RAFR: A high reliability replication algorithm for cloud storage

Data replication has been widely used as a mean of increasing the data reliability of large-scale cloud storage systems where failures are normal. Aiming to provide cost-effective reliability, and improve performance of cloud storage, this paper presents a A replication algorithm with forward recovery referred to as RAFR. This algorithm considers the inefficiency of hardware and guarantee the system to approximate the break times of the data and take over data transmission, which makes the system more reliably. We implemented RAFR in Hadoop Distributed File System (HDFS) and experiment results conclusively demonstrate that our RAFR is cost effective and outperforms default replication management of HDFS in terms of performance for large-scale cloud storage.