- Yes, Tableau only asks data that is required to render dashboard - 400 points (mark on the viz, value on tooltip, filter content - all of them are "points")
- Yes, filters are applied before bringing in the data to Tableau
- Majority of performance issues with large extracts is a file size itself and time associated with loading it for the first time. Most of the time Tableau has no need to load all of it plus in Tableau 10 we've made several improvements in that area.
So yes, it's possible to have large extract and fast dashboards. As usual it comes down to complexity of calculations and types of filters applied.
Thanks for the answer.
Just a couple follow up questions as I'm not entirely clear on your answers.
1. Regarding the filters. Are you saying that tableau does not transmit the underlying aggregate data. Instead it aggregates the filtered query in the server, then the server transmits the result set?
2. I didn't ask this in the original query. But, if I have the option to make the underlying data visible (View Data). Would it still just send the 400 data points mentioned above or does it send all the data?
By any chance do you know of a guide of a web article which discusses designing nice tableau dashes for speed.
- Correct. In most cases that's what's happening. Not sure if I understand what's "underlying aggregate data" though. Basically, if on the viz you'll go to Analysis > View Data menu - Summary tab will show you what was pulled from the database.
- When you move to Underlying Data (or Full Data) tab - it'll issue different query and will bring a different dataset. This particular query is typically limited to top 10000 rows so it won't be "all the data".
- Best Practices for Designing Efficient Tableau Workbooks, the V10.0 edition | The Last Data Bender
May I ask what's the underlying concern? Load generated on the backend? Billing associated with querying large amounts of data?