Investigations of Feasibility of Active Memory Heirarchy Support for Computational Field Simulation Problems

计算场模拟的并行处理工作 (CFS)问题突出了技术的需要， 支持大型数据空间问题之间的有效映射 存储器层次结构以有效的方式， 当处理器准备好使用它时，它就可用了。 到目前为止，典型的内存层次结构对 具有小工作集大小和良好局部性的程序 参考 CFS域中的问题通常涉及 巨大的数据集和解决方案技术， 整个数据集具有非常差的参考局部性。 的 本研究的重点是对软件的研究 允许应用程序员描述 应用程序的数据访问模式， 建立硬件系统以有效管理数据 在存储器层次结构的不同级别之间的移动， 以峰值速度向处理器提供数据。 这项工作将 包括模拟这个系统，使其有效 可以分析性能增强以确定 这种做法的可行性。

Work in parallel processing for Computational Field Simulation (CFS) problems has highlighted the need for techniques to support efficient mapping of large dataspace problems between memory hierarchies in an efficient manner so that data is available to the processors when they are ready to use it. Typical memory hierarchies to date have good support for programs with small working-set sizes and good locality of reference. Problems in the CFS domain typically deal with huge data-sets and solution techniques which make passes over the entire data-set have very poor locality of reference. The focus of this research is an investigation into software techniques for allowing applications programmers to describe the data access patterns for their application to allow a hardware system to be built to efficiently manage the data movement between the different levels of a memory hierarchy to provide data to the processor at peak speed. This work will involve the simulation of this system so that its effective performance enhancements can be analyzed to determine the feasibility of such an approach.