Non-relational data stores are growing in popularity, due both to the massive growth in the size of business datasets and to non-traditional access patterns caused by e.g. social graphs. Twitter faces both challenges, and so it is no surprise that we are making increasing use of NoSQL systems such as Hadoop, Cassandra, Redis, and our own open source social graph store, Flock. In this presentation, I will focus on how we use these systems at Twitter, with specific examples of where we ran into problems with a traditional MySQL-based architecture.