An Optimal Tree-Structured Repair Scheme of Multiple Failure Nodes for Distributed Storage Systems

Abstract

To reduce recovery cost of repairing multiple failed nodes, many repair schemes have been proposed for erasure codes based distributed storage systems. However, most of the existing researches ignore the network topology of storage devices. Motivated by such considerations, we combine delay repair schemes with network topology and propose a tree-structured model based on fountain codes with large value of (n, k, r) to improve the repair efficiency. More precisely, with the consideration of network topology, a new target named data recovery cost is defined to measure the efficiency of coded fragment download and source file reconstruction, and then the optimal recovery threshold is derived to minimize the average data recovery cost of general tree-structured model. Moreover, we analyze and compare the average data recovery cost of general tree-structure with different systematic parameters. To further improve the data transmission efficiency, an optimal tree-structured scheme based on improved tabu search algorithm (ITSAORT) is proposed. Compared with other algorithms, the ITSA-ORT scheme uses Prim algorithm to generate the initial solution and then uses special method to obtain the corresponding neighborhood structure. The experimental results show that the proposed scheme can find a globally optimal solution and obtain lower cost of data recovery. In addition, the ITSA-ORT scheme has lower computational complexity than the optimal tree-structured scheme based on particle swarm optimization algorithm (PSO-ORT) and the optimal tree-structured scheme based on firefly algorithm (FA-ORT).

Keywords

• Centralized communication model
• fountain codes
• delay repair
• network topology
• tree-structured model
• optimization algorithms