That's interesting and very promising. But what I miss in most discussions of relational vs NoSQL is some kind of separation of data model related, performance related and scalability related arguments. These are three rather different issues.<p>There is no way to have a benchmark alone show me what it means to denormalize my data model. Denormalization can help scalability and sometimes performance, but it has huge implications for maintainability, data quality, code complexity and even organisational matters.<p>I do understand the advantages of being able to run any algorithm on data and not be limited by SQL. But abolishing normalized data models altogether and replacing them with application code and custom data structures everywhere makes me very skeptical as someone who had to deal with data quality issues for many years.<p>So that's why I'm glad to see some performance issues of RDBMS improved by new storage architectures and also the innovation going on in terms of recursive queries to deal with hierarchical and graph data structures.