Research On Low Power Data Layout Technologies For Storage Systems

The explosive expansion of digital information motivates the emerge of large scale stor-age systems. With the advantages of both large capacity and low cost, disk has been the mainstorage media. A large number of disks are organized together to build mass storage systemswith high performance and high reliability. Mechanical characteristics of disks make diskbased storage systems one of the main sources of the energy consumption of data centers.Energy cost in the proportion of the total cost of ownership of data centers has been growingover time. Therefore, decreasing the energy consumption of disk based storage systems isthe key to the decrease the total energy cost and and then the total cost of ownership of datacenters.The working states of disks are divided into the following states, which are active, idleand standby. When responsing to an I/O request, disk must be kept in active state with con-siderable energy. Data layout of disk array determines the spatial distribution of I/O requestsissued to every disk, and further in?uences the working state and energy consumption of ev-ery disk. The conventional disk array data layout either overlooks the storage system energyconsumption or has lots of shortcomings in storage system performance and reliability as itreduces storage system energy consumption. Although solid state disk has advantages in bothhigh random reads performance and energy efficiency, its physical characteristics determinethat it has intrinsic shortcomings in both small writes performance and erasure cycles. There-fore, building mass storage systems with high performance, high reliability and low energyconsumption via optimizing the disk array data layout is a significant and urgent task.The conventional logging data layout schemes concentrate logging data onto specializedlogging disk to spin down mirrored disks to standby state to save energy. However the spe-cialized logging disk makes the storage systems suffer from the problems of single point offailure, performance bottleneck and high energy consumption. Therefore, a rotated loggingarchitecture RoLo (Rotated Logging) and the corresponding decentralized destaging mecha-nism are proposed. RoLo integrates the unused free space of redundant mirrored disks in a RAID10 system into a large logical logging space pool. By rotating on-duty logging spaceamong the free storage space of all the mirrored disks,and reclaiming the stale logging spacevia decentralized destaging, the reclaimed logging space can be reused, and thus the limitedmirrored disks free space can be used as a logical unlimited logging space. Since no extralogging disk is used in RoLo, it eliminates the additional hardware and energy costs and thepotential single point of failure and performance bottleneck of the conventional centralizedlogging architecture. The relatively large logging space reduces the disk spinning frequency,and thus can not only prolong the lifecycle of disks, but also reduce the storage system energyconsumption. The experimental results demonstrate that the rotated logging RoLo can notonly save up to 47.2% energy over RAID10, but also save up to 11.8% energy over GRAID,which adopts the conventional logging architecture with fixed and dedicated logging disk.The conventional file placement schemes concentrate files onto a few disks by only uti-lizing either temporal locality or spatial locality of file accessing, instead of fully exploitingand utilizing both bimodal distribution of file accessing and the complementarity betweensolid state disks and hardware disks in performance, reliability and energy consumption tooptimize the file placement. Therefore, a dynamic data layout reorganization scheme LoRo(Layout Reorganization) is proposed to further reduce the energy consumption of a mixedstorage system composed by both solid state disks and hard disks. By utilizing both theI/O access locality and the I/O access bimodal distribution of storage system I/O workloads,LoRo issues reads onto solid state disks, which have an edge in both random reads perfor-mance and energy efficiency, meanwhile LoRo issues writes onto hard disks, which have anedge in sequential writes performance. LoRo can reduce the energy consumption and im-prove the performance of the mixed storage system simultaneously. The experimental resultsdemonstrate that LoRo can save up to 68.2% energy over PDC which adopts popular dataconcentration scheme.Some disks can be spun down to low power standby state to save energy by replicatingtheir data onto the other disks. However, the energy saving of the conventional data layoutscheme which is usually deployed in single storage volume and adopts data replicationmechanism is constrained by both free storage space and storage workload intensity. Bothfree storage space and storage workload intensity are restricted each other and they limit the energy saving of storage systems. Therefore, a low power file layout scheme called AdapMig(Adaptive Migration) is proposed to improve the energy efficiency of multiple volumes basedstorage systems. The basic idea of AdapMig is to minimize the total energy consumption ofmultiple volumes based storage systems without sacrificing the predefined average responsetime via redistributing the files among multiple volumes periodically. AdapMig firstly modelsthe files distribution problems among multiple volumes based storage systems, and thensolves the optimization problem with multiple constrains by genetic algorithm, and finallydetermines the new file layout. AdapMig can effectively utilize the free storage space locatedin multiple volumes based storage systems and optimize the file layout among multiplevolumes based storage systems. The experimental results demonstrate that AdapMig cansave up to 26.3% energy over PARAID, which dose not optimize the file layout amongmultiple volumes based storage systems.