Welcome to distributed computing on accelerated hardware

How does it work?

Each atom is represented using classical equations of motion (Netwon's equation) and evolves according to a force field which models the chemical nature of each atom (Carbon, Oxygen, Hydrogen and so on) in its local environment. Bridging molecular, atomistic scales (of the order of nanosecond) with biological scales (micro-milli seconds) is a big challenge in computational biology.

Molecular dynamics (MD) is a simulation methodology which enables, for instance, the study of the dynamics of proteins in their environment. It is used by pharmaceutical companies for a wide variety of applications, including drug design, drug screening and in general to investigate protein function.

The impact of molecular dynamics would be much greater if faster ways to perform MD simulations were found in order to reach the time scales of biological processes (micro-milli seconds). These time scales cannot be simulated yet, despite the use of costly high performance supercomputers with thousands of processors. Specialized hardware like the Cell processor and GPUs could help to approach this goal.

In all-atom molecular dynamics simulations proteins, lipid membrane, water molecules, ions, and so on are represented by all their atoms. This is the most common molecular dynamics simulation performed by scientists but also the most expensive. The advantage that all the molecular specificity of the system is taken into account (for instance the water around the protein is often very important) but the computing cost is so big that often simulations can be run only on large and expensive supercomputers. Indeed, often these simulations use hundreds of processors at once.

PS3GRID-GPUGRID

The Cell processor and GPUs make suddenly possible to run all atom molecular dynamics on a single machine with a good return time. The large number of PS3 and GPUs available makes possible to use different computational protocols to exploit parallel independent runs and compute the energetics of the system (a fundamental thermodynamic quantity).

BOINC users are running entire molecular structures representing part of a cell membrane for instance with over 30,000 to 100,000 atoms depending on the problem. The results collected from several runs are analyzed. In this sense, PS3GRID-GPUGRID is not a single application or single problem project, rather it is a novel computational tool made available to computational scientists. We are in fact collaborating with other academic research group. Each new application will be described on the website and the scientific outcome reported.



Gianni De Fabritiis