A comment on the idea that the new “NoSQL” databaes may mean the era of the relational database may be over, at least in high volume OLTP. Despite what RDBMS verndors like Oracle may want you to believe, that era of the relational database never really began. To this day the world’s largest and busiest OLTP databases still run on legacy products like IBM IMS, CA-IDMS, and other CODASYL style databases. Data is hard to find but it is possible these systems run more transactions per day today than all relational products together.
IMS and IDMS first went live in late 1960’s and predate the existence of SQL, they are certainly “NoSQL”. The IBM data access method DL/I (or DL/1) for IMS is a put/get type architecture using keyed data segments, much like the NoSQL key/value pairs with structured values. There are differences, for example DL/I can store child keys in the data segments, which is what makes IMS a “heirarchial” database. I haven’t seen that in the new “NoSQL” databases but it could be done.
In a way it is shame to see all this effort to essentially re-invent something IBM, Cullinane and others built 40 years ago but then not everyone wants to buy an IBM 390 mainframe to run it. So we have NoSQL on more modest hardware and with more modern programming methods. But NoSQL is not so new but more another flavour of the way things have always been done in high volume OLTP.