Energy-Efficient Computing For Storage Sub-system

The storage system is one of the largest energy consumers within computer systems.Intensive research has been focused on lowering the energy consumption of storage sub system.And still, its performance is one of the bottle necks to the computer system's overallperformance. Hence, energy efficiency has become a central issue of the storage sub system.There have been various energy saving strategies and algorithms so far. However, thoseworks were solely for the purpose of energy saving and ignored performance enhancement. Withthe emergence of new storage medium solid state drive (SSD), we have more opportunities toimprove energy efficiency. As such, this paper aims for energy efficiency of storage subsystemby exploiting SSD. Strategies of"file"and"data block"granularities are proposed. The maincontributions are as follows:1. By using SSD as a large buffer for the hard disk(HDD),we propose a heterogeneous disksystem hetero-drive for energy efficiency of the storage systems by combining the SSD withHDD. The main idea is to cache"hot"data block into the SSD to create enough free time forHDD to spin down. Some novel data structure is contrived, and a decay enforcementreplacement algorithm is used for buffer management. To validate the effectiveness of ourproposed solution, we have performed extensive experiments with real world data traces. Theexperimental results show that our strategy achieves very competitive performance against thosetraditional buffer replacement algorithms. For all traces, the read/write rate improves by 265%and 25%; for OLTP data sets, our method reduces energy consumption by 28%.2. By exploiting the "exponential average" scheme, access frequency is collected andcomputed for each file; the files are replaced/relocated on the hetero drive according to itsreplacement cost, and such migration extends the idle period of the HDD. With the help of thedisk working model and "exponential average" approach to predict the next file access requestarriving interval, HDD makes reasonable state transition to reduce energy consumption to thefullest extent. Extensive experiments demonstrate that the proposed scheme can save energyconsumption about 40% and achieve a very promising result.