Research Of High Reliable Disk Array Technology Based On Intelligent Failure Prediction And Self-healing

With the rapid growth of the storage capacity of the data center, the reliability of the disk-based storage system is becoming one of the most important factors that influencing the reliability and availability of the entire computer system. However, the existing methods of improving the storage system reliability have some defects, such as fault redundancy is low, availability reduction when repairing the system, and the cost is high, so it can't meet the needs of practical applications.This paper takes the reliability of disk-based storage as of topic, and proposed an disk array storage system based on intelligent failure prediction and self-healing, which can reduce the risk of data lost and greatly improve disk array storage system's reliability and availability.What we have done are as follows:(1)We study the main ways to improve the reliability of storage. Based on deeply analyzing the operating principles of disks, we analyze the feasibility of some reliability techniques. We finally fix our mind on improving storage reliability by on line prediction.(2)We implement the collection of S.M.A.R.T information in the disk array environment, and adopt three attributes selection methods: sample mean, trend analysis and information gain to get the attributes set which are closely related to the states of running disks.(3)We study the principle and application of classify algorithm in machine learning methods. We implement Na?ve Bayesian Classifier and Support Vector Machine Classifier, and propose DTB classifier which is combined decision tree algorithm and boosting algorithm. Based on the performance testing, we reach the high accuracy of failure prediction while the low false alarm rate.(4)We propose a self-healing protection mechanism based on data migration. While applying the self-adaptive feedback theory, we use SFM-DMA to protect data on the disk which has been predicted to fail. The result indicates that the process of data self-healing can not only protect the data rapidly but also be transparent to the users.(5)Based on intelligent failure prediction and self-healing, we implement a high reliable disk array prototype called IFPS. We simulate the implementation process of the entire system through software, and also evaluate the feasibility and performance of the system. The results show that the reliability and availability of IFPS can be improved effectively.What we have done in our work provides a new mentality to enhance the reliability of disk-based storage, and some of the design concept and the key technologies are also helpful for improving the reliability of other equipments.