Efficient data reliability management of cloud storage systems for big data applications

Cloud service providers are consistently striving to provide efficient and reliable service, to their client's Big Data storage need. Replication is a simple and flexible method to ensure reliability and availability of data. However, it is not an efficient solution for Big Data since it always scales in terabytes and petabytes. Hence erasure coding is gaining traction despite its shortcomings. Deploying erasure coding in cloud storage confronts several challenges like encoding/decoding complexity, load balancing, exponential resource consumption due to data repair and read latency. This thesis has addressed many challenges among them. Even though data durability and availability should not be compromised for any reason, client's requirements on read performance (access latency) may vary with the nature of data and its access pattern behaviour. Access latency is one of the important metrics and latency acceptance range can be recorded in the client's SLA. Several proactive recovery methods, for erasure codes are proposed in this research, to reduce resource consumption due to recovery. Also, a novel cache based solution is proposed to mitigate the access latency issue of erasure coding