NoSQL: Not Boring

An accessible but technically grounded survey of the NoSQL landscape in 2012, structured as a response to the then-prevalent claim that NoSQL datastores would replace relational databases. The talk develops three arguments. First, the attractions of NoSQL — such as scalability, cost, flexibility, and eventual consistency — are real but apply to a narrower class of problems than commonly assumed. Second, the "boring" requirements that relational databases satisfy beyond transactional semantics, such as security, auditability, and high availability, are not incidental features; they are what valuable data eventually requires. Third, the relational ecosystem — SQL as a lingua franca, decades of tooling, interoperability standards, and the accumulated technical talent that knows how to use it — is itself part of what valuable data eventually requires. Ecosystems of this depth are not rebuilt cheaply. Note that the CAP theorem is commonly misread as a permanent trilemma authorizing weak consistency by default; a better reading is that good systems should maximize both consistency and availability during normal operation, and partition tolerance is the specific case requiring a tradeoff. Case studies include PayPal, Facebook, and Apple iTunes — all nominally associated with web-scale infrastructure, all running MySQL or Oracle for operations involving money or user-critical data. The observation: NoSQL is interesting and valuable work, but the constraints and tradeoffs still have to be mastered.Invited talk at the Intel Software Professionals Conference (SWPC), October 18, 2012. Companion to the author's HPTS 2011 position paper (doi: 10.5281/zenodo.20076475). Together the two documents constitute the primary written record of the author's "boring" thesis: that valuable data inevitably requires industrial-strength infrastructure.