18:00 - 19:00
Grenander 1 + 2
We build systems with the assumption of safety: transactions commit atomically, writes are visible now (or at least visible *eventually*) and reads see consistent cuts across our application. However, the databases we rely on often fail to provide these guarantees, especially under distributed systems failure conditions, such as network partitions, partial failure, and clock skew. We will learn how to experimentally verify correctness properties, through case studies of commercial and open-source distributed data stores.