The ‘traditional’ way of storing a database is on a mainframe or supercomputer, where all the information is stored in tables with the information all uniquely stored, frequently containing id references to other tables. For instance, an ‘orders’ table would have a customer id in it, and the ‘customer’ table would have their name and address. The programming language for databases like that is SQL - PostGres and Oracle are examples. That model gives you a lot of advantages - the data is always consistent, changes are either made completely or not at all - but every query has to go through one machine, so performance can suck, and you waste a lot of time ‘joining’ tables together for certain kinds of query.

If you’re storing eg. a blog with comments on it, that model doesn’t make sense. Each page has a varied selection of comments, comment will have a username and maybe their icon, which will rarely change, but will need to be evaluated by the database every time. It would make more sense to output the pre-rendered page as a JSON blob, and you could have a hundred machines with a few pages each to share the load. Updating people’s icons and adding new comments would need to be done by telling each machine to make a certain update if they’ve a copy of that page; you’d ‘eventually’ be consistent, but if you don’t care about that then you get a very scalable robust solution quite cheaply. Examples of such ‘NoSQL’ databases are MongoDB, Hadoop and DocumentDB.

Linux foundation have looked at DocumentDB’s license and said ‘yes, free enough for us’, so they’ll adopt it.