This document describes CFS, a Cassandra-backed storage system for Hadoop. It discusses the motivations for building such a system given Cassandra's strengths in scalability and real-time data but lack of support for ad-hoc queries. The solution presented uses Cassandra to store Hadoop file metadata and block data, allowing tasks to run locally for data locality. It describes how files are written by splitting them into blocks and updating column families for inode metadata and block storage (sblocks). Reads check the inode to determine blocks and request local or remote data as needed.