Research And Application Of Regenerating Codes In Distributed Storage System

Distributed storage systems can provide an effective solution to big data storage problems.In the storage system,the reliability of stored data is based on improving the performance of storage coding strategies.Among them,the regenerating code as a new coding strategy obtained from erasure coding strategy combined with network coding,has the optimal tradeoff between repair bandwidth and node storage,and reduces the possibility of network traffic during the repair process.The regenerating coding strategy has two extremal points in the cases of minimum storage and minimum repair bandwidth,namely minimum bandwidth regenerating code and minimum storage regenerating code.This paper focuses on the study of the regenerating coding strategy.The work is as follows:(1)Based on the existing(k+2,k)Hadamard MSR code,its coding matrix is improved and a new type of (k+2,k)Hadamard MSR code is proposed.In the new coding strategy,the computation load in the data repair process can be reduced by applying the modified Hadamard vectors to different locations in the coding matrices.The new coding strategy designed for repairing single systematic node,compared to the original(k+2,k)Hadamard MSR code,can complete the optimal repair of lost data with fewer operations in the Galois field.(2)Through the in-depth study of the locality in the locally repairable code,the new(k+2,k)Hadamard MSR code obtained above is combined with the locality to construct a new MSR local regeneration code.Thus,the MSR local regeneration code can have both locality characteristics and minimum repair bandwidth characteristics.Compared with the locally repairable code,the MSR local regeneration code not only can effectively reduce the repair bandwidth when single node failure,but also enables the system to implement multi-node repair.This provides a new idea for the design of the regenerating code.(3)Based on the VS2010 platform,the paper verifies the above two coding strategies.Simulation results show that the new(k+2,k)Hadamard MSR code can complete single-node repair with minimum repair bandwidth.The MSR local regenerating code enables the system to achieve multi-node repair.In this paper,two regenerating coding strategies are designed to apply to the single node repair and multi-node repair in the system respectively,which enrich the design ideas of storage coding strategies.