NGS: An Integrated Middleware and Language/Compiler Framework for Data Intensive Applications in a Grid Environment

EIA-0203846 Gagan Agrawal俄亥俄州立大学网格环境中数据密集型应用程序的集成中间件和语言/编译器框架我们建议开发方法，使数据密集型应用程序能够使用高级接口进行开发或指定，同时有效地利用网格资源。这将通过开发运行时技术、基于过滤器流编程的中间件以及用于将高级程序分解为一组过滤器的即时(JIT)编译技术来实现。我们的工具和技术将利用我们目标数据密集型应用程序的重要共性。我们的目标应用程序类别包括科学和商业数据密集型应用程序。我们正在开发一个名为DataCutter的中间件框架。DataCutter支持处理存储在广域网中的档案存储系统中的数据集。DataCutter将用于指定处理的接口导出为一组协调过滤器。虽然DataCutter提供了一个适合在网格环境中开发科学和商业数据密集型应用程序的框架，但它的编程接口相对较低。在这个项目中，我们提出了一系列高级语言接口，用于表示数据密集型应用程序。我们将数据并行Java、XML查询语言(XML Query Language，XQL)和挖掘操作符作为允许指定处理的接口，假设所有数据和计算周期在单个站点都可用。我们建议研究编译器技术，基于数据、计算、存储和网络资源的可用性，将这些应用程序分解为一组过滤器。

EIA-0203846 Gagan Agrawal Ohio State University An Integrated Middleware and Language/Compiler Framework for Data Intensive Applications in a Grid EnvironmentWe propose to develop methods that will enable data intensive applications to be developed or specified using high-level interfaces, and yet effectively utilize grid resources. This will be accomplished by developing runtime techniques, a middleware based upon filter-stream programming, and Just in Time (JIT) compilation techniques for decomposing high-level programs into a set of filters. Our tools and techniques will exploit important commonalities of the data intensive applications we target.Our target class of applications include both scientific and commercial data intensive applications. We are developing a middleware framework called DataCutter. DataCutter provides support for processing of datasets stored in archival storage systems in a wide-area network. DataCutter exports an interface for specifying processing as a set of coordination filters. While DataCutter offers a framework suitable for developing scientific and commercial data intensive applications in a grid environment, its programming interface is a relatively low-level one. The programmers need to decompose their application into a set of filters and specify the filters and their interaction.In this project, we propose a series of high-level language interfaces that can be used for expressing data-intensive applications. We target data parallel Java, XML query language (XQL), and mining operators as interfaces that will allow processing to be specified assuming that all data and compute cycles are available at a single site. We propose research in compiler techniques that will decompose such applications into a set of filters, based upon the availability of data, computing, storage, and networking resources.