I recently read an entry in The Database Column which suggested that Oracle, and indeed every other RDBMS, should now be regarded as a legacy technology, since they are based on the fact that RDBMS technology is now 25 years old.
The article advocates the move to column-oriented databases , which are particularly relevant for data warehouses. The idea (put simply) is that instead of selecting data from rows, it would be quicker for data warehousing technology to select columns from the tables, since only certain columns are typically required for data warehouse requirements.
Suggest that Oracle is a legacy technology, and you might think that the person suggesting it doesn’t know much about the industry. However, the article was written by Michael Stonebreaker – who knows a thing or two about RDBMS technology. Mike was a founder of Ingres, CTO of Informix and is currently professor at MIT . Clearly, with that kind of background, this is someone who knows a lot about the background of RDBMS – certainly more that I do!
However, one thing that does spring to mind is one of Mike’s other ventures – he is also founder of Vertica Systems, Inc. who, would you believe it, just happen to sell column-oriented databases! Now, I’m not arguing that a column-based approach is a method that should be considered for data warehousing solutions, but I would question the ascertation that there is something fundamentally wrong with Oracle that makes it unsuitable when compared with column-oriented databases. One of the claims that Mike makes in his post is that the new design results in a performance improvement of 50 times better than a row-oriented system. However, I would suggest that perhaps the root cause of the apparent slowness of Oracle is down to poor design, rather than a move to column-oriented technology resulting in such massive improvement.
It will be interesting to see whether there is any response that Oracle (or any other vendor) makes to this approach – I’m going to hedge my bets and suggest that a column-oriented database is probably a better thing for OLAP, whereas row-oriented is better for OLTP.
One thing I will give Vertica credit for is their wonderful slogan: "Lose Wait".