Research Initiation Award: Parallel Program Portability with Coarse-Grained Dataflow Computing

本文将研究一种用于在MIMD架构上开发并行程序的编程环境的粗粒度数据流（CGD）虚拟机模型。该范式将使用三个标准进行评估：(a)跨各种并行架构的可移植性，(b)对性能监控和调试的有效支持，以及(c)易用性。为了研究这种范式的可移植性，开发了一个用于在MIMD架构上进行可移植并行编程的运行时系统Charm。迄今为止，已经使用Charm实现了许多实际应用程序，从而使其成为一个有效的测试平台。为了研究CGD作为一种可移植编程范例的有效性，将在Unix工作站网络上开发Intrepid——一种用于性能监视、仪器和调试的集成环境。使用Intrepid开发的程序将能够在各种MIMD架构上有效运行，而无需更改。在单一环境下，Intrepid将能够比较不同MIMD计算机上并行程序的性能。这将允许识别高效、可移植的CGD范例实现所需的体系结构和操作系统级支持。

A coarse-grained data flow (CGD) virtual machine model of computing for the development of a programming environment for parallel programs on MIMD architectures will be investigated. This paradigm will be evaluated using three criteria: (a) portability across a wide range of parallel architectures, and (b) effective support for performance monitoring and debugging, and (c) ease of use. To study the portability of this paradigm, Charm, a run time system for portable parallel programming on MIMD architectures has been developed. To date, numerous reallife applications have been implemented using Charm, thus making it an effective testbed. To study the effectiveness of CGD as a portable programming paradigm, Intrepid - an integrated environment for performance monitoring, instrumentation and debugging, will be developed on a network of Unix workstations. Programs developed using Intrepid will be capable of running efficiently on a wide range of MIMD architectures without change. In a single environment, Intrepid will enable the comparison of the performance of parallel programs on different MIMD computers. This will permit the identification of architectural and operating system level support that is required for the efficient, portable implementation of the CGD paradigm.