AKULA-T：多核环境下评估和开发温度感知线程放置算法的框架

发布时间：2018-07-07 18:45

本文选题：温度感知线程映射 + 多核模拟器　；参考：《内蒙古大学》2013年硕士论文

【摘要】：目前,多核处理器已成为通用微处理器设计和实现的主流。多核处理器因多道程序在多核内同时执行导致其功耗过高,这些功耗以热能的形式向外散发,使处理器的温度不断上升。当处理器工作温度过高时,处理器性能下降,使用寿命缩短,比较严重时会损坏器件。因此,温度已经成为制约处理器性能和稳定的关键因素。比较典型的减少功耗并降低温度方法有动态电压频率调整和门控时钟等,但是这些硬件技术会带来不小的性能损失。如何从软件操作系统方面通过线程调度算法合理的调度当前任务达到减少功耗、降低温度和提高处理器可靠性的目的有着重要的现实意义。但是,开发新的线程调度算法面临非常大的挑战,一个是制定新的调度算法的实现难度大,另一个是算法测试时间长。一个算法从开始提出到真正实现,所花费的时间精力是巨大的,而且最后调度算法的有效性也是值得考虑的一个问题。很多时候花费巨大的时间和精力开发出的算法,如果效果不理想,则意味着巨大的浪费。本文提出了一个多核环境下温度感知调度算法的开发和评估框架。该框架可以帮助研究人员在其上进行多核微处理器面向温度问题的线程调度算法的开发和研究,可以节省人力物力。这个调度框架提供了一个API,允许开发人员实现和调试,而不需要修改内核或者使用系统调用,并提供了简单的评估模型。这个调度框架也有利于建立在真实机器上运行,而不需要修改调度算法。研究人员可以使用这个框架开发和评估的不同的调度算法,设计出合理的线程调度算法,使处理器实现高性能、低开销、高可靠性。最后通过实验表明,这个框架可以有效开发和评估新的温度感知算法。
[Abstract]:At present, multi-core processor has become the mainstream of universal microprocessor design and implementation. The power consumption of the multi-core processor is too high because the multi-channel program is executed simultaneously in the multi-core. The power consumption is distributed in the form of heat energy and the temperature of the processor is rising continuously. When the working temperature of the processor is too high, the processor performance will decline, the service life will be shortened, and the device will be damaged when it is more serious. Therefore, temperature has become a key factor restricting processor performance and stability. The typical methods to reduce power consumption and reduce temperature include dynamic voltage frequency adjustment and gating clock, but these hardware technologies will bring not small loss of performance. It is of great practical significance to reduce the power consumption, reduce the temperature and improve the reliability of the processor through the rational scheduling of the current tasks through the thread scheduling algorithm in the aspect of the software operating system. However, the development of a new thread scheduling algorithm is facing a great challenge. One is that it is difficult to implement the new scheduling algorithm, the other is the long testing time of the algorithm. From the beginning to the implementation of an algorithm, it takes a lot of time and effort, and the effectiveness of the final scheduling algorithm is also a problem worth considering. Many times it takes a lot of time and effort to develop the algorithm, if the effect is not ideal, it means a huge waste. This paper presents a framework for the development and evaluation of temperature sensing scheduling algorithms in multi-core environments. This framework can help researchers to develop and study the threading scheduling algorithm for multi-core microprocessors facing temperature problem, and can save manpower and material resources. The scheduling framework provides an API that allows developers to implement and debug without modifying the kernel or using system calls, and provides a simple evaluation model. The scheduling framework can also be used to run on real machines without modifying the scheduling algorithm. Researchers can use this framework to develop and evaluate different scheduling algorithms and design a reasonable thread scheduling algorithm to enable the processor to achieve high performance, low overhead and high reliability. Finally, experiments show that this framework can effectively develop and evaluate a new temperature sensing algorithm.
【学位授予单位】：内蒙古大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP332

【参考文献】