Tableau - Better to use table calculation or add a new dimension to the extract?

Question

I have a Tableau file that refreshes daily, pulling data from my company's data warehouse and then creating an extract of the data and creating a packaged Tableau file for employees to view through Reader. My two concerns are (1) how large the file is getting due to all the data in the extract, and (2) file performance (right now loading takes a bit but isn't terribly long.

I have a number of LOD calculations that I created as new dimensions, thinking that would improve performance. But now I'm wondering if it is actually better to have these as table calculations, that way there are less columns in the data extract, keeping the file smaller.

I tried searching but couldn't find the answer. Is it generally preferable to use a table calculation vs. adding the calculation as a new dimension? Thoughts? Thanks!

Answer 1

Are you able to make an aggregated extract?

If you don’t need people to drill into individual records from the visualization, then you can perform your aggregation as you make the extract instead of when generating the viz. That can lead to a much smaller extract and faster visualizations, with a few costs.

To make an aggregated extract, first hide all the fields you don’t need in your extract and the choose “agregate to visible dimensions” from the extract dialog. If you have date fields, you can choose the level to truncate (roll them up) to further condensing the extract.

The extract will apply the default aggregation function to each measure grouping by the visible dimensions. So change the default aggregation for measures first if needed.

When using this type of extract, be careful how you further aggregate your measures to make sure the result makes sense. Sums of sums is well defined, as in mins of mins. Other combinations may be suspect. You may even want to turn off Aggregation under the Analysis menu when working with a pre-aggregated extract.

This approach can save a lot of time and space, and works well when you know exactly what visualization to make. It doesn’t have the flexibility you get with a full unaggregated extract, but you can get a huge performance boost in return.

One option is to make both kinds of extracts, and build vizzes from the compact one to get a quick responsive dashboard and use a full extract to allow drilling downinto details