Disk Energy Saving Technology Research Based On Prefetching And Caching

A prominent problem faced by internet data service center is high disk energy onsumption which increases the system failure and decreases the reliability of the system. (?)here are some software and hardware level policies to save disk energy. At the software evel, based on the prefetching and caching method can merge disk I/O effectively, and makes he disk working full of the tasks, as well as extends the disk idle times. This is an important vay to achieve the disk energy saving. The thesis focuses on studying with the following ispects:Firstly, an energy efficient prefetching algorithm is proposed. To date, some problems still exist in traditional prefetching, for example, access sequence detection is too strict, (?)refetching size is too pessimistic, and read-ahead thrashings arise when prefetching pages are evicted prematurely under memory pressure. So, by introducing a trigger flag, it can make the iccess mode detection simply. On this basis, in order to support a large number of random (?)eading and writing, we relax the sequentiality criterion, and deal the traditional random iccess of large files with a sequential pattern. Meanwhile, its adaptive prefetching size can prevent read-ahead thrashing and reduce the disk energy consumption.Secondly, an energy efficient caching algorithm is proposed. After a deep study on disk page caching, an implementation of intercepting and dealing with disk I/O requests virtual device management model is designed. The model can intercept the requests from file system or application by virtual hard disk technology. It makes use of the request location correlation to reorder the I/O queue. Data in small writes is cached in a delay queue. All delay (?)/O requests will be written back one time when the disk activity is coming. When the disk is (?)usy, multi threads will be start. This algorithm can optimize the movement path of the disk (?)ead, as well as avoid the I/O congestion and improve performance.Finally, we realize the two algorithms and validate them on the Linux system. The experiments show that the energy effective algorithm extend the average disk idle time up to 10%, and improve the average data transfer rate nearly 2 times. In the same condition of workload, the disk energy reduced by the algorithms close to 17%. Experiments results show hat these approaches have good feasibility and practicality.