CHARMM (Chemistry at HARvard Macromolecular Mechanics) is a widely recognized scientific code for molecular dynamics and mechanics in computational biology. The code achieved good performance on many computer platforms in the recent past and was a considerable success in many areas of biology, in particular in research on prion infections. In particular the code runs well on the aging massively parallel multiprocessors manufactured by Cray Research Inc. in the mid 1990ies (T3D and T3E) with more than 100 processors and on low cost clusters of PCs with up to 16 processors.
Nowadays, the general trend of computing research leads away from the expensive supercomputers to cheaper clusters and ultimately to so called computational grid computing architecture, where every owner of PC can produce or consume compute performance in a way similar to a (still somewhat) visionary electricity production and distribution based on environmentally suitable micro power stations.
In this project, we will show the adaptation of the existing computational biology code CHARMM to the environment of widely distributed computing and the investigation of the performance and efficiency of such a migration. In particular, we will go through some critical issues such as the scheduling, the fault tolerance and the communication problems related to the migration, proposing some interesting solutions. For the proof of feasibility, we aim at a realistic setting using commodity hardware and well-known software components. Since the idea of widely distributed computing is quite well established, we rely on existing middleware to do the distribution of the computation. Therefore, we teamed up with "United Devices Inc." and are using their software to investigate feasibility of widely distributed molecular simulations.