Information: open this post in Google Chrometo view PDF file attached. ABSTRACTHighly available cloud storage is often implemented with complex, multi-tiered distributed systems built on top of clusters of commodity servers and disk drives. Sophisticated management, load balancing and recovery techniques are needed to achieve high performance and availability amidst an abundance of failure sources that include software, hardware, network connectivity, and power issues. While there is a relative wealth of failure studies of individual components of storage systems, such as disk drives, relatively little has been reported so far on the overall availability behavior of large cloud-based storage services.