Collaborative Research: CSR-AES: REEact: A Robust Execution Environment for Fragile Multicore Systems

本研究重点关注称为 REEact（鲁棒执行环境）的高级执行系统的设计、实现和评估，该系统动态地使应用程序的执行适应源于下一代多核 (CMP) 芯片中的异构性来源的运行时资源环境。我们感兴趣的异构性来源是那些影响各个内核和内存块最大性能的工艺变化、DFVS 等功率优化（这些优化会导致内核性能随时间变化以及由于热而导致内核关闭）。紧急情况以及可靠性影响导致核心由于现场发生的永久性故障而必须被禁用。 REEact 是一种虚拟执行环境 (VEE)，用于协调、控制和调整应用程序的执行。 它采用多种技术组合来调整硬件资源和应用程序的软件代码，以适应 CMP 的异构性质，从而在并非所有 CMP 内核和/或内存块都可用的情况下提供最佳性能和功耗解决方案。 这项研究通过支持使用这些架构进行高性能计算来影响 CMP 技术。通过有效的异质性管理策略，科学家、消费者和商业人士将更有效地使用高性能应用程序，从而在药物开发、金融市场预测和环境科学建模等领域取得更大进步。这项研究通过本科生和研究生参与研究项目以及 CMP 的课程和课程模块对他们产生影响。

This research focuses on the design, implementation, and evaluation of an advanced execution system, called REEact (Robust Execution Environment), which dynamically adapts an application's execution to the runtime resources landscape originating from the sources of heterogeneity in a next generation multi-core (CMP) chip.The sources of heterogeneity of interest are those that arise from process variation that impacts the maximum performance of individual cores and memory blocks, from power optimizations such as DFVS that result in varying core performance over time and core shut down due to thermal emergencies, and from reliability effects resulting in cores that must be disabled due to permanent faults that occur in the field. REEact is a type of virtual execution environment (VEE) that mediates, controls, and adapts the application's execution. It employs a combination of techniques to adapt both the hardware resources and the application's software code to accommodate the heterogeneous nature of the CMP to provide the best performance and power solution where not all of the CMP cores and/or memory blocks are available. This research impacts CMP technology by enabling the use of these architectures for high performance computing. With effective strategies for managing heterogeneity, scientists, consumers and business people will more effectively use high-performance applications, leading to greater advances in areas such as pharmaceutical development, financial market forecasting, and environmental science model¬ing. This research impacts both undergraduate and graduate students through their involvement in the research projects and through courses and course modules on CMPs.