Tableau Desktop allows two different types of connections (live and extract).
In general, it is always better to use an extract. Tableau Data Extract (TDE) are Tableau Software's file type and is created to work optimally with Tableau products. Using an extract will not only saves resources from the database (does not query the data base) but will allow the user to get information offline. That being said, depending on how frequent the data is being updated, some choose to use a live connection. If the data source updates frequently, Tableau Desktop will not show the updated data. Since Tableau Desktop will query the extract, it is expected that the data being seen might be old. If it is important to have the most updated data (live information), using an extract may not be the best. It is important to keep in mind that it is also possible to schedule a refresh for a extract on Tableau Server. If you know the data is being updated ever month at a particular time, it should be possible to schedule a refresh every month getting the most recent date.
I found a link to a whitepaper that addresses the concepts discussed above and should help in working both with the extracts and live connections:
Hope this helps.
Thanks for the response and white paper.
Sounds like I'd typically only use a live connection when I need up to date numbers, primarily when connecting to a database. If connecting to a local static excel doc on my machine would a live connection serve any purpose? Will an extract be faster?
That is correct. An extract should be faster compared to using a live connection. This being said, please keep in mind that all of this is workbook specific. In general, it is better to use an extract over a live connection. It also may be possible that a live connection could be faster than an extract.
You don't want everything to be an extract as that will impact the Tableau Server's performance and that can cause others to get rather grumpy
I encourage my users to use a live connection unless speed is of the essence and/or query/rendering times are taking too long -- a relative term but up to 15 - 30 seconds seems to be the threshold of acceptance and beyond that, "extract time!".
The data source doesn't matter, it's how much is being pulled.
For Excel a live connection is typically best. If you're pulling thousands of rows and 20+ columns from it then realistically the data should be dumped into a real database because <rant> EXCEL IS NOT A DATABASE APPLICATION!</rant>
You can also publish your extracts and re-use them across workbooks which will help reduce server extract clutter.