Relational database management systems like Oracle “should be considered legacy technology; [they are] more than a quarter of century in age and ‘long in the tooth.'”
This is not from some random disgruntled crackpot. It’s from Mike Stonebraker, a relational/SQL DBMS pioneer in the 1970s and the co-creator of the influential Ingres and Postgres DBMSs, written in his new blog.
Essentially, Stonebraker is saying that column-oriented databases are much faster than “traditional” RDBMSs, especially in OLAP and data warehousing applications. He claims that the new design (which is not really that new, since Sybase IQ and others already use it), combined with new compression techniques, result in performance that is 50 times better than row-oriented systems like Oracle.
But are column-oriented databases inherently faster, or is bad Oracle database design the root cause of the apparent difference?
Regardless, it turns out that Stonebraker has a new start-up whose product is (surprise!) a column-oriented DBMS. (I do like their clever slogan, however: “The tables have turned.”) Obviously, he has a vested interest in the downfall of Oracle, DB2 and the like. But is his critique just marketing bluster, or does he have a valid point? The blogosphere reaction is mixed: some are certainly crying foul that he is just pushing his product, while others agree that the RDBMS is dead. Still others take the middle road, agreeing that a row-wise design is better for OLTP, while column-wise is better for OLAP.
One wonders what Chris Date thinks about this “progress”?