LeanMD: A Charm++ Framework for High Performance Molecular Dynamics Simulation on Large Parallel Machines
Publication Type: MS Thesis
Molecular dynamics programs simulate the behavior of biomolecular systems, leading to insights and understanding of their functions. However, the computational complexity of such simulations is enormous. Parallel machines provide the potential to meet this computational challenge. To harness this potential, it is necessary to develop a program that can scale well on large parallel machines. Application domain programmers should be able to easily reuse the parallel program with minimal modifications to integrate their science routines and test their motivations. This thesis presents leanMD, a parallel molecular dynamics simulation framework written in Charm++ for PetaFLOP class supercomputers. LeanMD is designed to be scalable to large parallel machines (with tens of thousands of processors). LeanMD uses fine-grained spatial decomposition combined with force decomposition to enhance its scalability. The computation is modeled using a large number of virtual processors, which are mapped flexibly to available processors with assistance from the Charm++ runtime system. Charm++ allows the use of ``parallel" libraries to facilitate common operations such as 3-D FFTs. Charm++ also provides libraries for communication optimizations and has built in support for automatic load-balancing.
Vikas Mehta, "LeanMD: A Charm++ framework for high performance molecular dynamics simulation on large parallel machines", University of Illinois at Urbana-Champaign, 2004.