June 22–26, 2014
Leipzig, Germany

Presentation Details

Name: (03a) Implementing a Hybrid Parallel Overset Grid Algorithm for Computational Fluid Dynamics Applications
Time: Thursday, June 26, 2014
10:30 am - 11:00 am
Room:   Hall 4
CCL - Congress Center Leipzig
Breaks:10:30 am - 11:00 am Coffee Break
07:30 am - 10:30 am Welcome Coffee
Presenter:   Dominic Chandar, A*STAR
Abstract:   Simulating complex motions of bodies for fluid dynamics applications is still a challenge from the perspective of grid generation. One of the popular methods that is used for moving body computations is the overset grid method. In this method, component grids (structured or unstructured) are generated for each body under investigation and they are made to overlap a common background grid. The benefit of this approach is that the entire grid doesn’t have to be regenerated or deformed when there is a relative motion between the bodies. Also, overset grids can be useful for Adaptive Mesh Refinement (AMR) where generation of finer grids can be achieved in specific regions of interest. The objective of the overset grid algorithm is to classify which points on a given grid solve using the native P.D.E solver, and which points obtain the solution from the grid that it overlaps with. The entire process involves complex search algorithms, intersection checks, and setting up a communication table to be used with MPI. While this procedure is quite well known and easy to implement in a serial environment, it is still not entirely robust in a parallel environment due to the manner in which individual grids are partitioned using METIS. This work is essentially an extension of an earlier work with an implementation within an in-house fluid dynamics flow solver with additional multi-threading features. Apart from testing out the code on in-house (A*CRC) clus-ters, tests were also conducted using the Amazon Web Services (AWS) cloud. Online cluster creation on the AWS cloud was aided by starcluster – a tool to automatically configure AWS for several instances.

Dominic Chandar & Vinh Tan Nguyen, Institute of High Performance Computing, A*STAR