Parallelization solutions for the YNANO Discontinua Simulations.

Abstract

In the context of constant and fast progresses in nano technology, discontinua based computation simulations are becoming increasingly important, especially in the context of virtual experimentations. The efficiency of discontinua based nanoscale simulations are still limited by CPU capacity (the number of simulation particles in the system). It is accepted that parallelization will play an important role in solving this problem. In this thesis, two parallelization approaches have been undertaken to parallelize the YNANO discontinua simulations. The scope of the work includes parallelization of the YNANO using the shared-memory approach OpenMP and the distributed-memory approach MPI, and also includes a novel MR_PB linear contact detection algorithm which can be used under periodic boundary conditions. The developed MPI parallelization solutions are compatible with the MR linear contact detection algorithm used in the sequential YNANO, the developed solutions preserves the linearity of both MR_Sort and MR_Search algorithm. The overall performance and scalability of the parallelization has been studied using nanoscale simulations in fluid dynamics and aerodynamics.