Research On Reliability Technologies For Disk-based Storage Systems

With the explosive growth of digital information, the increasement of number of us-er groups and multimedia applications, the demands of mass information storage systems also have grown. The needs of users and data for storage systems are different, some us-ers and applications need to be provided higher reliability, some users and applications require the storage system providing lower response time. The needs like those make storage systems facing with the pressure of managing a large number of different data and storage systems must have high availability. Additionally, with the growth of the scale of massive information storage systems, the pressure of enterprise in energy con-sumption, cooling management and room space management room is grown. At today's storage systems, disk is the essential device to constitute mass storage system. Therefore, how to construct a disk storage system with high reliability, high performance, low power consumption and suitable for different applications becomes an important task.The function of disk scanning or scrubbing operation is identifying the potential failure sectors as early as possible. Obviously, the higher the scrubbing frequency is the higher reliability the system has. It may take a few hours for a scanning process to check the whole disk. The scrubbing process may result in a lower system performance. There-fore, how to choose the optimal scrubbing frequency has become a challenge to the ad-ministrator. In this paper, we present a novel scrubbing scheme to solve the challenge. In this scheme, an optimum scrubbing cycle is decided by keeping a balance between data loss cost, scrubbing cost, and disk failure rate. When the disk is running to 17520 hours, our research shows that the maximum cost of optimal scheme decreased by 73.3% and it has the highest reliability in a various disk scrubbing scheme.With the growth of information, new storage device are added into storage system. Additionally, the disk failures in large-scale storage can occur almost every day, thus, the reconstruction or expansion process is very common. These processes will bring the per-formance of front-end application to very poorly. In order to protect the performance of storage system, we propose a method based on grey theory (Seesaw) to achieve this aim. In this scheme, the laws of front-end application I/O requests are monitored.then control the reconstruction speed in order to provide enough bandwidth to front-end application. Compared to not taking Seesaw, our approach performs well on the write-dominated trace like Financiall and the speed-up ratio is 9.05 when RAID is in the reconstruction process.With the constant growth of data center scale, the latent sector errors become very common in data center. Two schemes are used to prevent the data loss in a single disk caused by latent sector error. One is the disk scrubbing technology; the other is the in-tra-disk redundancy code(IDR). The IDR scheme gains higher reliability at the expense of lower disk performance and the scheme isn't suitable for the application which re-quires high performance. Based the research that the sector which has heavy load has higher failure rate, we propose a scheme that replicate the popularity data block to im-prove reliability and improve the performance by choosing to access the data block which near to the disk head. Theory analysis shows that the scheme greatly improves the disk's reliability. The performance experiment shows that the disk access time will de-crease by 20% approximately when the disk is integrated this scheme.The archival storage systems which use Shamir threshold scheme split the data into n parts. In a (k, n) threshold scheme, it only need any k pieces of data to recover the original data. In this paper, we build an energy-efficient archival storage system Green-Arc. GreenArc distribute the data into n storage nodes by threshold scheme. k of n stor-age nodes provide the storage service, while the other n-k storage nodes remain standby mode in order to save energy. Experimental result shows that when archival storage sys-tem uses the same threshold, the lighter the load in the system. The more energy can be saved by GreenArc. With the increase of k value, the effect of saving energy is more re-markable.