Coding strategies to avoid data loss in cloud storage systems

摘要

Users that store their data in cloud storage systems want their data to be stored reliably and available every time. Even in case of failures the cloud must be able to serve its clients efficiently. For this purpose some mechanism is required that will provide easy and fast recovery of failed nodes. Many coding methods have been used in different versions of Hadoop's HDFS, HDFS-RAID and XORBAS for popular cloud systems. A study of coding schemes like the triple replication, erasure codes, local reconstruction code, regenerating codes, hybrid and double coding is presented in this paper. These are compared on the basis of storage space required, rate of code, number of nodes required for reconstruction and repair process and repair bandwidth in detail in this paper.