I took an interesting one-day course from DataStax on Apache Cassandra - a big data solution used by Netflix, Comcast, eBay, and dozens of other companies.
Like what Twitter uses, it's designed to be easily scaled horizontally, and provides great performance, but with data-duplication and at the cost of consistency; you get eventual consistency, which is fine for social media, but not so hot for banks or financial institutions.
DataStax sells a commercial version (support, extra tools), so I was expecting some bias, but what struck me as amusing (and a little over the top) was the presenter calling RDBMSs "legacy" all day long.
Cobol is a legacy language - it's been superseded by languages that can do everything Cobol can do, and more.
Cassandra, and other big databases, can't do everything that SQL Server, Oracle, MySQL, etc, can do. In fact, big-data databases are kind of crappy at a lot of what an RDBMS can do with ease. I am confident Netflix doesn't use Cassandra as their billing database, for example. But maybe they use Cassandra to see if there is a correlation between the movies and tv-shows you watch and your credit card of choice.
Fortunately, both solutions play well with each other, just as an OLTP system integrates well with a data warehouse.