One reason is that filters can not be properly defined at report stage because the whole set (which may not be essential) is taken out of the request. There are a few causes: 

• Create a request that returns many documents from the table(s), if lower aggregated records are sufficient. This can be verified by selecting the amount of marks at the underside of the Tableau Desktop. If this number is very big, you could draw a great deal of information from the database
• Use indigenous drivers: Tableau will suggest or require you to produce an extract of information so that you can proceed to work with a specific driver. The use of indigenous drivers instead offers better efficiency with ODBC links.
• Testing using a new tool: A useful way to find out WHETHER the slow query causes a slow workbook is to test the same request in a different instrument, for example, Microsoft Access or Microsoft Excel.

See userName\My TableauRepository\Logs for the request to be executed and see the log.txt file. Open this document and SCROLL up from below until you discover the following section: A query passed to the database is the section between the start-and-end query tags. This text can be copied and then used by an instrument such as Access or Excel. If it takes the same time to return as in Tableau, the issue is probably the query, not the instruments.

• Use excerpts: If you have problems with performance, create a table extract. These files can CONTAIN performance-oriented characteristics such as pre-aggregated hierarchical data and pre-calculated fields (reducing the quantity of visualization job). 
  • Tune your indexes: Ensure you have indexes on all columns that are component of the table joins Make sure you have indexes on any column that is used in a filter. Explicitly describe keyboards Foreign main relations For big datasets use table partitioning Use NOT NULL columns where possible. The Query Optimizer selects these statistics to find or UPDATE the effective plan.
  • (OPTIMIZE the mode of information: useful to generate summary tables if most of your requests need only aggregated information-not base level documents.