Understanding Fluid Flow in Microchannels using a CometCloud-based Federated HPC infrastructure

The ability to control fluid streams at microscale has significant applications in many domains, including biological processing, guiding chemical reactions, and creating structured materials, to name just a few. Recently, it has been discovered that placing pillars of different dimensions, and at different offsets, allows "sculpting" the fluid flow in microchannels. The design and placement of sequences of pillars allows a phenomenal degree of flexibility to program the flow. However, to achieve such a control it is necessary to understand how flow is affected by different input parameters.
The problem is challenging for several reasons. The search space consists of tens of thousands of points, and an individual simulation may take hundreds of core-hours, even when executed on a state-of-the-art HPC cluster. The individual simulations, although independent, are highly heterogeneous and their cost is very difficult to estimate a priori, owing to varying resolution and mesh density required for different configurations. Finally, because the non-linear solver is iterative, it may fail to converge for some combinations of input parameters, which necessitates fault-tolerance.