Research On Repairing Algorithm Of Node In The Distributed Storage System

In recent years,with the rapid development of information technology,the current society has gradually moved from the era of taking the computing devices as the core to the era of taking storage devices as the core,and data quantification has gradually become a trend.How to store these massive data effectively and reliably becomes a problem that needs to be solved urgently.Owning to the traditional centralized storage systems have limitations in reliability,scalability,etc.,distributed storage systems have gradually gained wide attention due to their features such as low cost and high scalability.In order to ensure the high reliability of the system,distributed storage systems often use redundant storage to acquire system reliability at the expense of a certain amount of storage overhead.The traditional redundant storage strategies mainly include replication strategies and erasure code strategies.In order to solve the shortcomings of the replication strategy in storing overhead and erasure code strategies in repairing bandwidth overhead,network coding technology is introduced into distributed storage systems,called regenerating codes,which is used to balance the storage overhead and repair bandwidth overhead of the system.This paper focuses on the research of node repair methods based on the regenerating code.The main tasks are as follows:(1)Node repair algorithm for distributed storage system based on MSR codeDue to hardware failures,human errors,and network failures and so on,some nodes in a distributed storage system are often unavailable and data stored on the nodes cannot be obtained.In this paper,such nodes are called failure nodes.In order to maintain the reliability of the system,designing a good node repair algorithm to repair failed nodes is very important for distributed storage systems.This paper proposes a new node repair algorithm for distributed storage system based on MSR code.This algorithm can repair single nodes exactly.The algorithm firstly groups the nodes in the system and stores the original files in groups.Each node has its own unique group and each group is independent of each other.Secondly,in each group,the original file data block is encoded with XOR algorithm and without any advanced operations in finite field.Finally,when decoding,each packet can be decoded independently at the same time.In addition,when there is a node failed in a group,by connecting to the surviving node in the same group and downloading a small amount of data,a new node can repair the data in failed node with XOR algorithm,which can reduce the disk I/O overhead and repair complexity.(2)Multi-node cooperative repair algorithm based on MBR codeIn addition to single-node failure,it is common that multiple nodes failed at the same time.And some distributed storage systems use delay repair,that means,repair operations will not start unless the number of failed nodes reaches a certain number.Therefore,it is also necessary to study the multi-node repair algorithm for distributed storage systems.Compared to decomposing multi-node repairs into single node repair one by one,cooperative repair of multiple nodes can reduce repair bandwidth overhead.This paper studied the multi-node cooperative repair method based on MBR code and proposed a new multi-node cooperative repair algorithm based on MBR code.Theoretical analysis shows that the proposed method achieves the theoretical minimum of its repair bandwidth.