全部博文（436）
分类： LINUX
2012-12-10 20:35:05
《Survey of Energy-Cognizant Scheduling Techniques》读书报告
这一次，我们采取小组分工阅读文章，我主要读3.2和3.3.
3.2Algorithms using DVFS and DPM
使用dvfs和dpm算法
Dvfs动态电压和频率调整
Dpm 动态能量管理系统
DVFS通过降低电源电压和时钟频率来降低能耗，但这样做会延长程序执行时间。
Early research efforts exploiting DVFS to reduce energy consumption arose in the context of real-time systems, where tasks’ deadlines are given and their WCETs (Worst-Case Execution Times) are typically known beforehand.早期的研究工作利用dvfs减少能源消耗产生的背景的实时系统，任务的最后期限已给出，并给出了其wcets（最长执行时间）通常是预先知道的。Many algorithms proposed in this domain have leveraged the performance slack available in real time applications许多算法在这一领域的提出已经改变限制实时应用的性能。The common approach is to minimize energy consumption by reducing the performance of the processor (DVFS level) but without causing the applications in the workload to miss their deadlines常见的方法是减少能源消耗，降低了处理器的性能（dvfs水平），但是没有造成程序工作量而错过最后时间。Choosing the right DVFS setting for combo applications is a delicate balance between reducing power consumption and increasing execution time.选择正确的dvfs设置的组合应用是一个降低能耗，提高执行时间之间的微妙的平衡。
DPM利用一些电子设备可能保持者空闲状态却消耗着能量所以可以把它们关闭来减少能耗。
3.3执行物理资源映射
执行物理资源映射
讨论了工作迄今在这部分处理所有计算内核是完全等价，因此并没有解决这一问题的映射执行物理背景。
improve performance in the presence of core frequency variation such as first utilizing cores that support the highest frequency or mapping high-IPC threads to high-frequency cores.提高性能存在核心频率变化首先利用内核，支持最高频率或映射high-ipc线程高频磁芯。The authors propose compile-time support that will estimate the workload assigned to each core (in aparallelization of nested loops scenario) and apply DVFS to the cores with less work assigned to them such that all threads will finish at roughly the same time. This makes each core consume the minimal energy possible to get its portion of the loop done, while keeping the execution time for the whole loop unchanged. 作者提出了编译时间，将估计的工作量分配到每个核心（在并行的嵌套循环的情况）和应用的核心dvfs少工作分配给他们的所有线程完成大约在同一时间。这使得每个核心消耗能量最小可能得到它的循环部分，同时保持执行时间为整个循环不变。
问题：
一些英文词汇翻译过来很奇怪，无法理解
选择正确的dvfs设置的组合应用是一个降低能耗，提高执行时间之间的微妙的平衡。不太明白。
每个线程度量μ是干什么的，具体作用
