Distributed Storage Codes Based On Piggybacking Framework

The rapid development of the Internet,5 G and their related industries has led to an explosive growth of data,which poses a huge challenge to the storage of big data.In order to solve the problem of failed nodes in big data storage,distributed storage codes with fault-tolerance and saving storage resources have become one of the key technologies in the development of big data.As a class of distributed storage codes,piggybacking codes,with their excellent node repair performance and high storage efficiency,coupled with their low complexity and flexible design,have attracted more and more attention in recent years.This thesis mainly focuses on the fault-tolerant storage of data,aims at failed nodes in distributed storage,and provides theoretical bases for distributed storage codes of big data and mobile data.More-over,this thesis still provides technical supports for reliable and efficient storage of massive data.Firstly,this thesis gives some overviews of the traditional data fault-tolerant technologies,i.e.,multiple copies mechanism and MDS codes.Then the thesis introduces three significant distributed storage codes,i.e.,Regenerating Codes?RGC?,Locally Repairable Codes?LRC?and piggybacking codes.These overviews include fundamentals,current situations and their respective advantages and disadvantages,which can lay the foundation for further research.Secondly,two improvements have been made in a piggybacking framework for repairing parity nodes.Under the premise of keeping the complexities of coding and decoding ba-sically unchanged,the first improvement which aims at repairing parity nodes reduces the repair bandwidth of both systematic nodes and parity nodes to varying degrees,and the de-sign of the improvement becomes more flexible.The second,focuses on the systematic nodes,is valuable to sacrifice the repair bandwidth of a small proportion of parity nodes in exchange for the repair bandwidth of systematic nodes with more nodes.Thirdly,the generalized piggybacking codes are studied in depth and some problems on their the repair of parity nodes are pointed out.Most significantly,this thesis puts forward a multi-node repair strategy for piggybacking codes for the first time based on the generalized piggybacking framework.After analyses,this thesis obtains the influence of factors on the minimum average repair bandwidth ratio of multi-node.Such as the number of repairable n-odes,the proportion of protected instances,the proportion of designed instances,the number of systematic nodes and so on.The method has some referable value in designing piggy-backing codes with excellent multi-node repair performance.Finally,this thesis proposes a novel double-layered piggybacking framework?D-PB-1?to re-pair both systematic nodes and parity nodes effectively at the same time via optimizing the proportion of reserved substripes₍?6?and piggybacked substripes_{?6?,which greatly reduces the repair bandwidth.After analyses,when the number of systematic nodes6}and parity nodestends to infinity,the average repair bandwidth ratio of all nodes approaches to zero.Subsequently,the improvement of the double-layered piggybacking framework?D-PB-2?is put forward,which result in a lower repair bandwidth by changing the construction of the piggyback blocks belong to systematic nodes.Compared with other piggybacking codes,the two classes of double-layered piggybacking codes not only have flexible design require-ments but also obtain the optimal comprehensive repair efficiency,especially for D-PB-2.In addition,this thesis also presents a multi-instances model for the double-layered piggyback-ing framework,which reduces the number of substripes in the framework without sacrificing repair performance.Last but not least,analyzing a construction method for the parity nodes of MDS codes,this thesis associates an extreme case of the improved double-layered pig-gybacking codes and reveals the fact that their repair processes are the same essentially.Moreover,this thesis also derives a hypothesis on the lower bound of node repair for the basic piggybacking framework theoretically.