Modeling Parallel Molecular Simulations on Amazon EC2
Xiangqiang Xu, Gabriel Dunham, Xinghui Zhao, David Chiu, Jie Xu · 2015
Cloud computing has been widely used by computational scientists and engineers as a means to run large-scale simulations while circumventing capital investment of hardware. However, a challenging problem is to accurately estimate how much cloud resources that a specific computation requires, in order to execute computations in a cost-effective way. In this paper, we use a real-world molecular dynamics (MD) simulation as a motivating scenario and present our work in modeling parallel execution of such a computation on the cloud. Our model estimates the workload of an MD simulation at fine-grained detail, and based on that estimate, calculates the time required to run the simulation. The accuracy of the model has been evaluated using various types of MD simulations on different scales.