The Research On Optimization Of Topology Sensitive Repair Technology In Distributed Storage System

Node failures often occur in large-scale distributed storage system. With the intro-duction of codes, it allows us to reduce the storage costs while ensure the reliability by data regeneration. When a node fails, it’s necessary to regenerate new redundant data in new node, referred to as newcomer node to ensure the reliability. We mainly focus the transmission costs and regeneration time in data regeneration. In data regeneration, the less traffic in network means less transmission costs. And the duration time directly influences the reliability and availability of overall system. However, these two metric-s are highly dependent on the network topology, which is unfortunately overlooked in related work. A ideal network model assumes that two nodes in storage system are di-rectly connected and the link is symmetrical. And the difference between the ideal and the actual topology model will lead to serious underestimated performance. Although the MDS codes have been acknowledged can guarantee the reliability with less storage overhead, the existing MDS codes rely on the ideal topology model, the newcomer n-ode fetch data directly from the providers and regard the transmission distance and the heterogeneous of link between nodes, so the regeneration time if often limited by the bottleneck link and the transmission costs were covered by the ideal network model.In this paper,1) For optimization of data repair technology in network coding framework, proposed a construct algorithm based on Stenier-Tree and give a theoret-ical analysis between repair bandwidth and repair traffic.2) Design a heuristic tree-structure regeneration protocol and analyzed the performance of regeneration proto-col by experiment. It provides support for regeneration of data in distributed storage system. The result show that the regeneration bandwidth has been greatly improved without incurring excessive traffic and reduce the duration time.