Hacker News new | past | comments | ask | show | jobs | submit login

I think one of the reasons that NoSQL databases have been oversold is that a lot of projects don't have people on them who are good at engineering databases. The result is that folks use ORM's badly.

If you are going to use the database just to store data structures form your program, you might as well use NoSQL db's. However, in most cases, you get integration and migration wins by:

1) Placing your engineering effort on the database. Looking at the sort of real world data you are collecting, modelling it well in the database and then presenting an API to the application. The API will either be a relational one (i.e. views) or a procedural one (stored procedures). After a couple of iterations, the schema shouldn't be being fundamentally changed too much though there could be some minor tweaking.

2) Now, with a good API you can build an application on the database using a methodology of your choice. This could be done in an agile way.

Now if integration is not a goal, then sure you can do all the data validation in your application and you can use NoSQL databases. But relational databases are also powerful integration tools in their own right. I can't imagine LedgerSMB, for example, doing well on anything else for this reason alone.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: