Research Of Piggybacking Design For Systematic MDS Code In Distributed Storage Systems

Erasure codes are increasingly used in distributed storage systems of the big data era to ensure data availability and reliability.With the continuous expansion of the scale of distributed storage systems,the phenomenon of storage node failures has become more and more common.However,erasure codes have a problem of large bandwidth cost during repair of failed nodes.?n order to solve this problem,Piggybacking framework has been proposed,and has attracted much attention because of its advantages such as MDS property,high code rate,low number of instances,and low repair bandwidth.Piggybacking design on the existing erasure codes based on the Piggybacking framework can reduce the data download during the repair of the failed node.In this thesis,we combine the systematic MDS code,and studies several mainstream Piggybacking designs based on the Piggybacking framework.By analyzing the coding structure and the repair algorithm of the failed node,it finds the defects in partially Piggybacking designs that repair the failed node.For the repair of systematic nodes and parity nodes,several improved and innovative Piggybacking designs are proposed.This thesis mainly includes the following research contents:(1)An improved Piggybacking design for repairing systematic nodes is proposed.The Piggybacking design RSR-? can effectively reduce the repair bandwidth of systematic nodes,but there are problems of high coding complexity and repair complexity.By optimizing piggyback rules,an improved Piggybacking design(?PBD)is proposed,which effectively avoids the problem of solving the equations over a finite field.A comparative analysis with RSR-? shows that ?PBD not only maintains the same average repair bandwidth rate as RSR-?,but also has lower coding complexity and repair complexity.(2)A nested Piggybacking design for repairing systematic node is proposed.The REPB in the Piggybacking designs has the characteristics of fast and efficient encoding and decoding.By combining the idea of piggybacking across instances and using the REPB coding structure,a nested efficient repair Piggybacking design(NREPB)is proposed.Comparative analysis shows that NREPB can further reduce the repair bandwidth of systematic nodes compared to REPB.(3)A balanced allocation piggyback design for repairing parity nodes is proposed.The original piggyback design OPD can effectively reduce the repair bandwidth of parity nodes in RSR-? and RSR-?.Considering the balance when piggybacking across instances,a balanced allocation piggyback design BAPD is proposed.By analyzing the repair process of systematic nodes in RSR-? and RSR-?,new piggyback rules across instances are given,and applies the piggyback rules of BAPD and OPD to RSR-? and RSR-? respectively can conclude that BAPD can reduce the repair bandwidth of parity nodes in RSR-? and RSR-? more than OPD.