Multi-node repair based on GA_PSO with Fractional Regenerating Code Combined with Prior Replication

Abstract

Erasure codes can improve the reliability of modern Distributed Storage Systems (DSS) by preventing data loss and nodes failure. Regenerating code is a class of erasure codes that allow for repairing of failed nodes. However, regenerating code increases the amount of the participating nodes and its coding parameters are difficult to determine. In addition, it has huge computational overhead and low repair efficiency that prohibit its applications. Hence, we first propose a fractional regenerating code combined with prior replication with uncoded repair. Simulation results show that it can reduce repair bandwidth and computational complexity by increasing the number of high prior nodes. Second, we formulate the problem of computing multiple failure repairs cost using the proposed code as a redundancy scheme. We model the problem as an Integer Linear Programming problem (ILP) and solve it by Genetic Algorithm_Particle Swarm Optimization (GA_PSO) algorithm. We present results of repairing bandwidth cost for our proposed algorithm in two scenarios to evaluate the effectiveness of the solution approaches. Simulation results demonstrate that GA_PSO can get smaller repair bandwidth cost than GA.