摘要
云计算的广泛应用导致数据中心的产生.数据中心的能效的高低不仅涉及到电费,还关系到否符合环境法规.作者通过修改Hadoop YARN编程模型,使用RAPI的能耗限制功能来降低应用程序中计算失衡时的能耗.目的是测试在不会明显地降低性能的条件下,通过RAPL,接口控制CPU的能耗是否有效.通过实验表明,在同样的负载下,Phadoop架构在分块矩阵乘法上相对于原来的Hadoop架构的能耗降低了34%.
An increased adoption of cloud computing has lead to a greater concentration of hardware in massive named datacenters.It is essential for these datacenters to be energy efficient not only to cut down on electricity costs but also to be in compliance with environmental regulations.The author implemented an enhanced version of Hadoop YARN framework that utilizes RAPL's power capping feature to mitigate computational imbalances in an application and to reduce CPU power consumption, named Phadoop. The purpose of the experiment is investgating whether it is beneficial to use RAPL interfaces to conserve the energy consumption ofa CPU in a cloud-based workload without significant loss of performance. Experimental results indicate a reduction in energy consumption of Phadoop up to 34% compared to Hadoop.
出处
《计算机系统应用》
2015年第11期235-241,共7页
Computer Systems & Applications