This is a first article to see how things look.
For a year or more I've been storing temperature readings from 3 probes several times a minute.
The probes themselves are all hooked up to a Nanode that's still in "dev mode" (i.e. a mess on my desk), albeit that one of the probes is actually dangling outside and so providing useful temperature readings (which are input into Cortex - that runs our heating - but more on that another time). The data is sent by the nanode to a server where some php code deciphers it and puts it into a mysql database. More php supports graphing it and a bit more php formats data into xAP packets and sends it to Cortex.
I found that I was changing the database schema when I wanted to add something and because I was using the database as an integration point, with multiple pieces/apps reading from it I would then have to make matching changes to each part at the same time (ironically I realise this is a pain I've been on the receiving end of in my day job for the past couple of years too).
This got me wondering if using a full relational database was really necessary, whether it was use using it as an integration point and maybe some of the "new-fangled nosqls" would be better - given I'm mostly just storing "a few" key/values in a relatively flat structure.