Mapping Polygranular Parallel Processing to Shared, Heterogeneous, High-Performance Reconfigurable Computers

Peterson将多粒度并行程序(P3)映射到共享、异构、高性能可重构计算机(HPRC)摘要该项目旨在开发性能模型和映射算法，作为设计基础设施的理论基础，以支持在共享、异质、高性能可重构计算机(HPRC)上执行多粒度并行程序。该项目将开发一个数学框架，用于表征在HPRC系统上执行的多粒度并行应用程序的性能。此外，该项目将研究以最佳方式将应用程序映射到HPRC系统的计算效率方法，从而显著提高HPRC和相关计算平台的设计自动化效率。鉴于嵌入式系统和网格都成为普适计算的基础，该项目将专门解决与共享、异构环境中的多粒度并行应用程序执行相关的问题。该项目涉及新兴计算机系统体系结构领域，其更广泛的目标是提供必要的理论性能建模和映射基础设施，以有效地利用现有和新兴硬件平台的多粒度并行程序。因此，该计划支持各国通过网格和包括片上系统在内的嵌入式系统来满足下一代计算需求的努力。HPRC平台的计算能力为图像处理和图形算法等一系列问题领域提供了显著的性能和成本改进。

PetersonMapping Polygranular Parallel Programs (P3) to Shared, Heterogeneous, High Performance Reconfigurable Computers (HPRC)AbstractThe proposed project aims at the development of performance models and mapping algorithms as the theoretical foundation of a design infrastructure to support polygranular parallel programs executing on shared, heterogeneous, High Performance Reconfigurable Computers (HPRC). The project will develop a mathematical framework for characterizing the performance of polygranular parallel applications executing on HPRC systems. Moreover, the project will investigate computationally efficient approaches for optimally mapping applications to HPRC systems, providing dramatic improvement in design automation efficiency for HPRC and related computational platforms. Given the emergence of both embedded systems and the Grid as the basis for ubiquitous computing, the project will specifically address issues related to polygranular parallel application execution in a shared, heterogeneous environment.This project addresses the area of emerging computer systems architectures with the broader objective of providing the theoretic performance modeling and mapping infrastructure necessary to effectively exploit current and emerging hardware platforms with polygranular parallel programs. Hence, the program supports national efforts to meet next-generation computational needs with the grid and with embedded systems including system-on-chip. The computational capabilities of HPRC platforms promise significant performance and cost improvements for a spectrum of problem domains, such as image processing and graph algorithms.