Comment by CurtMonash

http://www.dbms2.com/2008/02/16/stonebraker-database-taxonom..., from 2008, may provide some historical context.

Anyhow, I saw this only after hitting my alcohol tolerance for the night, so I haven't made it all the way through on a first read. That said:

1. Mike writes confusingly about MapReduce. In one place he calls it a "data model". That's wildly incorrect. In another he says that Hadoop was introduced as a MapReduce clone. That's a more minor error, mainly in product naming.

2. Mike also oversells the success to date of columnar analytic RDBMS. That said, he's at least directionally correct. But Oracle and Teradata (specifically in its classical row-based mode) aren't dead yet.

3. I think Mike slightly misinterprets what's going on with SparkSQL. It's not directly in the analytic RDBMS category, and those who try to use it as such often give up. Rather, there are data processing pipelines, and SQL is used in certain necessary and high-volume steps.