Acquisition of a Beowulf Class Cluster Computer to Support Interdisciplinary Faculty Research and Graduate Student Training Using Numerically Intensive Statistical Modeling Methods

在国家科学基金会重大研究仪器奖的支持下，锡拉丘兹大学麦克斯韦尔学院政策研究中心(CPR)的高级研究员J.S.拉辛教授将在CPR中建立一台32-双处理器节点的Beowulf级集群计算机，以支持使用数值密集统计建模方法的教师研究和研究生培训。Beowulf级集群计算机将使用现成的硬件组件、开源软件和相当程度的平台独立性，以相对较低的成本提供超级计算机计算能力。J·S·拉辛教授将在CPR计算机顾问的协助下监督Beowulf级集群计算机的购买、安装和操作，该顾问在Beowulf集群计算机方面的培训将得到该奖项的支持。拉辛预计，最初将有9名教职员工和20至30名研究生从这台计算机的购买和实施中受益，随着其他Maxwell School研究人员开发项目来开发Beowulf集群计算机的能力，这些数字将会增长。应用统计研究正在迅速发展，从基于参数模型和推理的范例，依赖于渐近近似，到基于非参数模型和基于重采样的推理。这些信息丰富的大型数据集进行非参数分析的时机已经成熟，但它们的计算负担几乎阻碍了它们在这类数据集上的应用。这也适用于有望改进参数建模的方法。这些计算障碍可以通过使用并行计算环境来克服，其中最实用的形式是Beowulf类集群计算机。CPR至少有9名教员，他们要么使用数值密集型计量经济学方法并在这一领域拥有专业知识，要么经常与大数据集建模的计算方面作斗争。有了这台集群计算机，我们不仅可以教他们这些方法的理论，还可以教他们如何将它们应用到大型的真实世界数据库中，比如他们毕业后将使用的数据库。广泛的影响：锡拉丘兹大学的麦克斯韦尔公民与公共事务学院，政策研究中心所在的地方，进行高质量的社会科学研究，培训研究生担任学术和研究职位，以及世界各地负责任的政府和非政府机构职位。如果该项目获得资助，研究人员将能够利用该集群将尖端统计方法应用于我们中心保存的大量数据集，所有这些都将导致改进政策分析和增强对经验现象的理解。最后，当然，在集群上开发的软件将公开提供，从而最大限度地发挥投资的更广泛影响。我们非常兴奋地看到，在即将到来的一年里，这一资源可供教职员工和学生使用。简而言之，我们将能够将最新的数值密集统计方法应用于目前仍无法计算的丰富数据集。

With support from a National Science Foundation Major Research Instrumentation award, Professor J.S. Racine, Senior Research Associate in the Center for Policy Research (CPR) of the Maxwell School at Syracuse University, will establish a 32-dual-processor-node Beowulf class cluster computer in CPR to support faculty research and graduate student training using numerically intensive statistical modeling methods. The Beowulf class cluster computer will provide supercomputer computing power at a relatively low cost using off-the-shelf hardware components, open source software, and a substantial degree of platform independence. Professor J. S. Racine will supervise the purchase, installation, and operation of the Beowulf class cluster computer, assisted by the CPR computer consultant, whose training in Beowulf cluster computers will be supported by this award. Racine anticipates that initially 9 faculty members and 20 to 30 graduate students will benefit from the purchase and implementation of this computer, and that these numbers will grow as other Maxwell School research faculty develop projects to exploit the Beowulf cluster computer's capabilities.Applied statistical research is rapidly evolving from a paradigm based upon parametric models coupled with inference, which relies on asymptotic approximations, to one based on nonparametric models and inference based on resampling. These large information-rich datasets are ripe for nonparametric analysis, but their computational burden all but prevents their application for such datasets. This also applies to methods that hold the promise for improved parametric modeling. These computational obstacles can be overcome by using a parallel computing environment, the most practical form of which is a Beowulf class cluster computer.CPR has at least nine faculty who either work with numerically intensive econometric methods and who have expertise in this area, or who routinely struggle with computational aspects of modeling large datasets. With this cluster computer, we will be able to teach them not only the theory of these methods but how to apply them to large, real world databases such as those they will work with after graduation.Broader Impacts: The Maxwell School of Citizenship and Public Affairs at Syracuse University, in which the Center for Policy Research is located, conducts high-quality social science research and trains graduate students for academic and research positions, as well as responsible government and non-governmental agency positions around the world. Should the project be funded, researchers will be able to leverage the cluster to apply cutting-edge statistical methods to numerous datasets housed by our center, all of which will result in improved policy analysis and enhanced understanding of empirical phenomena. Finally, software developed on the cluster will, of course, be made publicly available, thereby maximizing the broader impacts of the investment.We are extremely excited to have this resource available for use by both faculty and students in the upcoming year. Simply put, we will be able to apply the latest numerically intensive statistical methods to rich datasets which currently remain beyond our computational reach.