The vague termed NoSQL (not only SQL) database groups object-oriented databases. There is a very wide palette of different architectures (like document stores, column stores, graph databases, and so on). However, they have at least one common point--they store data as structured but not necessarily consistent objects. They are often not ACID-compliant, although they also offer reliability and stability in their own ways. As they store objects (not tables), stored data don't have to be consistent while redundancy is not a great issue.

NoSQL databases are generally used for different purposes than relational databases; therefore, a direct comparison is impractical. In general, object-oriented databases offer better scalability so they are almost exclusively chosen for big data over relational databases. They often use query languages other than SQL but it is also common to use APIs and function calls for querying and processing data. Finally, a lot of NoSQL DBMSs are vertically extendable, which means we can cluster them effortlessly and extend the servers they can use by putting more machines into the rack, giving them more resources.

Based on the previous statements, NoSQL databases are superior to relational databases in almost every aspect. So why should we even consider using relational databases? Well, NoSQL databases have several drawbacks--they are very difficult to set up, configure, and fill with data. Furthermore, it is harder to enforce data integrity, and the lack of transactional support in some NoSQL DBMSs make concurrent editing harder. With smaller datasets (that is, a few GBs), NoSQL databases give only a negligible performance boost. Thus, if relational databases can give a good fit for our data, setting up a NoSQL database is just not worth the time and effort.