I was reminded, at IBM Innovate 2013, that IMS, the mainframe hierarchical database on which I first learnt what DBA skills I have, is still very much alive and well. In fact, another day, I'll talk about some real IMS innovations that are on the horizon.
IMS is still a fine database for applications that suit it - high value, high volume, transaction processing, I guess, with a well-defined and stable access path. In fact, it's still an extremely fast option at the highest scale; and competitive with many newer databases. One recent innovation is support for SQL access and IBM claims that it's seeing a definite move back to mainframe solutions for some large customers. IMS can probably be best seen as a transaction manager these days - it is still a strategic IBM product for transaction management (although the paper actually quoted refers to 2004).
IMS has always been a fast DBMS; its issue has been that it doesn't really provide 'data independence', which means that the database needs to be designed for specific applications/data structures by highly skilled DBA staff and coders need to be aware of the database structure in order to write efficient code; whereas with a relational database such as DB2, any SQL query that specifies the right target will work, and the database optimiser takes care of performance. This may be becoming less of an issue (for reasons I don't really have time to go into here, but might write about later on) but overall, including people and training costs, a relational database like DB2 should usually be a cheaper option than IMS for a given level of performance and application sophistication, even if IMS Fast Path, say, has better absolute speed and lower operational overheads when used properly; and sometimes absolute performance is more important than cost of ownership. In other words, although a relational database should probably be the default choice for business processing today, there are still some use cases where less fashionable kinds of database are actually a better choice.
I am reminded that IT is still very much a fashion industry. IMS, being a hierarchical database, is ideally suited to XML processing, and perhaps IBM was a bit slow to exploit that but now it's too late, the moving finger of fashion has passed on from XML and people now want JSON instead (see Philip Howard's comments here). ims isn't seen as a fashionable addition to someone's cv today - even if it is still a good solution, technically, to many large organisations' data processing needs. similar fashion 'victims' include the 4gl (alive and well and very productive in the form of Uniface, which now dare not call itself a 4GL any more) and the mainframe generally, which should probably be the automatic default choice for hosting cloud computing but, I suspect, often isn't.
Perhaps, as IT matures, it should be moving on from fashion-based technology choices towards fact-based choices? Is that a flying pig I see in front of that cloud?