Center for Petascale Computing
The parallel software architecture of NAMD was developed
over ten years ago, and its scalable design continues to serve the biomedical research community.
However, the needs of the scientific driving projects for petascale computing require NAMD to
run much faster simulations of moderately-sized systems (350,000 atoms) while also being able to
run extremely large systems (70 million atoms) scaled effectively to large numbers of processors.
The technological improvements and optimizations to be developed will include the parallel scaling
of long-range electrostatics; the optimization of NAMD's memory usage, the implementation of
support for parallel input and output, the development of hierarchical and topology-aware load
balancing support, the implementation of advanced fault tolerance support, and the improvement
of NAMD's usage of multiprocessor nodes. This latter development will exploit recently developed
Charm++ features that are only available to applications using a new parallel framework, to which
NAMD will be adapted.
Investigators: Laxmikant Kale, Klaus Schulten, and Robert Skeel