Creating Commercial Parallel Statistical Software

描述（由申请人提供）：统计复杂性的增长 方法以及这些方法在不断增长的数据集上的应用， 研究人员已经超过了单处理器计算能力的增长 电脑在生物医学竞技场中，这一点最为真实， 复杂的估计和模拟模型已经激增。 由于服务器市场的原因，多处理器计算机被广泛使用， 但是在这样的计算机上运行的并行统计软件 市售。我们建议扩展Stata，一个统计系统广泛 被研究人员用来开发多处理器计算机。的目标 该项目旨在通过以下方式促进对复杂统计模型的估计 利用多处理器计算机的优势。 我们的初步分析表明，2处理器计算机上的Stata可以 预计执行统计分析的速度约为 单处理器计算机; 4处理器计算机，3.0倍快; 16 处理器计算机，速度是6.4倍。在第一阶段项目中， 将修改Windows和Stata for Linux，并对 可以并行化的Stata部分和观察到的性能 改进.具体的评价标准将适用于这些 测量. 拟定商业应用： 该项目将为商用统计软件包Stata的并行版本创建一个概念验证版本。如果这个概念验证版本的性能符合我们的预期，我们将在随后的第二阶段项目中生产一个商业分发版本。目前还没有现成的商业统计软件包可用于多处理器机器的并行化。一个商业上可用的并行Stata将服务于这个不断增长的市场。

DESCRIPTION (provided by applicant): The growth in complexity of statistical methods and the application of those methods to ever-growing datasets by researchers has outpaced the growth in the computing power of uniprocessor computers. Nowhere is this more true than in the biomedical arena, where complex estimation and simulation models have proliferated. Multiprocessor computers are widely available owing to the market for servers, but parallelized statistical software to run on such computers is not commercially available. We propose to extend Stata, a statistical system widely used by researchers, to exploit multiprocessor computers. The goal of the project is to facilitate the estimation of complex statistical models by exploiting the advantages of multiprocessor computers. Our preliminary analysis suggests that Stata on a 2-processor computer can be expected to execute statistical analysis about 1.8 times as fast as a single-processor computer; a 4-processor computer, 3.0 times as fast; a 16 Processor computer, 6.4 times as fast. In the Phase I project, both Stata for Windows and Stata for Linux will be modified and measurements made of the fraction of Stata that can be parallelized and the observed performance improvements. Specific evaluation criteria will then be applied to these measurements. PROPOSED COMMERCIAL APPLICATION: This project will create a proof-of-concept version of a parallelized edition of the commercially available statistical package Stata. If this proof-of-concept version performs as we expect, we will produce a version for commercial distribution in a subsequent Phase II project. There is no commercial, off-the-shelf statistical package that is parallelized for multiprocessor machines. A commercially available parallelized Stata would serve this growing market.