Research On Reconstruction Technology Of Fault Tolerant Storage System Based On Block Codes

With the advent of the era of big data,the erasure code technology in fault-tolerant storage systems has received more and more attention due to its low storage overhead and high fault tolerance ability.The reconstruction efficiency of erasure code technology is the bottleneck of the wide application.At present,the researches on reconstruction technology mainly starts from two aspects: coding scheme and reconstruction process.The block code is a new type of erasure code that improves the efficiency of data reconstruction from the perspective of coding scheme.It has the advantages of simple coding structure,low reconstruction overhead,et al.However,the coding parameters of the existing block codes are not flexible enough to achieve dynamic ratios of storage overhead,fault tolerance ability,fault tolerance rate,and reconstruction overhead.Moreover,there are few researches on the performance improvement of the block code in the reconstruction process,and it is difficult to achieve an effective combination of the coding scheme and the reconstruction process.In view of the above problems,this paper studies the coding structure and algorithm of existing block codes and the main techniques in the reconstruction process and improves the reconstruction technology of fault-tolerant storage systems from these two aspects to enhance the efficiency of data reconstruction.The research work of this paper is divided into the following two parts:(1)This paper proposes a dynamic coding scheme based on block codes called DLRC(Dynamic Local Reconstruction Codes).Four parameters are selected to characterize DLRC:the number of initial data blocks,the number of global parity blocks,the number of local parity blocks,and the number of coded blocks involved in the calculation of each local parity block.The coding structure and coding algorithm of DLRC and the main flow of reconstruction using DLRC code are given in this paper.Then we analyze the ratios of storage performance,fault tolerance ability,fault tolerance rate and reconstruction overhead by adjusting the coding parameter values and experimenting with the corresponding coding algorithms.The advantages and disadvantages of DLRC in the above four performances are verified by comparing DLRC with the commonly used erasure codes.(2)This paper proposes a tree-structured reconstruction method based on network topology and computing capability(TR-NTCC).Firstly,the weighted undirected graph of thenascent node and the reconstructed participating node is generated by using the inter-node network distance as the weight and the corresponding adjacency matrix is obtained.Then,according to the MRTree(Minimum Reconstruction Tree)algorithm mentioned in this paper,minimum reconstruction trees are constructed and determine if the minimum reconstructed trees is unique.When the minimum reconstruction tree is unique,this minimum reconstruction tree is the optimal reconstruction tree.When the minimum reconstructed tree is not unique,the computing power of the different nodes in different reconstruction combinations is judged,and the minimum reconstructed tree constructed by the reconstructed participating nodes combined with strong computing ability is selected as the final optimal reconstructed tree.The experiment verifies the reconstruction performance of TR-NTCC by comparing the total network transmission distance of TR-NTCC and star reconstruction method and the reconstruction time of different reconstruction combinations.