Hey Aditya -
This is a really big question - and the answer is "it depends".
If the underlying data system can answer the questions fast enough that you aren't waiting too much in Tableau, try connecting live.
If the answer takes too long to come back, consider an extract (Import Data). However, you won't always want to use this rule of thumb as the ERP system could have billions of rows in it....
I agree with you. Yes, its a big question.
The reason I asked is, most of the BI systems till now (Cognos, OBIEE, BO etc) all have a warehouse system(DB) in place and a lot of effort goes in there designing it.
I am curious about the approach for Tableau (taking into account the fast data engine), let us say I have a ERP system where my DB size is around 2TB obviously I do not want to send analytical queries against my OLTP system.
I am thinking
1. Do we use import data?
2. Do we build a warehouse and query against it (Connecting Live)?
If we use option 1, can Tableau DB (Viz/Fast Data Engine) handle such a huge amount of data?
The Ideal approach will be option 2:- Build a warehouse and query against it (Connecting Live).
However, You can always create the Extract(small set of data) on OLTP System
while importing use Filter Conditions to do so(Similar to OBIEE, BO where you create your metadata and define Joins and create Aggregate Measures)
Once you have extract, develop your reports and schedule to refresh the Extract based on your report requirement