Computational Chemical Engineering on a Dedicated Beowulf Cluster

摘要：David S. Sholl, Lorenz T. Biegler and Steinar Hauan机构：卡内基梅隆大学提案号：0094407这是一项设备拨款，用于购买32节点的Beowulf集群，以支持卡内基梅隆大学三个研究小组使用的先进计算研究。Beowulf集群代表了高级计算的一种优势架构：它们的构建成本低廉，配置灵活，并为各种科学计算任务提供了强大的计算环境。共享集群的三个研究小组将进行分子动力学和计算化学、建模和可视化方面的研究，以支持过程合成和大规模离散和连续过程优化。因此，集群将服务于需要在几分钟和几小时的时间尺度上达到高峰性能的计算，以及在几天的时间尺度上保持持续的计算性能。这些互补的概要文件意味着，通过共享集群，它的功能将得到比任何单独的组更充分的利用。集群还可以很容易地在不同的使用模式之间分配资源，而不会产生显著的计算开销。该集群最初将把pi的计算能力提高约三倍，并将在接下来的几年中扩展，并将在未来五年内将其计算能力提高一个数量级以上，而不会使现有节点过时。具体来说，这三个小组的研究包括开发新的算法：计算化学中的计算，特别是分子动力学和蒙特卡罗模拟，从原子尺度结构的细节确定宏观材料特性；工艺设计和合成的精确工艺建模，包括相平衡、工艺替代方案的表示和评估以及设计见解的可视化；涉及离散和连续决策以及先进分解策略的大规模稳态和动态过程的优化。

ABSTRACTPI: David S. Sholl, Lorenz T. Biegler and Steinar Hauan Institution: Carnegie Mellon UniversityProposal Number: 0094407This is an equipment grant to provide funds to purchase a 32 node Beowulf cluster to support research in advanced computing to be used by three research groups at Carnegie Mellon University. Beowulf clusters represent an advantageous architecture for advanced computing: they are inexpensive to construct, flexible to configure, and provide a powerful computing environment for a broad variety of scientific computing tasks. The three research groups that will share the cluster will be conducting research on molecular dynamics and computational chemistry, modeling and visualization to support process synthesis and large-scale discrete and continuous process optimization. The cluster will thus serve computations that require high peak performance on time scales of minutes and hours, along with sustained computational performance on time scales of days. These complementary profiles mean that by sharing the cluster, its capabilities will be used more fully than they would be by any group individually. The cluster can also easily be operated in a way that distributes resources between different usage modes without significant computational overhead. The cluster will initially increase the PIs' computational capabilities by about a factor of three and will be expandable over succeeding years and will increase their computing power by over an order of magnitude within the next five years, without rendering existing nodes obsolete.Specifically, the research of the three groups consists of the development of new algorithms for: Calculation in computational chemistry, particularly in molecular dynamics and Monte Carlo simulations to determine macroscopic material properties from details of atomic-scale structure;Accurate process modeling for process design and synthesis that includes phase equilibrium, representation and evaluation of process alternatives and visualization of design insights; andOptimization of large-scale steady state and dynamic processes that involve both discrete and continuous decisions along with advanced decomposition strategies.