Constructing Scientific Applications as Heterogeneous Distributed Programs

目前，许多科学应用程序都是由一系列计算步骤构成的，每一个计算步骤都由一个单独的程序实现，并使用文件在步骤之间传输数据。在这个项目中，提出了一种基于结构化应用程序作为异构分布式程序的替代方案。这种方法的潜在好处包括改进对远程高性能机器的访问，增强与应用程序交互的能力，以及提高应用程序设计人员的灵活性。通过使用系统软件基础设施对用户透明地处理通信和异构性，这些好处几乎不需要额外的努力就能实现。这项研究的具体目标是开发新的应用程序编程技术和软件支持系统，以促进科学应用的这种类型的结构。第一部分涉及到使用Schooner RPC系统继续为此目的调整现有工具，以及研究与构建可以利用这些功能的新应用程序代码相关的问题。第二部分包括为软件系统确定适合科学应用需要的适当特征，然后构建这样的系统。所有这些工作都建立在先前涉及异构计算的研究之上。

Many scientific applications are currently structured as a series of computational steps, each of which is implemented by a separate program with files being used to transmit data between the steps. In this project, an alternative based on structuring applications as heterogeneous distributed programs is proposed. Among the potential benefits of this approach are improved access to remote high-performance machines, enhanced ability to interact with the application, and increased flexibility for the applications designer. These benefits can be achieved with little extra effort by using a system software infrastructure to handle communication and heterogeneity transparently to the user. The specific goal of this research is to develop new application programming techniques and software support systems to facilitate this type of structuring for scientific applications. The first part involves continuing work into adapting existing tools for this purpose using the Schooner RPC system, as well as investigating issues related to the construction of new application codes that can exploit these capabilities. The second part involves determining appropriate characteristics for a software system attuned to the needs of scientific applications, and then constructing such a system. All of this work builds on previous research involving heterogeneous computing.