are assembled into a meta-application. This can be achieved by adding another layer of procedure around the individual applications, as is done in parameter sweep, optimization, and uncertainty quantification methods. Such meta-applications can be considered “many-task computing” (MTC) applications, as they are assembled of a series of tasks, each of which may be as complex as a standalone application or simple as a procedure call.