Study On Fault Tolerance Technology Of Data For Distributed Storage Systems

With the rapid development of Internet technology,there is an explosive growth in the amount of data.Distributed Storage Systems(DSSs)have become a mainstream way of storing massive data today.With the continuous expansion of DSSs,node failures have become frequent.In order to achieve effective repair of node failures,data redundancy schemes based on different coding techniques are introduced into DSSs.To address the shortcomings of traditional fault-tolerant techniques,regenerating codes combined with network coding theory reduce the the consumption of bandwidth resource for repairing failed nodes,but such codes need more nodes to participate in the repair process and have high disk I/O overhead as well as computational complexity.Furthermore,Locally Repairable Codes(LRCs),which require only a small number of nodes to be accessed in the repair process,and Fractional Repetition(FR)codes,which do not involve the operation of encoding and decoding in the repair process of nodes,have been proposed one after another.This thesis puts forward three reliable storage schemes of data for DSSs based on the existing LRCs and FR codes,and the specific research contents are as follows:(1)Considering the failure of multiple data blocks in storage system and how to provide parallel access for hot data,the construction method of binary single-check LRCs with(r,t)locality of information symbols is proposed based on resolvable balanced incomplete block design,and the constructed LRCs are the codes with optimal minimum distance.In particular,the LRCs are also codes with optimal code rate when the availability t = 2.Furthermore,a class of LRCs that can tolerate more data blocks failures is constructed by combining the incidence matrix constructed based on the resolvable balanced incomplete block design with the generation matrix of Cauchy Reed Solomon codes.The LRCs constructed are still the codes with the optimal minimum distance.(2)Through the research on the relevant structures of LRCs,it is found that most of the existing structures are for systems with given parameters,and the parameter selection of LRCs is not flexible enough.In this thesis,LRCs with availability t = 2 and LRCs with locality r = 2 are constructed by using the combination relationship of triangular association schemes.Further,the LRCs constructed based on triangular association schemes are extended to construct LRCs that can achieve arbitrary locality r > 2 and availability t > 2.The constructed LRCs with(r,2)locality are the codes with the optimal code rate,and the constructed LRCs with arbitrary locality r > 2 and availability t > 2 are the codes with the optimal minimum distance.(3)Considering that the parameters of the actual DSSs may change dynamically over time,a class of FR codes with the same storage capacity that can flexibly select the size of nodes and number of repetitions of coded blocks in the system is constructed based on the rotation operation matrix,and then a class of heterogeneous FR codes with different storage capacities is constructed by changing the rotation mode of the rotation matrix.The two FR codes constructed are universally good FR codes,and under certain conditions,the constructed FR codes with the same storage capacity are the optimal FR codes,and they are also the FR codes with the optimal minimum distance and the optimal reconstruction degree.