Attributes of Molecular Dynamics Calculations: Accounting for CPU Cycles

Molecular dynamics is a widely used simulation method, which, together with electronic structure methods, quantum dynamics, and Brownian dynamics, plays an important role in modeling processes at the atomic level. Molecular dynamics is also a widely studied method. To model systems more accurately and for longer time scales with improved turn{around time, scientists continually seek eecient algorithms to utilize fully modern computing resources. Because it is a computationally challenging application of importance, molecular dynamics is often used to evaluate computing systems, placing it in benchmark suites such as SPEC (System Performance and Evaluation Cooperative) and SPLASH 28]. As a benchmark, MD is often used alongside other applications to evaluate computer A common application of a benchmark uses total time to ascertain the eecacy of an algorithm or computing system. The details about the benchmark that should be considered will vary according to the goals of the study and what is being measured. To compare two parallel molecular dynamics algorithms, for example, one should have a handle on the relevant input parameters and their eeect on the total instruction count. Similarly, to evaluate the speedup of a parallel molecular dynamics algorithm requires some knowledge of the simulation beyond the number of atoms. Other parameters, such as pairlist update frequency and cutoo radius can have a dramatic eeect on the workload 5, 7]. The results of computing system evaluations can vary with a benchmark's input parameters to varying degrees. For example, we have found in our study on the Cray C90 that the percentage of branch statements executed in a representative simulation of a protein in water ranges roughly from 5% to 10% (Section 4). On the other hand, not constraining bonds in the solute was relatively unimportant for the properties we examined. Our goal in this paper is to highlight a number of the key issues responsible for diversity among molecular dynamics calculations. Attention to some of these details can be important when evaluating a molecular dynamics algorithm or when using the MD application to evaluate a computing system. Through our analysis, we show where processor cycles are consumed in this oating point intensive calculation, and how performance can be degraded by complexities arising in simulating complex molecular systems. 1 The Molecular dynamics calculation Molecular dynamics simulates the motion of a collection of atoms using Newton's equations of motion 20]. A molecular dynamics simulation is considered a chaotic process. In other …