Collaborative Research: Efficient Detection and Alleviation of Scalability Problems

本研究提案的重点是开发工具支持，为科学程序员提供查询可扩展性问题并将此信息与源代码关联的能力。此外，我们认为工具应该能够建议和评估优化转换，以减轻这些问题。这将构成对当前性能分析实践的重大改进。其关键的智力价值在于提供了一个自动化框架，用于检测可伸缩性问题并将其与源代码关联起来。我们将在ASCI代码上试验我们的框架，这是为了强调高性能集群。首先，这两个PI正在努力创建一个跨学科的教育和研究计划。第二，学生将接受高性能计算的教育。最后，拟议的工作允许技术转移到一个新兴领域的广阔竞技场，如集群计算以及已建立的领域的SMP和大规模并行计算。开发的框架和工具将普遍提供给研究社区和高性能计算实验室。

The focus of this research proposal is to develop tool support to provide the ability for scientific programmers to inquire about scalability problems and correlate this information back to source code. Furthermore, we believe that tools should be able to suggest and evaluate optimizing transformations to alleviate these problems. This would constitute a significant improvement over current performance analysis practice.The key intellectual merit is in providing an automatic framework for detecting scalability problems andcorrelating them back to source code. We will experiment with our framework on the ASCI codes, whichis intended to stress high-performance clusters.The broader impact of this work is in three main areas. First, both PIs are working to create an interdisciplinary educational and research program. Second, students will be educated in high-performancecomputing. Finally, the proposed work allows for technology transfer to a wide arena of emerging fields,such as cluster computing as well as the established areas of SMPs and massively parallel computing.The developed framework and tools will be made generally available to the research community and high-performance computing labs.