 |
| How to remove fields, then aggregate remaining data |
I was surprised to find that it does not, in most respects.
Dropping unused fields means removing columns that are not used in any of the worksheets in the Workbook. This is done in Tableau by editing the data source and selecting "Hide All Unused Fields", demonstrated in the gif above. (Note: No other platform automates this as an out-of-the-box feature.)
Getting rid of unused fields is a go-to performance recommendation for in-memory modes in Tableau, Power BI, and Qlik. One might expect that dropping unused fields would reduce the RAM consumption and improve the calculation performance of the workbook. However, when I tested this in Tableau, I was surprised to find that the RAM usage was unchanged when I removed unused fields, saved, closed the solution, cleared the cache, and reopened the smaller extract. Implied is that Tableau is smart about what it brings into memory compared to leaving on disk.
Once I saw the RAM unchanged, I was unsurprised to find the calculation performance stayed the same without the extraneous columns, too.
While interesting, dropping unused fields may still have other performance benefits worth considering.
Two minor reasons to slim down your extracts come to mind: First, fewer columns may speed up the refresh process by transferring less (narrower) data to Tableau from the data source. Second, the saving and opening of the extract is likely to be faster because it will be smaller.
However, a major performance benefit may be unlocked as a result of dropping fields: aggregating the remaining fields. Aggregating reduces the number of rows in the data source, which means less data to load and process for users as they interact with the solution. This is the box that is checked in the gif above. (Note: No other platform automates this as an out-of-the-box feature, either.) This aggregation achieves the original desired benefits of reducing the RAM footprint of the solution and speeding up calculations.
So, dropping unused fields is nice, but if you want users to experience better performance, you must aggregate the extract, as well.