Exploiting multi-megabyte on-chip memory hierarchies

在几乎所有现代电子设备的核心，如计算机、手机、MP3播放器、游戏机和网络路由器，至少有一个处理器。处理器是一种可以处理以数字表示的信息的设备。处理器非常有用，因为它们可以通过传感器(例如麦克风)和执行器(例如扬声器)与环境交互。处理器在科学探索中也非常有用，因为它们可以用来评估复杂的物理现象模型，如化学、生物和星际相互作用，这些模型过于昂贵或不切实际，无法以其他方式进行研究。今天，许多日常活动和许多科学探索都依赖于优化、廉价的处理器的可用性。从历史上看，在降低成本的同时，可以不断提高处理器的性能和其他特性(例如，能源消耗)。与80年代初的计算机相比，如今一台典型的计算机的成本至少低了10倍，而功能却高出约1000倍。通信和计算行业的大部分都依赖于这一趋势。改进处理器的一个关键机制是计算机体系结构，它研究如何在给定现有制造技术的情况下构建处理器，同时考虑到目标应用(S)。计算机体系结构面临着持续的挑战，因为底层技术和目标应用程序的属性随着时间的推移而发生重大变化。这项研究的目标是通过应对新的挑战和利用基础制造技术的不断萎缩以及多媒体和面向互联网的应用的出现带来的新机遇来改进处理器。解决的挑战包括性能、设计复杂性和能源消耗。这项工作的目标是设计一种片上存储器层次结构，提出粗粒度管理和跟踪，并利用其能力更好地预测和优化应用程序需求。将从这项工作中受益的主要是大型数据中心(如用于基于互联网的服务和大型组织的数据中心)中使用的高端性能导向型处理器。

At the core of virtually all modern electronic devices such as computers, cell phones, mp3 players, game consoles, and network routers there is at least one processor. A processor is a device that can manipulate information represented as numbers. Processors are extremely useful because they can interact with the environment via sensors (e.g., a microphone) and actuators (e.g., a speaker). Processors are also extremely useful in scientific exploration as they can be used to evaluate complex models of physical phenomena such as chemical, biological and inter-planetary interactions that too expensive or impractical to study otherwise. Today many everyday activities and much of scientific exploration rely on the availability of optimized, inexpensive processors. Historically, it was possible to continuously improve the performance and other characteristics of processors (e.g., energy consumption) while reducing cost. A typical computer today costs at least 10 times less and is about 1000 times more powerful than a computer of the early 80s. Much of the communications and computing industry relies on this trend. A key mechanism for improving processors is computer architecture, which studies how to build processors given the available manufacturing technologies while taking into consideration the target application(s). Computer architecture faces continuous challenges because the properties of the underlying technology and the target applications change significantly over time. This research is targeted at improving processors by addressing new challenges and taking advantage of new opportunities that stem from the continuous shrinking of the underlying manufacturing technology and from the emergence of multimedia and internet oriented uses. The challenges addressed include performance, design complexity, and energy consumption. This work targets the design of a on-chip memory hierarchies proposing coarse-grain management and tracking, and leveraging its capacity to better anticipate and optimize for application needs. It is primarily high-end performance-oriented processors used in large data centers (such as those used for internet-based services and in large organizations) that will benefit from this work.