Dynamic Replication Technique for Micro-Clouds Based Distributed Storage System

In this paper, we investigate a dynamic replica placement strategy to achieve minimum repair time during a complete loss of data center in large-scale distributed object storage systems while keeping the high availability and durability of the system as well as its balanced load and minimum network latency. Specifically, we address whether random placement is good for latency, availability, durability and load balancing, we also find out if there are better alternative placement strategies with better load balancing, moreover, we research which level of working unit - object or bucket - gives the best performance during migration. we show that a dynamic placement strategy reduces latencies, improves load balancing, and achieves faster repair times. In addition we show how the strategy with object placement help to get higher repair bandwidths and better load balancing with different object popularities, and the drawbacks of object migration in getting a higher repair bandwidth.