全部博文(436)
分类: LINUX
2012-12-10 23:24:17
主要精读了4.2-4.4的内容
非预测性的主动热管理
Thermal-aware scheduling solutions is considered dynamic.These solutions are proactive since they attempt to facilitate favorable thermal conditions(促进良好的散热条件)
"Non-predictive": they rely on temperature measurements/estimations of the current chip state and do not create thermal models to forecast future temperatures(依靠温度估计目前芯片状态,不产生热模型来预测未来的温度,这就叫做非预测性的热管理)
存在的问题:虽然现代的芯片都装有软件可读的热传感器,但是读数受噪音干扰,以及存在其他可靠性问题。而且热传感器仅存在于芯片中的特定位置,as such,preprocessing temperature readings prior to utilizing them in any scheduler can increase its accuracy and effectiveness.(why ?为何利用上述问题能增加准确性和增强效用?)
因此便有人提出在离线时construct a thermal equivalent RC model for the circuit and reduce its complexity to make the model manageable.这个model 可将不可靠的传感器数据转换为可靠的温度读数。
有人提出了一套可以纳入操作系统,并且可以评估这些处理器中的效用的热点缓解技术。这个技术的two observations 不理解
线程(thread)指中的程序的调度单位。 (不理解线程与这个热管理有什么关系?)
什么是strategy throttle??不理解
预测性主动热管理
与非预测性主动热管理的区别:非预测性没有热模型来预测芯片的温度,而预测性热管理有这样的热模型。二者的目标和策略是相似的。
预测性热管理的缺陷:it adds significant complexity for the implementation both in terms of the code needed and the calculation time.(它增加了所需代码和计算时间两方面在实施时的复杂度)
优势:the scheduler becomes significantly more powerful being able to anticipate thermal crises as well as take actions that not only mitigate problems observed at the moment,but also reduce the probability of future crisis.(可以预见热危机并及时采取行动,这样就需要总去检查,降低发生危机的概率)
How temperature predicting models can extend the thermal-aware policies described earlier?
Example :(热模型)the rate of change of temperature for an application is proportional to the difference between its current temperature and its steady state temperature.
ABTM和CBTM
ABTM:用最小二乘法预测线程温度
CBTM:用微分方程、初始核心的温度,应用程序的稳定温度等来预测未来温度。