Research On Key Technologies Of Energy Conservation Or Disk-based Storage Systems

Energy conservation and environmental protection are major issues for governmentsaround the world today. Chinese government has put energy conservation on the importantposition in the national economy work. In recent years, with the rapid growth of the totalglobal digital information, the energy issue of storage systems has become increasinglyserious. How to reducing the energy consumption of storage systems has become aparticularly urgent issue and a hot research topic in the field of computer science. Therefore,to expand research in this area has significant academic value and practical significance forpromoting the development and application of energy conservation technologies for storagesystems.This dissertation starts from the energy conservation principles of disk-based storagesystems; its purpose is to improve the energy efficiency and performance of disk-basedstorage systems. It analyzes the energy saving requirements and key energy conservationtechnologies of disk-based storage systems and then focuses on key technologies of energyconservation for disk-based storage systems, including disk power management, datamanagement for cache disks, disk load balancing and system model of energy-awaredisk-based storage systems. The main research works and innovatives of this dissertationinclude:(1) This dissertation proposes a disk power management strategy with access modeself-matching and performance guarantee named Access Mode Matching and PerformanceGuarantee (AMM-PG). To overcome the weakness of existing disk power managementstrategies, AMM-PG uses data histogram of idle interval lengths to correspond to the diskaccess mode and exports the corresponding time-out threshold, to realize reasonable powermanagement of disk which adapt to the disk access mode, and then to improve the energyefficiency of the disk. This strategy also adds the consideration that when disk return to idle state from standby state to some disk access modes and proposes a performance guaranteemethod. It determines the maximum time that a disk stays in standby state by performancerequirements and data histogram of idle interval lengths, so that it can meet the performancerequirement and maximize the energy saving at the same time. Theoretical analysis andexperimental results show that compared with the fixed time-out strategies and adaptivemanagement strategies share, AMM-PG has better energy efficiency.(2) This dissertation proposes a load adaptive energy-aware cache management methodnamed Workload-Adaptive LRU (WA-LRU). To overcome the weakness of existing cachemanagement methods, WA-LRU distinguishes the energy priority of the data disksaccording to the number of hot data and the intensity of data accesses. When data in cachedisk need to be replaced, it always replace the data block comes from the disk with thelowest priority to match the disk access pattern and improve the energy saving effect. Themethod also use the Least Recently Used (LRU) queue base on life cycle management toidentify and eliminate the bottleneck effect of the cache disks, thus avoiding the low energyefficiency problems caused by less data from multiple data disk be cached. Theoreticalanalysis and experimental results show that compared with the existing Least Recently Used(LRU), Power-Aware LRU (PA-LRU) and Partition-Based LRU (PB-LRU) method,WA-LRU has better energy-saving effect.(3) This dissertation proposes a multi-queue load balance method for energy-awaredisk-based storage system named Multi Queue Load Balance (MQLB). MQLB focuses onthe problem of energy conservation solutions base on overall storage nodes that the diskswhich the load skews to become overloaded easily. MQLB divides data into multiplequeues by popularity according to the data access frequency, the temporal locality of accessand other factors. It only maintains the data which have been frequently accessed recentlyas migration objects and uses hot-degree to measure the disk load pressure bring by a HotData Unit (HDU). When a disk becomes overloaded, MQLB determines the appropriateLoad Press Ratio (LPR) according to the degree of overload, and then chooses appropriateHDU according to the LPR to carry out data migration. Theoretical analysis and experimental results show that compared with the existing Popular Disk Concentration(PDC), Workload-Adaptive Management (WAM) and Energy Model-based File MigrationStrategy (EM-FMS) method, MQLB can adjust the load more effectively and quickly andintroduce a smaller overhead.(4) This dissertation proposes a new energy-aware disk-based storage system modelnamed Dynamic Tree-based by Massive Array of the Idle Disks (DT-MAID). DT-MAIDfocuses on the bottleneck problems of current energy-aware disk-based storage systems andproposes an idea that research on energy-aware disk-based storage system model fromsystem perspective. The design ideas of DT-MAID are: it uses two-layer dynamic treestructure to organize cache disks and divides the cache disks in the dynamic tree into tworoles of Master and Slave; dynamic tree manager orients load to Master, in favor of otherSlave into standby state to improve the energy efficiency of the system; Slave nodes are notalways in the rotation, they will enter standby mode to save energy when their load is lightand meet the set conditions; when the load on Master is heavy or the accessed data coverageof the system is large, the dynamic tree manager will choose an appropriate node fromSlaves not in standby state or recover a Slave in standby state to take over its work, toimprove the system's service capacity and energy efficiency; The theoretical analysis andexperimental results show that DT-MAID has good energy saving effect, service qualityand reliability.