Z. Cui and Z. Lan (USA)
Performance evaluation, High performance computing, Cluster computing, Message passing, Molecular dynamics
One of the most exciting and difficult challenges in biology area is to understand the interactions of complex biological systems. Computational simulations have become increasingly important in enabling progress in biological research. GROMACS is such a software package that provides a variety of tools to enable biological simulations. In this paper, we investigate and analyze the performance of GROMACS in two widely used cluster environments: a Linux cluster and a Unix cluster. The performance is studied from three aspects: overall performance, communication characteristics, and load balancing characteristics. Our analysis shows that GROMACS suffers from three major performance issues: (1) it cannot scale well beyond 16 processors; (2) the underlying communication scheme of GROMACS is not efficient; and (3) it does not include any dynamic load balancing scheme for efficient execution on parallel and distributed systems. Besides analyzing the application and locating the bottlenecks, we also provide several candidate solutions to improve the performance of GROMACS on large-scale cluster systems.
Important Links:
Go Back
