A Self-regulating Compiler Framework for NoC Based CMPs

过度的电力消耗正成为我国日益严重的问题。大公司、数据处理中心和政府实验室目前正在花费数十亿美元来冷却和驱动他们的计算基础设施。此外，计算机的过度功耗会导致非常高的温度，最终影响其可靠性。这项研究涉及开发新颖的、基于软件的方法，以降低功耗，缓解相关的热问题，同时仍保持高性能。研究人员专注于基于芯片多处理器（CMP）系统的NoC（片上网络）运行时适应技术。所开发的技术是动态编译器的一部分，该编译器在运行时观察应用程序和机器状态，并使用一组辅助线程自动调整执行以适应动态变化的条件。这些帮助线程和应用程序线程共享相同的CMP/NoC资源，并在运行时进行协作和调整，以确保控制功耗，不超过热限制，并保持所需的性能水平。

Excessive power consumption is becoming an increasingly important issue for our nation. Major companies, data processing centers, and government labs are currently spending billions of dollars to cool and power their computing infrastructures. In addition, excessive power consumption by a computer leads to very high temperatures, eventually affecting its reliability. This research involves developing novel, software-based approaches to reduce power consumption and to mitigate associated thermal issues, while still maintaining high performance.The investigators focus on runtime adaptation techniques for NoC (Network-on-Chip) based CMP (Chip Multiprocessor) systems. The developed techniques are part of a dynamic compiler which observes both application and machine status at runtime, and adapts execution automatically to dynamically changing conditions using a set of helper threads. These helper threads and application threads share the same CMP/NoC resources, and cooperate and adapt at runtime to ensure power consumption is under control, no thermal limits are exceeded, and the required level of performance is maintained.