Continuous performance optimization on odern Architectures

如今，多核可伸缩性给从移动到数据中心的所有平台的开发人员带来了越来越大的挑战。包括英特尔、AMD、IBM和ARM在内的主要处理器供应商之间的激烈竞争正在引领计算机体系结构的新创新，系统软件和应用程序开发人员才刚刚开始利用这些创新。在这些平台上开发软件的复杂性越来越高，开发人员现在比以往任何时候都更容易接触到处理器体系结构的性能特性。影响其应用程序的性能、性能隔离，甚至安全。本研究的目标是提供工具和见解，帮助开发人员充分利用硬件。我们主要集中在三个方面：首先，设计和实现一个连续的性能分析系统，帮助开发人员理解现代处理器中一、二和三阶性能可度量的意义。其次，构建一个可移植的运行时，以帮助开发人员优化使用允许应用程序控制缓存行为的新指令，这可以说是对开发人员收益最大的最复杂问题之一。第三，探索如何使用该运行时和几个新的处理器特性来对应用程序进行分区，以确保性能、正确性和安全性。

Today, multicore scalability poses increasing challenges to developers across all platforms frommobile to the data center. The intensive competition between the major processor vendors includingIntel, AMD, IBM, and ARM is leading new innovation in computer architecture that systemssoftware and application developers have only begun to take advantage of. There is an increasedcomplexity of developing software on these platforms where developers are now exposed morethan ever to the performance idiosyncrasies of the processor architecture. Impacting performance,performance isolation, and even security of their applications.The goal of this research is to provide the tools and insights to help developers make the bestuse of the hardware. We focus on three pieces: First, design and implement a continuous profilingsystem that will help developers make sense of the first, second and third order performance metricsavailable in modern processors. Second, build a portable runtime to help developers make optimaluse of the new instructions that allow applications to control cache behavior, arguably one of themost complex issues with the largest gain for developers. Third, explore the use of this runtimeand several new processor features for partitioning applications for performance, correctness, andsecurity.