Having watched an equally interesting and entertaining panel discussion, “The Aarhus 6”, at GOTO Aarhus, it strikes me – again! – as weird and funny that NoSQL discussions almost always seem to end up discussing performance and availability.
In my experience – and Martin Fowler actually noted this towards the end – most development teams and organizations don’t care about the availability of their database. Not as in “don’t care at all”, but as in “availability demands are so that an ordinary OPS team can make the database sufficiently available”.
In other words, almost no organizations need five nines, four nines, or something like that. Most organizations, if they’re honest and don’t pretend to be more important than they are, probably don’t even need two nines!
Thinking about this, it’s funny that the discussions almost never discuss how easy the databases are to get started with, and how easy it is to store data in them. Matt Dennis even managed to talk almost for his entire 50 minutes in his “Big Data OLTP with Apache Cassandra” without touching on how data is actually stored in Cassandra. It’s only because someone in the audience asked about it, that he said the word “column family”.
Chris Anderson did at some point however, comment that “the winner” (i.e. the NoSQL database that development teams will end up choosing), might be selected “because it’s easy to install”.
This is actually spot on, in my opinion! I think most development teams are better off prioritizing ease of installation and operation, and ease of usage, far far over operational quality attributes like insane scalability and availability.