Archives of the TeradataForum

Message Posted: Mon, 19 May 2014 @ 09:31:56 GMT

	<Prev	Next>		<<First	<Prev	Next>	Last>>

Column partitioning is when we have no PI for a table or join index. I like this quote from the document:

column partitioning is usually not suitable for workloads that often select both a significant number of rows and project many columns from a table. However, column partitioning might be suitable if a request selects a significant number of rows, but projects only a few columns, or conversely, if a request projects many columns, but only selects a small number of rows.

I can see that we can do aggregates more efficiently and faster than row-based because of the storage and also they don't need index. However, with advanced in-memory analytics, these days, I feel columnar is just accessory. Do it on the fly :).

I have not implemented Columnar solution. I am involved in many POCs of Hadoop, NOSQL databases and have to make recommendations. I feel that we implement features only if there is ROI. this is my viewpoint.

	<Prev	Next>		<<First	<Prev	Next>	Last>>