Research On Self-Healing Disk Array Technology

With the amount of data becomes the level of PB in science technology, Science Computing has become data-intensive. Data-Centric computing model needs a new standard for the availability and reliability of data. Massive, high-reliability and high-performance storage devices become the basic elements of HPC. At the same time, inevitable failures of storage components grow with the capacity and performance's enhancement. As the control system pose increasing complexity, failure rates of storage system increases rapidly.Be transparent to the users, that is, processes of the corrupt data recovery was run without decreasing the performance of data's read/write operation. Transparent to users was also named"self-healing"and it becomes another goal of storage system.This article mainly researches on the technology of storage system's capacity, performance, reliability and the ability of being transparent to users.The Self-Healing technology which addresses the symptom rather than cause becomes a principle of disk array designing. Based on the basic disk read/write process and combined with the reliability level model in the theory of reliability derived the disk failure model tree both in the form of top-down and bottom-up analysis. Then design and realize the disk monitor algorithm T2SU with some appropriate parameters for the environment from the leaves of the two model trees and the characterization of the current disk state from the S.M.A.R.T attributes. Respectively, T2SU makes use of the threshold method to determine the quality of the environment as well as the expansion of self-learning algorithm and self-learning algorithm to estimate the compatibility of the state. Accompany with the theory of statistics, the prediction technique of the disk's MTTF in accordance with probability distribution form from now on was created. RAID 5T disk array structure with "self-healing" technology was designed according to disk status monitoring. This disk array architecture was transparent to users when a disk failure happens with the TRDP disk data migration and restructuring strategy.In order to achieve high-performance of data's write/read operations, a multi-RAID structure: RAID 6 (4) / 5T (8) was designed by general computer components such as motherboard, CPU and others. The form of redundancy information data and source data in this array was plane. In accordance with the principles of hierarchical management, after the SNIA's standard about RAID 6 was achieved, it will generate data both of source data and parity data by Liberation algorithm and deployment all the data of RAID 6 as the data source of the following RAID 5T arrays. With the re-parity information, At least Multi-RAID can still work after 6 disks'failure.