Virtually testing conversion of tables to column-organization

You can virtually convert row-organized tables to column organization to determine whether the conversion would improve the performance of a query workload.

Before you begin

About this task

The conversion takes place only in memory and for the purposes of testing. No physical changes are made to the structure of the tables that you select.

You can do this virtual testing either after the Workload Table Organization Advisor recommends tables to convert to column organization or without running this advisor.

If you want to run a virtual test after the advisor gives recommendations, you can use the recommendations as a baseline and select or deselect other tables that are referenced by the query workload. Then, after you run the virtual test, you can compare the estimated performance improvement with the performance improvement that is estimated from the recommendations. For example, suppose a query workload references tables A, B, C, D, and E. You run the Workload Table Organization Advisor, which recommends converting tables B, C, and E and estimates the performance improvement at 80%. You decide to run a virtual test conversion of all 5 tables and find that the estimated performance improvement is still 80%. Therefore, converting tables A and D provide no benefit.

If you want to try figuring out yourself which table conversions might improve the performance of a query workload, you can choose not to run the advisor and instead try virtually converting different combinations of tables. For example, if a query workload references tables A, B, C, D, and E, you might test virtually the conversions of these combinations of tables:
  • A, B, C, D, E
  • A, B, C, D
  • A, B , C
  • B, C
  • and so on.

Procedure

To test virtually the conversion of row-organized tables to column-organization:


Feedback