| Summary: | 碩士 === 大同大學 === 資訊工程學系(所) === 103 === With cloud computing technology was developed, performance is a key point, however it is limited by the hardware devices. When the performance is unable to improve effectively, energy-saving becomes an important issue. This paper uses Hadoop's distributed compute framework called MapReduce, and exploits the distributed storage architecture of Hadoop's distributed file system HDFS to do the energy-saving test.
After referring many energy-related technical, we found a method had a better effect on energy-saving that is GreenHDFS. It is a subproject of the Apache Hadoop projects, and evolved from HDFS. In this paper, we know the internet use peak time and off-peak time through Bai-du statistic's survey. When the capacity of data processing is too high, we don’t let the node into sleep mode, it maybe not only can't save energy but paying for more energy. Therefore, we design an energy-saving management policy for the user to access data from the node in off-peak time. It will reallocate and manage the task on nodes. As the time from peak time to off-peak time, the capacity of data processing will slowly down and some nodes maybe idle. To avoid this problem, some of the nodes will go into sleep mode, and allow other nodes to access data. The sleep node will be awaked for large amounts of data into the node.
The Experimental results shows, compared to the original method, in off-peak time to allow the idle node going into sleep mode can effectively reducing the overall energy consumption of 21%.
|
|---|
