Dynamic Replication Management for Object-Based Storage System

Abstract

Data replication has been widely used as a mean of increasing the data availability of large-scale storage systems where failures are normal. Aiming to provide cost-effective availability, and improve performance and load-balancing of large-scale storage cluster, this paper presents a dynamic replication management scheme referred to as DRM. A model is developed to express availability as function of replica number. Based on this model, minimal replica number to satisfy availability requirement can be determined. DRM further places these replicas among Object-Based Storage Devices (OSD) in a balance way, taking into account different capacity and blocking probability of each OSD in heterogeneous environment. Proposed DRM can dynamically redistribute workloads among OSD cluster by adjusting replica number and location according to workload changing and OSD capacity. Our experiment results conclusively demonstrate that DRM is reliable and can achieve a significant average response time, and load balancing for large-scale OSD cluster.