Personal computer, workstation, server, and networking hardware continues to see declines in its price to performance ratio, meaning that commodity clusters today are able to achieve performance achievable only by expensive, custom designed supercomputers just a few years ago. This allows researchers to develop more complex and powerful algorithms which allow for richer simulations of biochemical processes. In particular, more accurate theoretical models can be used and simulations can be longer at a given level of theory, providing direct access to events that previously required too much computer time to simulate accurately. The lab has taken advantage of the cheap, powerful commodity hardware by deploying Intel Nehalem based compute nodes as part of the LoBoS cluster., which has allowed for a substantial gain in computing power. Additionally, these nodes will be upgraded to use faster networking hardware (QDR rather than DDR InfiniBand). CHARMM has been converted from Fortran 77 to Fortran 95 as part of a broader effort to modernize the program. This modernization of the code will allow for easier integration of new features into the program and provide for an overall better development experience. Current work being performed in this lab includes enhancement of the MSCALE code to interface with other simulation programs and adding support GROMACS style angle parameters and cut-offs, which allows for compatibility with coarse-grained molecular models such as MARTINI. The CHARMMing program is under active development. New features added within the last year include the ability to build coarse-grain models and to calculate oxidation/reduction potentials of Iron/Sulfur clusters embedded in proteins. Additional testing and validation of the software has been accomplished and the structure of the code has been improved. Development of techniques and software that will allow CHARMM to run on XMT eXplicit Multi-Therading (XMT) is a general purpose parallel architecture aimed at reducing single task completion time. XMT provides good performance for Parallel Random Access Model (PRAM)-like algorithms, allowing easy programming. We are developing techniques and software so that CHARMM will be used with XMT.
Our research aims to port key concepts of molecular simulation to XMT, including fast Fourier transform, eigenspace operations and short range n-body problem. Our FFT implementation reveals significant performance gaps between XMT and state-of-the-art processors. Future research will be based on implementation of other CHARMM modules, and also realizing molecular simulation sub-programs on XMT. The long-term goal is to completely port CHARMM to XMT.
Showing the most recent 10 out of 15 publications
